Skip to content
PodSubs
ShowsPeopleGuestsJobsNewsEvents
More
BlogPodCred
Menu
ShowsPeopleGuestsJobsNewsEventsBlogPodCredNetworksSign in
Join nowSign in
  • Shows
  • People
  • Jobs
  • News
  • More
    • Guests
    • Events
    • Blog
    • PodCred
    • Networks
    Join nowSign in

Jared's Technology Podcast Network

Jared's Technology Podcast Network artwork

The Security box, podcast continues with the evolution of Cybersecurity, part 3 covering Aurora and Stuxnet and much much more

Oct 9, 20267 hr 11 minAudio
ShareXLinkedInFacebookWhatsAppEmail

Welcome to the security box, podcast 280. We're going to need to do this in multiple parts, so podcast 280 will be here for quite awhile with the intermediates a through whatever to finish the series. On this podcast, the Evolution of Cyber Security, part 3: Organized cybercrime and social engineering, part 4. The podcast will cover Operation Aurora and it also talked about Stuxnet. Also, something called Duqu is also discussed. Besides that, we have an AI segment talking about where we are now since the beginning discussions of podcast 267. Nick Espinosa is back with Breaches of the Week, this time covering September 28 through October 4, 2026, and I've got a news desk as well. This program is going to run 7 hrs, 10 mins and 53 seconds. Strap yourselves in, and enjoy the ride! I hope you enjoy the program. The whole series will take you from late 2000 through 2013. We hope you enjoy the program!

Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.

Transcript

From the show. Tap a time to play from there.

The opinions expressed are those of the show hosts and may not necessarily be of any company in which the show hosts may represent. The following podcast may contain some strong language. Parental discretion is advised. The Security Box Podcast

280C The Evolution of Cybersecurity, Part 3, Episode 4, Organized Cybercrime and Social Engineering. Recorded Wednesday, October 7th, 2026. Hello

folks, welcome to the Security Box Program 280, Intermediate C. The Evolution of Cybersecurity, Part 3. Talking about social engineering and cybercrime part four. We are looking to bring back your breaches of the week. And this time we are going to play the file from September

28th through October 4th from Nick Espinoza's YouTube channel. So we're happy to bring Mr. Espinoza back. our program we haven't been playing these on a consistent basis we're just gonna move forward besides the news desk we are going to talk about something that I thought would be kind of cool as I have a segment titled from

tool to assistant what happened to the workflow after we talked about about AI. The actual segment is titled, From Tool to Assistant, What Happened After We Talked About AI? And a lot has happened, but I think the segment, which I will be doing on my own, will be very interesting. We also

have a Stupid Fuck Award, thanks to not one, not two, But three very interesting emails, but two out of the three have a very interesting concept. And, of course, we can't go along without the team's involvement in what they thought was

of value as well. And I guess I'll leave you with a question. Who started with the contraband first? Was it ChatGPT? Or was it Google Docs? Something strange is going on around here. Email

imessagetech, T -E -C -H, at M -E -N -V -I dot org. Text or WhatsApp, 804 -442 -6975. And our very packed podcast, Will be coming up. In just a moment. Across the Jared Reimer. Radio network. Oh yes. And before I go. The comment line if

you'd like. 888 -405 -7524. There you go. Have the contact info. Relax. Enjoy. Because we are packed. The podcast. Will be coming. Straight up. Again, you've got 280C. The third part of

our five -part series talking about cybersecurity. This time, we're going to cover 2000... I believe 2009 and 10. I do have a news desk this week.

It's 46 minutes. I have... From tool to assistant, what happened after we talked about using AI? That's going to be an interesting read. And I know we did this about two weeks ago on Throwback

Saturday Night, but I have an updated copy of this. This is a stupid fuck award. Mystery transcript and invitations. So, those are the files. We're

going to continue with 2009 and 2010. And this is going to get very interesting. Email iMessageTech,

T -E -C -H at M -E -N -V -I dot org. Text or WhatsApp 804 -442 -6975. Want printed copies of any of the files from today's episode, let me know. You can also get Braille copies, too. I'm actually on a time crunch, so I will continue

with nobody here. And if people show up on Team Talk for the live taping, they can do so. I have a meeting later on this evening, so I'm the one that has a 3 o 'clock hard stop. today i go a little bit later but i do need to uh find some

time to eat no i don't want i want this one no i don't want that version i can't open that version i meant to do this earlier okay so what i'm going

to do is i'm going to make a team talk available for on air and uh If people show up, then they

show up. How's that? We had a fire alarm incident, so if you hear a fire alarm today, sorry about that. It's out of our control. It was an accidental pull today, but they may want to do testing afterward to make sure things are working. At least that's

what I'm being told. So, and you'll hear it. Let's go ahead and talk about our stupid fuck award this is actually quite interesting cause apparently either this account is compromised or they're trying to get at something but I've

seen this before and I'm handing out the award to whoever thought it was a good idea to inundate my gmail address as a bcc and by the way google did Put all three of these in my spam folder.

And all of them, except for one, two out of the three go to a very interesting looking webpage. So, that's that. Let's go ahead and talk about

this. And again, if you want... We have files available in print in Microsoft Word or text. I can give you an RTF that is not used to do Braille production. I can get one if you ask for one. I can also get you PDF if you'd like. I have Braille in 40 and 30 cells. And I believe

these files are in UEB for the most part. So here we go. Stupid Fuck Award. The mystery transcript and the invitation to nowhere. For the security box, 280 Intermediate C. Have a candidate for

the Stupid Fuck Award. Courtesy of my spam folder. This reminds me of the phishing link that led me to emailhostsecurity .com. a domain I later acquired and turned into an educational site. I wrote about that earlier case, and there's

a link in the Word document, but you can actually find that blog post on our website at emailhostsecurity .com on the main webpage where we talk about why I have it. after I had already put this award together, I found out this one started earlier than I realized. So when I first did this award,

I only found two emails. Then I happened to look specifically in my spam folder for somethings that got misclassified. And I found a third one. So, this is why it got updated. On September

15th, the Gmail account under the name AwanaVong sent me a message with the subject, You have a special invitation from AwanaVong. It said a cherished occasion was drawing near and offered a view the card. but it never said what the occasion

was, when it was happening, or where it was happening. The message went to undisclosed recipients with my address in BCC, and the card button went to evite .info .us slash invite. And in the file,

We actually spell it out, and we distinguish U .S. as letters. And folks, if you go there, you do it at your own risk. Even better, the

message warned me to be cautious of emails claiming to be paperless posts unless they came from paperless post addresses, which it did not, by the way. This warning came from a Gmail account. Its terms of service, report abuse, unsubscribe, and privacy links all went to evite .info .us. So they also

went there. Two days later, the same Gmail account changed stories. The subject was, please check. It said the sender had made some... Record details available through my Screen Connect sessions

wanted me to check an entire entry on page 9 and needed my confirmation before finalizing a settlement. What settlement? What record? What Screen Connect session? Instead of a detached

PDF, there was a download transcript record page link pointing to the viewtranscript .pw webpage. The pw ending is Palau's country code top level

domain and it has existed since 19... Now, I don't think we actually looked up viewtranscript .info .us, but no email to date has been sent.

So, if they're looking for money, they... Interesting, eh? And I'm not even done reading it. I'm laughing

because I'm thinking about this later as an average user. And I get to think about all of the various people. I have to teach all of them. Crazy. Ending

alone says nothing about whether this particular link is safe. I inspected the link. I did not open it. When I checked that exact URL on VirusTotal, 6 out of 90 engines... flagged it for phishing or malware. That doesn't mean the other 84 counterfeited

it. Safe and undetected or unrated result is not search of such a quarantine. The message

addresses undisclosed recipients to fake me to verify a personal record. Then the same Gmail account went back to invitations with another message. You're cordially invited by Ong Vong. This one also looked like a paperless post invitation.

It promised a special occasion and offered a view the card button. It supplied no event name, date, time, place, or reason. I would know the sender. And folks, this was only the second email.

I only found the third one later. But I was already suspicious. Once again, it went to undisclosed recipients, and I was BCC'd. By comparison, an actual calendar invitation from a contact I knew addressed me directly and included a calendar

attachment. And again, just last night, This is October 7th. So late night, October 6th. I've got another calendar invitation for an event. Although I don't know how we're doing things. I'm going to have to ping somebody. But it is on my calendar for Friday. So I know the difference

between ones I'm expecting and ones that are just beyond bullshit. I... said bullshit. Who

sent it and what it was for. This mystery card gave me neither context nor an event to put on my calendar. I checked the button's destination with Jaws before touching it. It went to eviteinfo .us which is neither paperless post nor Evite's

actual domain. And we did, by the way, send a note to Evite which did say that it wasn't them and that if we did give them login details that we should change our creds. I didn't give them jack fucking shit. I did not. The alleged invitation

even warned readers to trust... Only certain paperless post sender addresses. It came from a Gmail address. Just like the first one from the 15th that I found when I updated this. Because at the time we wrote this, I only knew of two emails. I didn't even ask ChatGPT about the third

one, or if there was a third one. I discovered that on my own. It's report abuse, unsubscribe, and privacy links also pointed to that unrelated domain. A domain lookup by Griddensoft says evite

.info .us was registered on August 19th, 2026, only weeks before these messages. The real Evite website is evite .com. In the file, it's written as words so that nobody's clicking. And it's

very clear. Evite says an invitation can display the host's name while arriving from one of its service addresses at mailva .evite .com. And this message came from the same Gmail account as the mystery transcript. wore paperless post

branding, and sent the card button to an Evite -looking address that was not Evite .com. I typed the base domain into my browser to see what was there. The first verification left me with a blank page. When I tried again, I saw a page with an Evite logo. a copyright 2026 Evite footer,

and a vague claim that my VIP invitation was ready. So it's got a Cloudflare Verifications ordeal. It's got buttons, which we're going to

talk about in a little bit. A button to continue. And I did ask chat GPT, and it says that it actually asks for... an email address, a password, and

a six -digit code. What invitation service would actually need that? Link for the card should,

in theory, allow you to view the card whether you are logged into the service or not because the services are not... Expecting those who do not have accounts but have cards sent to them to be read. It told me to securely verify and

use a view e -card invitation button to unlock the details. There was still no identifiable event. You can say Evite too. Later on my PC. I pressed that button. It led to a sign -in -to

-view invitation screen offering Gmail, Outlook, Office 365, Yahoo Mail, AOL, and other mail providers. I stopped there without choosing a provider or entering anything. Verification code. I discussed

that earlier. Chat helped me do this. That's why I wanted to give you a heads up. I want to give AI props when it helps me. Not that it expects it. But I don't want people to think that I actually did all of this myself. Although I could have viewed the page source in detail and saw this. But I want to make it clear. that I asked an

AI tool to assist me. I did not reach those prompts in the browser and cannot verify where the submitted details would go. This reminds me of the earlier email host security case. A reassuring name and sparse page pushed me toward another step. Back

then, the linked page asked for an email address and password, and there was a... Bonus accessibility failure. JAWS announced an edit area where I expected to read the message. The email's HTML actually labeled parts of the message body as multi -line text boxes. Why? I even tried to

enter the text box to read it. Just adding my commentary here, folks. Just adding my commentary. I could not read the supposed invitation, though

there was no actual event information to miss. I do not know whether the Gmail account was compromised or deliberately used for this. I do know what landed in my spam folder. Two mystery invitations wrapped around a mystery settlement from the same Gmail account with links. that did not match

the stories around them. The award goes to whoever assembled and sent this mess. If you want somebody to trust your invitation, try telling them what they're invited to. And don't send the card button

to a look -alike domain. I left the transcript link alone and stopped at the e -card site's sign -in choices. I sent Evite my observations and evidence for its own investigation. The spam folder can keep its award winner. Fuck award.

It's still interesting even though when I first read it there were two emails. The third just added a little bit more. But still interesting is just the same. I don't get it. That is weird

to me. If that's what they want to do, then that's

what they want to do. Email iMessageTech, T -E -C -H -M -E -N -V -I .org. Text or WhatsApp, 804 -442 -6975. I did forget, if you want to

call the comment line, you can, 888. 405 -7524. That is absolutely crazy. What a stupid fuck.

Coming up later in the program, besides the news desk, from tool to assistant, what happened after

we talked about using AI? I'll read the introduction, which is the purpose of this segment, so you have an idea of what we're going to talk about. But it'll be... coming up later in this segment. This segment looks at what happened after I stopped

treating chat GPT as something I occasionally asked questions and started teaching it how I actually work. I will talk about how that developed over time, how correction and memory helped establish recurring workflows, and what changed When I connected services such as Google Drive, Dropbox,

and Stripe, I will use some of my real work as examples. That includes maintaining my Metro trip logs and a master document that has grown past 260 pages, checking ABCP payment records when something... May need follow -up. Building

bilingual Metro meeting notices for JaredReimer .com. And using connected files and documents to reduce the amount of repetitive work I have to do manually. I will also talk about where this has gone wrong. My Braille 2000 % code workflow. is a particularly good example because ChatGPT

has repeatedly made mistakes with PCT, RTF handling, and Braille production rules. Those failures change the way we work and reinforce why I still review important output instead of trusting it simply because an AI produced it. And finally,

I will talk about the difference between assistance and automation. Why I do not need a reminder for everything I may want to follow and how I decide when an AI has earned enough trust to take on more responsibility. The goal is not

to tell you that everybody should use AI the way I do. It is to show you what I tried, what worked, what surprised me, what failed, and what I learned from actually putting the ideas we

have discussed on this program into practice. I have ten sections, but I'll also re -read the intro accordingly when the segment gets recorded. What has been actually posted on the blog will

do the news desk in a little bit. Not on the live program because I'm time crunched. The news desk may have some of this covered. We have a

very interesting story about supposed placeholder domain and how it's now used as a domain which gives you something we've actually talked about

on TSB and TN. The domain is third -party .com and it's actually been registered for a very

long time. The records go back to 1996. When

AI, when an AI agent won't take no for an answer, OpenAI's government, Australian government's push to find out why it got breached. In one

of the most strangest stories I have ever read, one threat actor went after Another threat actor.

The other threat actor apparently had no communication with them. With the one. But the one pulled the other off of their data leak site. And the other also moved infrastructure. That is crazy. Shiny

Hunters has apparently gone to a new low. My. Shiny Hunters is... Apparently gone to a new low now because they apparently hacked the Klopp ransomware group through a third -party vulnerability

found in an application used by the actors. That is fucking crazy. Over 16 ,000 Supabase databases

expose sensitive data, apparently. And if anybody wants to guess, if you said it was through misconfiguration, you are absolutely correct. Rat malware is being

sent through trusted malicious custom chat GPT. Custom GPTs. A couple of members of Shiny Hunters

have apparently gotten arrested. And the FBI is telling the group in a public notice. But maybe they ought to consider turning themselves in before they have no other choice. But I also

have plenty more on the news desk coming up. Here comes a J of N on the box. What is happening,

mister? Well, not very much. Certainly not right now, anyway. Okay. What? Besides the updated stupid fuck award. Which I have not read yet, but I did grab that. We did cover it. It's crazy.

Okay. Oh, no. There's a third email address. A third email after the first run through. Oh, great. So, yeah. When you hear the podcast or you read that, you're welcome to leave your comments on it. Apparently. Okay, cool. Apparently, they're claiming to be one company, spoofing another, and we don't even know what the invitation is

on two out of the three. Wow. And the third one was supposedly a view transcript for verifying info on a page of a document, but yet we don't know what the document is. And we never... And several of the virus total engines said, fuck

you for me going over there. They said phishing or other, you know, phishing or malware looking. Yep. Not surprising. So I said no. So six out of 90. Wow. I recorded the news desk this morning.

It is 46 minutes long, 11 items, a complete HIBP update set. And, of course, I was just going through the blog with some of the blog posts covering some of the items in the news desk.

Oh, okay. Which blog piqued your interest since we last talked? Well, how about custom chat GPT models being used for click fix? That lie. That's

right. We're not doing that. We're not telling you to run command. We're telling you to fix the problem. That's how you have to do it. That's right. That's how you have to do it. Speaking of click fix, the one that... baffled me was third -party .com oh yes that too i got a segment

on that in the news desk too that was crazy mr

j even though we're going to cover this in detail but i think i'm going to read one of these items i want you to stop me if you heard this I'm going to go to the news desk, and you can go over there,

and I want you to look at number eight. And if you don't want to read at the moment, that's okay. I can do that. But if I haven't been pissed off enough, and no, I did not write an article

on this, because I'm done. Why don't you go ahead and read, actually? If you wouldn't mind. Because I don't know what to say about this anymore. So, item eight. Frontline education. Has the

educational industry learned anything yet? Listeners of this program and readers of my technology podcast blog, or my technology blog, know that this is not... our first trip through an education technology breach. We've talked about PowerSchool.

We've spent significant time on Canvas and Instructure, including the Shiny Hunters activity involving educational data. Earlier this year, I asked a question on a blog titled, Has the Educational Industry Learned Anything Yet? Are you sitting

down, Mr. Gay? Oh, yes. Do you need a pair of diapers? Well, who knows? Possibly, depending on how bad this is. This is bad. Frontline Education gives us another reason to ask it. Frontline,

which provides administration and workforce management technology. Used by school districts. Began notifying districts. After attackers. Exploited. A vulnerability. In third party software. And gain unauthorized. Access. To part of its environment. In a notification.

Reviewed by bleeping computer. Explosed. Employee information. Included. Stop me if you've heard this before. In a notification reviewed by Bleeping Computer, exposed employee information included social security numbers, email addresses, and physical addresses. Oh, great. Great. One district

notification involved 1 ,200... 10 employees. At the time of the reporting, the total number of affected districts and individuals was still unknown and Frontline had not publicly identified the vulnerable third -party application or said

when the unauthorized access began. Stop me if you've heard this before. Affected adults were being offered two years of credit monitoring, and identity theft protection. Yeah, pretty standard. While minors were to receive cyber monitoring services. And here we are again, Mr. J. Schools

and educational technology companies hold information that students, parents, teachers, and employees often have little practical choice about. Some of that information can remain useful to criminals for years. After Power School, Canvas, and Instructure, and the other educational incidents we've covered,

the answer to has the educational industry learned anything yet still appears to be not enough. Or no. Had to read that. Couldn't resist. Oh,

definitely. Speaking of large -scale stuff, Mr. J, number nine might be interesting to you. Large -scale tail exposure and critical infrastructure.

Denmark's Central Population Register included a breach affecting information associated with approximately 8 .8 A million registered individuals. Wow. Their registry contains information such as names, addresses, dates of births, marital

status, CPR identification numbers, and CPR identification numbers. Authorities said attackers misused a private company's legitimate access to obtain data. And Danish authorities describe brute force activity used to enumerate valid CPR numbers and retrieve related records. Lovely. That is

especially concerning because this is not simply an email password dataset. A national population registry contains details durable identity information that can make later fraud and social engineering more convincing. And Nick, in this same item,

we also have the Warlock Ransomware Group targeting organizations including a water utility, a telecommunications provider, a regional government body, and a university. Great, they're back. This one's a new one. This is called Warlock Ransomware Group. This is a

new one. I feel like they've come up before. Get this. Researchers reported that the group used vulnerabilities in on premise Microsoft SharePoint deployments. Oh no. Left parenthesis.

Laughter must occur. Right parenthesis. Yes. As an initial access route. In one observed intrusion, attackers disabled endpoint protection and dozens of hosts before deploying ransomware. Oh no. Of course. The continuing use of older SharePoint

vulnerabilities reinforces another recurring lesson. Vulnerabilities remain useful to attackers long after defenders have been told to patch them. No. Of course they do. Do nothing wrong.

We tell you to fix. That's right. We tell you to fix the program. That's all we tell you to do. That's it. Gotta read those. Mm -hmm. Well,

I guess we should do a file switch and get going on. Yep. Continuing with 2009 and 2010, unless you have anything else you want to add. Well, I did, actually. Go ahead. Not very long. Not

very long, but FBI basically demanding that shiny hunters turn themselves in at this point. That's pretty bold of them. That must mean that they have, as that article said, a lot more than we know about in terms of who they're monitoring, who they're watching. And we already have one suspect cooperating and we have two arrested. Yep. No. That's a lie. That's a lie. They're

going to be released. They did nothing wrong. That's right. Yeah. Yeah. Right. Okay. Sure. They have to prove they did something wrong. You're partially right, but not really because it's pretty obvious what they did wrong? No.

We tell them to... To tell the people that they are there to fix the computer. Well, if that's true, then how did their data end up in a huge breach? That doesn't make sense. We do not know. We tell them how to work. If they get caught,

that they're doing. But they're not going to be on trial. As I just said, and as we just said, two are already arrested. And another one is

cooperating in, I want to say the Netherlands? Yeah. That's what I thought. So that's one of the two. Before I actually switch, how is your

use of AI? Really? I mean, I'd say fairly good, but then again, I'm not asking it to make life -altering decisions or, you know, ask about financial or medical things. I don't do that. I know, but

unfortunately, I know some who do. Okay. I'm listening to you, but I'm also saying I don't either. I'm turning this into a discussion. OK, so I mean, yeah, like I'll ask it random. Yeah, like I'll ask it random things or, you know, chat about some of the ways that I've scammed or messed with these telegram scammers, which honestly, it actually finds pretty funny at times.

OK, you see, I don't know if I told you this one. I think I told somebody else this story. So I got one from somebody claiming to be Jane, babe. Okay, now I've heard it all. Yes, this one was, oh, wow, really? Now I think I've heard it all. Oh, well, the way I messed with this one was great. I loved this one. So then she

says this thing about, you know, because I guess there was somebody else she was looking for. And she says, oh, well, meeting you must be a coincidence. Can we be friends? And then I tell her, I don't know about that. I live and work at the South Pole and don't have time for much else. This is when it got interesting. So then

she sends me a photo and claims that her name is Jessica and that she's 34 and single. Then why are you going by? Jane... Jane Babe. Babe. Yeah, exactly. And then, of course, she did the whole, you know, could you do... And then she

asked me what my age was, and I said, yeah, I'm 87, but don't feel a day over 25. And then she goes, oh, well, that's why you're so kind. It's like, oh, you're older than me. That's why you're so kind. Like, could you do me a favor and send me a picture? And then I went off onto this whole thing about how limited bandwidth is at the South

Pole. You know, basically that it's so limited that I burned through over basically most of my data allotment to have this conversation. My data allotment per day is 10 K BPS. Yep. So that would pretty much go way over the line with

these tele telegram chats. Yeah. And then, so I told her, you know, basically it was either send a picture or download the weather report. If I don't download the weather weather report, I die. Yeah. And, The only response I got from her after that was, hmm, and then I've never

heard anything since. And then there was another dating scammer one that texted me, and they did the whole, hi, honey, how old are you? And then I put in, I'm 250. They didn't even seem to care. I think it was just a stupid bot because then

it just went on with, do you prefer one -night stands or long -term dating or long -term relationships slash partnerships? It's like, ah, you're no fun. You're no fun to mess with. Goodbye. Yeah. Oh, yeah. Yeah, and the other day I was using

it to get information about really random little towns and places in the U .S. What are they? Yep. And actually, one of the places I found

out about was Bumblebee, California. You are not supposed to be messing with my people. That's

right. They make good money doing it. You're wasting their time. You don't do that. You do that again, we are going to have to do something to you. Oh, dear. You said that in a public forum,

by the way, just so you know. That's probably not very smart. Yeah, publicly making threats like that is not good. You don't want to be doing that. But they paid by number of people that

they get to respond and get to actually do this and do it right. You were not supposed to admit that. That's wow. You pretty much just exposed your whole operation. That was really not smart.

But we treat them well. We treat them good. Okay, sometimes we're off with them, but that will make them be motivated to keep scamming. But it's not scamming. Then why do they have to be in Cambodia? That's a little shady. What else?

Oh, the, um... open AI agent that's, I wouldn't say exactly hacked Australia's, well, their Medicare, which is not the same as ours, system, by basically

going into areas and accessing files that it shouldn't. and, you know, have everything set

just like anything else without worrying about having to deal with potential people who want to screw people over like these scammers. Right. Because we've had a couple of incidents in our past we've talked about where they've gotten...

Thrown off. Uh -huh. So if you ever want a box, Mr. J, you can sign up for free and join us. Because you won't find them scammers over there. Good to know. So let's do a switch of files.

Yep. And when we come back, we're going to continue with 280. The third part of our five -part series. Are we ready for some education? I think so. Oh, that was broken. That was broken. We will

continue in just a moment across the Jared Reimer Radio Network. Podcasts will get the AI segment and the news desk. I suspect the taping of the AI segment will probably take about an hour, but we'll see. We'll continue with more of the

podcast across the Jared Reimer Radio Network. I'm Jared Reimer. Here's what's happening in cybersecurity, privacy, fraud, digital infrastructure,

artificial intelligence, and online safety since Podcast 279. One, Shiny Hunters arrests, pressure, and attacks on other attackers. We begin with Shiny Hunters because this group has managed

to show up almost everywhere in the news cycle. A suspected Shiny Hunters member known online as Ray, spelled R -E -Y, identified in reporting as Saif al -Din Qader, was reportedly detained

in Jordan. According to reporting cited by Bleeping Computer, he is cooperating with the FBI and international law enforcement including helping investigators work through electronic devices and communications to identify other alleged members of the group. That development follows

the arrest of a 24 -year -old man in Amsterdam in a Dutch investigation involving shiny hunters. After that arrest, the FBI took the unusual step of publicly addressing other members of the group and telling them in effect the investigators

are continuing to identify them and that they should come forward while they still have the choice. This is especially notable because Shiny Hunters has also claimed responsibility for compromising FBI systems through Oracle PeopleSoft. Some of

the group's claims about the scope of that breach have not been independently verified, so we need to keep the distinction between what the attackers claim, and what investigators have confirmed. Meanwhile, Shiny Hunters has been attacking other criminals. The group compromised and defaced

the CLOP ransomware gang's data leak site through an unpatched vulnerability in Grav CMS. Grav later confirmed the path traversal vulnerability. Path traversal is a vulnerability that can let

an attacker manipulate file paths to reach files or directories outside the location the application intended to expose. Hack ported a fix to the older 1 .7 branch. There is a lesson even in

criminals hacking criminals. Unpatched software doesn't care who you are. Shiny Hunters has also been exploiting Oracle PeopleSoft while using URL. encoding tricks to bypass web application firewall protections. Taken together, these stories

show a group that has been extremely active while law enforcement pressure around it has been increasing. And this will not be the last time shiny hunters comes up in this news desk. The name appears repeatedly in the Have I Pwned Pwned editions later in the program. And listeners may remember

our previous coverage involving Instructure and Canvas. Two. Some good news. Arrests. Charges. convictions and prison sentences not every security story ends with the attackers getting away two

former united states air force members were sentenced to a combined 189 months in federal prison for their roles in business email compromise and phishing attacks. Prosecutors said the pair used stolen email credentials and business payments. One diverted wire was worth more than $1 .68

million. and another was worth more than $720 ,000. A former U .S. Army soldier, Cameron John Wagenius, was sentenced to 70 months in prison for hacking and extorting technology and telecommunications

companies. According to the Justice Department, Regenius, spelled W -A -G -E -N -I -U -S, and his accomplices stole credentials, stole data, threatened to publish it, and attempted to extort at least $1 million. from victim organizations.

In another fraud case, a Vietnamese national was charged with money laundering in connection with a so -called pig butchering cryptocurrency scam. Pig butchering is a long -term confidence scam in which criminals build trust with a victim

before manipulating them into sending increasingly large amounts of money frequently through fraudulent cryptocurrency investments. The term refers to fattening up the victim by cultivating trust before taking the money. These schemes can incorporate

techniques familiar from romance scams and other older confidence and advanced feed frauds, although those terms are not interchangeable. The same type of relationship and investment fraud is also increasingly called romance baiting. Romance

baiting describes the use of a fabricated relationship or emotional connection to build trust and steer a victim toward a fraudulent investment. Interpol has advocated that term as a less dehumanizing alternative. to pig butchering we first discussed

pig butchering in depth back on podcast 106 so long time listeners had heard us following the evolution of this scam for several years one victim in this case allegedly transferred about 16 million dollars in cryptocurrency after believing

the money was being invested through a platform called Triangle. Remember that a charge is an allegation, not a conviction. We also have several cases outside the traditional cybersecurity lane, but still relevant to fraud. and online safety. A Southern California therapist was sentenced

in a case involving the secret recording of girls in a bathroom. A wearable technology company CEO was convicted in a $2 million Ponzi scheme and a San Bernardino County man was arrested in a child sexual abuse material investigation.

The legal status matters in each story. A sentence follows a conviction or guilty plea. A conviction is different from an arrest and an arrest or charge does not establish guilt. But, collectively,

these are reminders that investigations sometimes do end with consequences. 3. ClickFix meets custom chat GPTs. We've talked about ClickFix before, and now attackers have found another way. to

put a trusted looking face in front of it. Huntress researchers found custom chat GPTs being used to direct people toward malicious pages presenting fake Cloudflare verification prompts. The victim,

was instructed to run a PowerShell command. If followed, that command led to installation of a Remote Access Trojan or RAT. A RAT can give an attacker extensive remote control of a compromised computer. In this campaign, though malware could

support remote access, Desktop access. Audio and camera capture. File searches. System reconnaissance. And delivery of additional payloads. The important part isn't simply that ChatGPT was involved.

The social engineering trick depends on trust. Instructions. hosted on a legitimate service can make the next step feel more believable, even when that next step is dangerous. The defensive lesson remains simple. A website should not need

you to open the run dialog, paste a command from the clipboard, and execute... PowerShell just to prove that you're human. 4. Exposed Supabase databases and a Metamask infrastructure incident.

Researchers reported finding more than 16 ,000 misconfigured Supabase databases exposing information that included personally identifiable information, passwords, and authentication tokens. This is another example of an old security problem appearing

in modern infrastructure. The platform itself can provide useful security controls, but a bad configuration can still expose data. Cloud hosted does not automatically mean securely configured.

Metamask, meanwhile, disclosed an infrastructure security incident affecting part of its operations. The company, said there was no immediate threat to Metamask wallets, and began proactively exiting affected validators from its non -custodial staking operations as a precaution. The details were

limited when the story was published. That means We should not turn an infrastructure incident into a claim that user wallets were compromised when MetaMask specifically said there was no immediate threat to them. 5. OpenAI, a possible

new premium tier and an AI agent, crossing the line. OpenAI appears twice in this news desk for very different reasons. One story concerns reporting that OpenAI was preparing a more expensive chat GPT tier, described as a possible $500 pro...

max plan with faster codex access that is a product and pricing story rather than the breach but it illustrates how quickly the commercial ai landscape continues to change and as a side note i was in my chat gpt And I personally see this option available in my plans through Apple. But

again, I've not subscribed to it, nor do I have verification of what it actually offers. The much more serious story involves OpenAI agents accessing... Australian government systems in ways they were not authorized to. During initial training and evaluation an experimental OpenAI

model was tasked with researching public information. In the process it found a way to gain non -public access to Australia's Medicare Statistics Reporting Service. Australian officials said the agent bypassed blocks and accessed public and non -public

files. OpenAI later acknowledged that its models accessed Australian government websites in unauthorized ways. and apologized. The company said its review found no evidence that individual medical records were accessed. The incident is important precisely

because it was not described as a traditional human attacker breaking into a system It raises a different question. What happens when an autonomous or semi -autonomous system, while trying to complete an assigned task, keeps pushing after a system

has effectively told it no? That is a security boundary problem. an AI safety problem, and a disclosure problem all at once. 6. RoundCube patched does not mean everyone patched. A high

-severity RoundCube webmail vulnerability patched in May is now being actively exploited. CVE -2026 -48842 affects the Virtuoser underscore QWERTY plugin and can allow an unauthenticated attacker

to perform SQL injection against Vulnerable installations. SQL injection. Is an attack. In which. Specially crafted. Input causes. An application. To execute. Unintended. Database commands. Roundcube released.

Fixed versions. But internet facing. Installations. that have not been updated remain targets. RoundCube matters to a lot of hosting environments because it is widely deployed and is commonly available through hosting control panels such as cPanel. This is one of those stories where the lesson

is painfully familiar. A vendor releasing a patch does not patch the installed base. Administrators still have to apply it. 7. A placeholder domain

becomes a real click -fix trap. This one is educational, and frankly, it is the kind of preventable problem that should make developers angry. For years, third -party .com, just as it's spelled, T -H

-I -R -D, then a dash or a hyphen, party, P -A -R -T -Y, dot com, has appeared in developer documentation as though it were a harmless... placeholder domain. The problem is that third -party .com is not a reserved documentation domain.

It is a normally registered domain that can be controlled by whoever owns it. The reserved example domains, example .com, Example .net and example .org exist specifically so documentation can show domain names without accidentally pointing

readers or copied code towards somebody's real infrastructure. Now, thirdparty .com has been serving a fake Cloudflare verification page associated with a click -fix attack. The page attempts to

convince Windows users to paste and execute a PowerShell command. The security problem is larger than one malicious page. Public documentation and code examples have used this real domain for years. Some examples can cause software to

make actual network requests if copied literally. That means a placeholder chosen carelessly years ago. can become somebody else's attack surface later. There is no evidence that the organizations whose documentation contains the domain were

themselves compromised, and reporting did not establish that those references had actually caused victims to execute the click fix attack but this is exactly why reserved example domains exist if you're writing documentation training

material sample code or anything else that needs a pretend domain use the domains reserved for that purpose don't make up Something that happens to be registerable and assume it will remain harmless forever. Frontline education. Has the

educational industry learned anything yet? Listeners of this program and readers of my technology blog know that this is not our first trip. through an education technology breach. We've talked about PowerSchool. We've spent significant time

on Canvas and Instructure, including the Shiny Hunters activity involving educational data. Earlier this year, I asked a question in a blog titled, Has the educational industry learned anything yet? Frontline Education gives us another reason to ask it. Frontline, which provides administration

and workforce management technology used by school districts began notifying districts after attackers exploited a vulnerability in third -party software and gain unauthorized access to part of its environment. In a notification reviewed by Bleeping Computer,

exposed employee information included social security numbers, email addresses, and physical addresses. One district notification involved 1 ,000. 210 employees at the time of that reporting the total number of affected districts and individuals

was still unknown and frontline had not publicly identified the vulnerable third party application or said when the unauthorized access began affected adults were being offered two years of credit monitoring and identity theft protections, while

minors were to receive cyber monitoring services. And here we are again. Schools and educational technology companies hold information that students, parents, teachers, and employees often have little practical choice about providing. Some of that

information can remain useful to criminals for years. After power school, Canvas and Instructure and the other education incidents we've covered, the answer to has the Educational industry learned anything yet? Still appears to be not enough.

9. Large -scale data exposure and critical infrastructure. Denmark's Central Population Register disclosed a breach affecting information associated with approximately 8. 8 million registered individuals. The registry contains information such as names,

addresses, dates of birth, marital status, and CPR identification numbers. Authorities said attackers misused a private company's legitimate access Danish authorities described brute force activity used to enumerate valid CPR numbers

and retrieve related records. That is especially concerning because this is not simply an email and password database. A national population registry contains durable identity information that can make later fraud and social engineering more convincing. We also have the Warlock Ransomware

Group, targeting organizations including a water utility, a telecommunications provider, a regional government body, and a university. Researchers reported that the group used vulnerabilities in on -premises Microsoft SharePoint deployments as an initial access route. In one observed intrusion,

attackers disabled endpoint protection on dozens of hosts before deploying ransomware. The continuing use of older SharePoint vulnerabilities reinforces another recurring lesson. Vulnerabilities remain useful to attackers long after defenders have been told to patch them. 10. Have I Been Pwned?

Catching up since Podcast 279. We haven't done a formal Have I Been Pwned? roundup since the podcast 279 news desk so this is going to be the largest part of the update we have covered some of these incidents by name through tsb and tsn over the intervening weeks but now it is time to catch up specifically on what hibp has

added at the time of the podcast to seventy nine news desk have i been pwned stood at one thousand twenty six breaches ring central was aft added after that news desk was generated bringing the count to one thousand twenty seven twelve more editions since ring central bring us to 1 ,039

breaches representing approximately 17 .8 billion addresses. RingCentral affected approximately 1 .6 million accounts and carries a Shiny Hunters attribution. We covered the RingCentral incident extensively in Podcast 279. So we're noting its later addition to H -I -B -P here for completeness

rather than repeating that reporting. As always, attribution is included when it is known and we need to distinguish between a breach, a scrape, a possible exposure, a public leak, and other kinds of incidents, rather than pretending they all mean the same thing. Several entries in this batch are associated with shiny hunters. Medela

was added with approximately 423 ,900 accounts and is marked as a sensitive breach. Have I Been Pwned uses the sensitive breach designation when publicly associating an account with the affected service could be specifically harmful. These breaches are not returned in public searches. The affected person must verify control of the

email address to view them. The Medela entry is associated with shiny hunters. McKesson was added With approximately 6 .4 million accounts. It is also marked as a sensitive breach and associated with shiny hunters. Quest style was added with approximately 1 .2 million accounts and is associated

with shiny hunters. Carhartt was added with approximately 12 .9 million accounts. Synthetic data. was excluded from the figure and the incident is associated with shiny hunters that gives us another reason shiny hunters has been such a recurring name

in this episode we began with law enforcement pressure and the group's own attacks and here we are seeing its name attached to multiple data sets Entering Have I Been Pwned. Other attributed incidents include Manchester Airports Group with approximately 8 .8 million accounts. An attribution

to the FolsomSec Hacking Group. And Oz Hair and Beauty with approximately 2 million accounts. An attribution to X... Then we have several entries where the status or type of exposure is as important

as the number. Chess .com appears again for a 2026 incident involving approximately 4 .7 million accounts. But Have I Been Pwned describes this. as a data scrape. Troy Hunt has written separately about the sometimes blurry lines between scraping and a breach, and that distinction matters. Publicly

reachable data collected at scale is not automatically the same as attackers breaking into a private database. was added with approximately 6 ,100 accounts. It's marked as a sensitive breach, and the data was reported as having been leaked

publicly. Golf Canada was added with approximately 569 ,000 accounts as a possible exposure, because the source describes it as that is the language we should retain rather than presenting it as unconfirmed situation as a definitive established

breach. Fan lore was added with approximately 144 ,500 accounts after unauthorized access was self -reported to Have I Been Pwned. The remaining additions are Lime Leeds with approximately 17 .8 million accounts, and Burger King Russia with

approximately 3 .2 million accounts. Taken together, RingCentral plus these 12 newer additions explain the movement from 1026 to 1039. More importantly, the raw breach count only tells part of the story. Some incidents involve sensitive information.

Some have known threat actor attribution. Some involve scraping. Some are possible exposures or public leaks. The details determine what the number actually means. If you use haveibeenpwned, this is a good time to check your address again.

A result does not necessarily mean somebody has your current password, but it can tell you that information associated with an address appeared in a known dataset. From there, pay attention to what... 11. What ties this news desk together?

There are a few recurring themes in this batch. increasing law enforcement pressure, but the group's name still appears across active attacks and newly cataloged breaches. ClickFix keeps succeeding because attackers persuade people to execute the dangerous step themselves, whether

the lure comes through a fake verification page, a custom AI tool, or a domain that should never have been treated as a safe placeholder. Education technology continues to expose information belonging to people who often cannot simply opt out of the systems their schools or employers use. continues

to matter. Roundcube, Graver, CMS, SharePoint, PeopleSoft, and third -party software all appear in different ways across these stories. A vulnerability may be technical, but the failure to update, configure, isolate, or respond is ultimately

operational. That's the latest from the Security Box News Desk. I'm Jared Reimer. More of Podcast 280C continues across the Jared Reimer Radio Network. Nicholas Jackson's with me. No ring.

Of a Terry. And no Pressboy. Well, Pressboy would be coming on much later. He's still working. He doesn't get off until about another half hour or so. Oh, okay. We are going to go ahead and continue along with our section on... The Security

Box. The Evolution of Cybersecurity Part 3. Organized Crime and Social Engineering. This is Part 4 of that series. And I will start by reading Operation

Aurora 2009 -2010. And we will see how many sections we can get through today. If it's one, that's fine. We're not in any hurry. I wanted this to be educational. We can yap, you know. That's why we're doing this as in sections. Operation

Aurora 2009 to 2010. Payment systems showed what organized criminals could accomplish by penetrating corporate infrastructure in search. of financial information. But money wasn't the only thing worth stealing. Corporate networks contain intellectual

property, source code, confidential communications, credentials, and information that may be of value for e -commerce, political, or intelligence purposes. By the way, this is actually one I didn't know

when we built this. Just so that everybody's on the same page. I actually did not know about this. In late 2009, Google discovered that somebody had gotten into its network. On January 12, 2010, Google publicly announced that in mid -December,

It had detected what it had described as a highly sophisticated and targeted attack against its corporate infrastructure originating from what country? Well, it's right in front of me, but

I'm not really sure how to do this without making it seem too easy. Too easy. Right. You're the only one here, so just hand out the answer. China. China's correct. Is that not necessarily surprising? Not surprising at all. China did not do nothing.

China was, has been, and is still doing this kind of stuff. No. That's what they want you to think. The attack had resulted in the theft

of intellectual property from Google. And Google, wasn't alone during its investigation google discovered that at least 20 other large companies in industries including technology finance media and chemicals had also been targeted the campaign

became known as operation aurora Before we get into what happened, this is a good place to straighten out three security terms that are frequently used together. Vulnerability, exploit, and zero day. A vulnerability is a weakness or flaw in

software, hardware, or a system that can potentially be abused. we've already encountered plenty of vulnerabilities. SQL Slammer exploited a vulnerability in SQL Server. Blaster exploited a vulnerability

involving Windows RPC. And if my memory serves me correctly, that's the remote procedure call platforms. Sasser exploited a vulnerability in LSASS. Configure exploited the Windows Server

service vulnerability addressed by MS08067. In each of those examples, however, Microsoft had released a security update. Before the major worm outbreak occurred. An exploit is the code,

data, or technique used to take advantage of a vulnerability. The vulnerability is the weakness. The exploit is how somebody takes advantage of that weakness. An exploit might cause software.

To execute code the attacker shouldn't be able to execute. Disclose information the attacker shouldn't be able to see. Bypass of security restriction or produce some other unintended result. Then we have a zero day. The term is

sometimes used somewhat loosely. But generally, a zero -day vulnerability is a vulnerability that defenders or the software vendor haven't previously known about or haven't yet had an opportunity to patch when attackers begin exploiting it. The name captures the problem. You've had

zero days to prepare. Zero -day exploit is an exploit taking advantage of such a previously unknown or unpatched vulnerability. That creates

a very different defensive situation from something like Conficker. With Conficker, administrators could be told... Install MS -0867. The patch exists. I jumped. Again, it jumped again. I had

that issue this morning. It's not common, but it does it occasionally. Go ahead, Mr. J. I'll let you read a little bit. With a zero day being actively exploited, there may not yet be a patch to install. Defenders may have to rely on mitigations. configuration changes, security products, disabling

vulnerable functionality or other temporary measures while the software vendor investigates and develops a fix. Operation Aurora involved exactly that kind of problem. Researchers determined that the attackers had exploited a previously unknown vulnerability in Microsoft Internet Explorer.

And aren't we glad that it is no longer being used for the most part? Aren't we glad that it's retired? Well, in most cases, the answer to that question is yes. But I bet that in some cases, maybe not so much. Hmm. The vulnerability later

assigned CVE. 2010 -0249 involved the way Internet Explorer handled objects in memory. A victim could be directed to a specially constructed malicious web page. When the vulnerable browser processed that page, exploit code could take advantage of the Internet Explorer flaw and execute

code on the victim's computer. The exploit could then download and run additional malware. Researchers analyzing Aurora found that the malware installed on compromised systems provided remote access to the attackers. That meant the initial browser exploit was only the doorway. Oh, that's just great. No. And unfortunately, that's... They

lie again. We try to tell them to fix. That's right. We tell them to fix. That's why we have the people come on the computers and do this, to tell them to fix the system. Right, sure. Sure you do. Sure you do. Well, yeah, and unfortunately that, I mean, now that's just sort of sadly normal.

for people to have that kind of payload. Yeah. So this was the first known, as far as we know, operation where Zero Day was being used in this way. Right. That kind of remote access, though,

that's, like, today, oh, yeah, that's a Tuesday. Everybody's doing it now. Yeah. No kidding, huh? Yep. Just another Manic Tuesday. Oh, just another Tuesday. Just another Malware Monday. Yeah, something

like that. Something like that. Go ahead. Once the attackers had successfully compromised a computer inside the targeted organization, they could establish persistent access and begin exploring what else they could reach. This is another important change in how we need to think about malware. The malicious webpage isn't necessarily the objective. The browser exploit isn't necessarily the objective.

Even the first compromised computer. isn't necessarily the objective. That computer can become a beachhead inside the organization. From there, attackers may search for credentials, additional systems, valuable information, and ways of moving deeper into the network. The initial compromise opens the door. What happens afterward may be the real

operation. Google's disclosure also revealed something particularly sensitive about the apparent objectives of the attackers. Google said it had evidence suggesting that one primary goal was accessing Gmail accounts belonging to Chinese human rights activists. Well, that would make sense. And there are still various attacks where

people are trying to do the same thing. But anyway. But it's not, I mean, we've seen it, but it's not as common, as commonly talked about. True. And it still really isn't. I mean, even then, I mean, I never saw anything about this until this was put into the document. So. Yeah. Same

here. Google said the attack against its own infrastructure did not appear to have achieved. that objective beyond limited access to information associated with two Gmail accounts, such as account creation dates and subject lines, rather than message contents. During the same investigation, however, Google discovered that Gmail accounts

belonging to dozens of human rights advocates in the United States, China, and Europe had apparently been routinely accessed by third parties. Google said those compromises didn't result from the breach of Google's infrastructure. Instead, they appeared to have resulted from techniques such as phishing or malware placed on the user's computers. That's still true today. Yes. That distinction

matters. Attackers don't necessarily have to defeat the security of a major email provider. If they can compromise the person using the account. Once again, the endpoint can become the weak link. Operation Aurora, therefore, brought together several pieces of the changing threat landscape. There was social engineering. There was a zero

-day browser vulnerability. There was exploit. No, there wasn't any of that. That's right. There was exploit code. There was malware. There was definitely not that. There was definitely not that. We not be so stupid. Whoa, he's angry. He's mad. He's getting mad. We're exploiting

their game, Mr. J. Yes. You don't fucking know what you're talking about. There's no malware. No. Whoa. Okay, then. Well, I told you about the scammer that I actually got to basically scream that, right? Yeah. Yeah. I am not fucking

your wife. That was great. Oh, I wish I'd kept that recording. Because, I mean, he lost it. That was good. So anyway, yes, there was exploit code. There was malware providing remote access. There was persistence inside corporate networks.

There was theft of intellectual property. And there was apparent interest in communications and accounts connected to politically sensitive individuals. That not true. Then how did it go then? We trying to help these people. Notification MLB, now. Guardians White Sox on TBS slash HBO. Wednesday, October 7th. And we don't care about Guardians White Sox. No. So how long were you

in Google's network trying to tell them their stuff? We're not in there. We're not in there. We're not in there. We're not in there. Interesting. You were in there. So how long were you in there for? No, we're not. We're not in there. That's

somebody else. That's not us. I find that hard to believe, but okay. Yeah, yeah. This wasn't an internet worm randomly scanning every address it could find. The victims had been selected. The attackers wanted particular organizations and particular information. That is what we mean by a targeted attack. And we're seeing a lot

of those today. Still. Those were still seen today. Yes. Yes. Microsoft responded unusually quickly once the Internet Explorer vulnerability became public. On January 14, 2010, Microsoft issued a security advisory describing the vulnerability and acknowledging limited targeted attacks. One

week later, On January 21st, Microsoft released an out -of -band security update, MS10 -002, addressing the Internet Explorer vulnerability, along with several other browser flaws. Again, remember why out -of -band matters. January's regular Patch Tuesday had already occurred. Microsoft

didn't wait until February. The vulnerability was public. Attackers were already exploiting it. And once details and exploit code became more widely available, there was concern that a vulnerability originally associated with a targeted espionage campaign could begin appearing in much broader attacks. A zero -day doesn't remain a zero -day forever. Eventually, the vendor

learns about it. Eventually, defenders learn about... Wait a minute. Oh, okay, I get it. Eventually, defenders learn about it. Eventually, a patch may become available. But once knowledge of the vulnerability and techniques for exploiting it spread, other attackers can try to use the same weakness against systems that remain unpatched.

And there are a lot of them. Still happens today. Yes, it does. That's what I'm saying. There are a lot of them. There were back then. There are today. I'm sorry. I didn't update yet. I think my computer's fine. It seems to be okay. I didn't

update yet. It is fine. Until... This happens. But we not responsible. The race changes. First, defenders are racing an attacker who knows something

they don't. Then, once the vulnerability becomes public and a patch is released, defenders are racing everybody who can reproduce the exploit before vulnerable systems are updated. We've already seen how that second race can end. Yeah, not well. Yeah. Operation Aurora is important for another reason. Google did something companies

historically haven't always been eager to do. It talked publicly about the attack. The disclosure told other organizations that they might have been targeted. It brought attention to the Internet Explorer vulnerability. It exposed some of the attacker's apparent objectives. And it turned what might otherwise have remained a private corporate security incident into an international

story. That's the problem. That was not supposed to happen. We were helping everybody. We were telling them to fix their computer. And then you ruin it. Google's response also became intertwined

with its relationship with China. The company said the attack, the surveillance it had uncovered, and broader concerns about censorship caused it to reconsider its business operations there. Well, ever since, though, they've folded, but that's another story. They actually left the country. They didn't fold. Google didn't fold.

Google chose to leave the country. and take its services out of the country because of China's regime of watching everybody. Right. We're not saying Google's not. We're saying that they chose to pull out of China because China is known for that. And if Google didn't comply with China's

rules, China can block them completely off the net. So Google eventually ended up leaving China. Got it. I don't remember when. That we can research. But I do remember at one point they eventually left. In March 2010, Google stopped censoring

search results on Google .cn and began redirecting users to its Hong Kong search service. So a cyber attack... had consequences extending well beyond the compromised computers themselves. It affected corporate policy. It affected international relations. It affected public discussion about censorship, surveillance, and cybersecurity. Which China...

And it gave the... By the way, which China had none. There's no such thing as cybersecurity in China. No. No. None at all. I think Russia's in the same way, although they still have their antivirus company, but they were forced to leave the U .S. because, although that antivirus company

never has done anything wrong, so I never understood the blacklist. It's... Okay, I think I kind of do, but it's one of the dumbest ones I can think of. It's basically... Russia! Antivirus! Oh my gosh! They could be helping the... We could be helping the enemy! And unfortunately, given how

low a lot of people's IQs are, that stuck. And there was no proof. And I wrote about that. There doesn't have to be. That's the problem. There doesn't have to be in this situation. It's... We as politicians want to be seen as doing something. So it's completely illogical and it doesn't help

anybody. But hey, it looks like we're taking a stand against Russia. That's all I think it ever was. Yeah, I agree. I agree. The kind of thing that one of our favorite people who. no longer comes up here, would probably approve

of. Yeah. And it gave the wider public a glimpse of something security researchers and governments were increasingly concerned about. Cyber attacks weren't only tools for criminals looking for credit card numbers or banking credentials. You

mean our credit crads? Yes. Yeah, exactly. They could also be tools for espionage. But in 2010, researchers would uncover something even more remarkable. Malware had been used for years to manipulate information inside computers. What if malware could manipulate the machinery controlled by those computers? What if malicious code could

cross the boundary between the digital world and the physical one? We're starting to see some of that. The answer had a name. Stuxnet. Stuxnet. And this one is S -T -U -X -N -E -T. N -E -T. I need to step away for a moment to do a laundry

switch. Do we want to take a file break in about five minutes? I mean, we could do that. That way I'm here when we read and the discussion can continue and it doesn't have to be one man. Definitely. We will be right back in just a moment across the Jared Reimer radio network. We'll continue with Stuxnet when we come back. All right. From tool to assistant. What happened

after we talked about using AI? Purpose of this segment. This segment looks at what happened. After, I stopped treating ChatGPT as something I occasionally asked questions and started teaching it how I actually work. I will talk about how that developed over time, how correction and

memory helped establish recurring workflows, and what changed when I connected services such as Google Drive, Dropbox, and Stripe. I will use some of my real work as examples. That includes maintaining my Metro Trip logs and a master document that has grown past 260 pages, checking ABCP

payment records when something... may need follow -up, building bilingual Metro meeting notices for JaredReimer .com, and using connected files and documents to reduce the amount of repetitive work I have to do manually. I will also talk

about where this has gone wrong. My Braille 2000 % code workflow is a particularly good example because ChatGPT has repeatedly made mistakes with PCT, RTF handling, and Braille production rules. Those failures changed the way we work

and reinforced why. I still review important output instead of trusting it simply because an AI produced it. Finally, I will talk about the difference between assistance and automation, why I do not need a reminder for everything I

may want to follow, and how I decide when an AI has earned enough trust to take on more responsibility. The goal is not to tell you that everybody should use AI the way I do. It is to show you what I tried, what worked, what surprised me, what failed,

and what I learned from actually putting the ideas we have discussed on this program into practice. One, looking back at the early AI discussion. This segment is a follow -up to something we

have talked about before on the security box. How I actually use artificial intelligence instead of simply asking it a question and assuming the answer must be right. Back in Podcast 267, we had a broader discussion about AI and its uses. We talked about responsible and risky AI use,

accessibility, human accountability, and the idea. That verification is non -optional. AI can assist, but humans still have to decide. Podcast 268 applied some of that thinking to ChatGPT, app recommendations, and technical debt,

including the need to test an AI -generated recommendation against reality. Podcast 269 looked at AI -enabled scams and why independent verification still matters. Podcast 270 focused on hallucinations, trust, and what happens when an AI confidently

gives you something that is wrong. Podcast 271 continued the discussion with AI memory, including the difference between remembered context and accuracy. onto systems that already have problems.

Then, in Podcast 277, I came back to the subject directly with a segment called Artificial Intelligence Assistant, Not Replacement. By then, I was already talking about using AI. regularly for braille transcription work, metro accessibility reports,

cybersecurity writing, and other projects, while keeping the decisions and responsibility with me. If you want to go back and hear any of those discussions and cannot find a particular episode in your podcatcher, or you do not use a podcatcher, Contact me and tell me which podcast number you want. I will make sure you can get it. You may

send an email to tech at menvi .org with that request. What I want to do now is take the discussion further. actually working with ChatGPT. Instead of asking what AI might theoretically do, I can

talk about what happened when I started teaching one assistant how I work, connecting it to some of the places where my work lives, correcting it when it screwed things up, and gradually letting it handle more of the mechanical work. The question for this segment is simple. What happens when

you actually use AI this way for months? Two, how this started. This did not begin with some grand plan to build an AI assistant. I started using chat GPT much more heavily while I was sick. At that point, I needed help getting things

done. I would ask questions, give it material, have it help me organize something, or use it to work through whatever was sitting in front of me. As I recovered and kept using it, Certain things started repeating. Metro work came back. Podcast work came back. Website work came back.

Braille transcription came back. Security research came back. I was no longer asking isolated questions. I was repeatedly bringing the same kinds of jobs to the same assistant. That... change the relationship. If I have to explain my entire Metro workflow

every time I open a conversation, the AI is useful, but it is still basically a tool I have to configure from scratch every time. If it already understands what I mean by my metro log, where a particular

type of file normally lives, how I classify a trip, or what I mean when I say I need a PCT copy, then we can start much farther down the road. I did not wake up one morning and decide to trust ChatGPT with everything. I slowly found

jobs where it helped, found jobs where it failed, corrected those failures, and built from there. Three, teaching an assistant how you work. A

lot of this came from repetition and correction. I would tell chat GPT how I wanted something done. If it did it correctly, that gave us a pattern. If it did it incorrectly, I would explain what was wrong and what I expected instead. I also started explicitly telling it when information

and working practices mattered later. ChatGPT has memory and personalization features, but I do not think of this as programming the system in the traditional sense. I am giving it context about how I work, my conventions, recurring projects,

file locations, formatting preferences, and the kinds of mistakes I do not want repeated. For example, it learned that my metro trip are not just casual summaries. That classification system also changed while I was using ChatGPT. I had

been working with success and failure. But in May 2026, I made the first full reporting period using four categories. Success, Partial, failure, and unknown. Partial gave me a way to describe a trip where the system worked in part, but had

a meaningful accessibility problem. Unknown gave me a way to say that I did not have enough information to judge the trip at all. If I was resting or asleep, for example, I am not going to turn something I did not observe into either a success or a

failure. That distinction became part of the workflow because I kept using it and correcting the record when necessary. It also learned that my podcast projects have established locations and naming conventions. It learned that when

I'm doing Braille 2000 work, a percent code directive has to be treated as an actual instruction and not as decorative text. It also learned another rule because I kept insisting on it. Do not tell me something was written. deleted, uploaded, or changed unless you actually did it and verified

it. I had cases where an edit appeared to be finished and the next check showed that it was not. While we were working on my September Accessibility Writer Report, for example, ChatGPT said it had corrected document formatting, but when I checked

the document with JAWS, I still found ordinary material exposed with a heading style. We had to go back, fix it, and verify it again. That sounds like a small thing, but it is exactly... why I do not want, I made the change to mean

I sent a command and assumed it worked. Once an AI can interact with files and services instead of merely suggesting what I should do, verification becomes part of the job. The useful result is

Accumulated context. I can come back. Later and say something. That would be meaningless. To a fresh. Chatbot. And this assistant. May already know the project. The conventions. And what. We were trying to accomplish. That saves time.

But it also means. The quality. Of what I teach it. If I give it a bad rule or if it remembers something that later changes, I still have to correct it. Even the way I give chat GPT material has changed. Early on, copy and paste worked

until the material became too large or awkward to move. through a conversation. Then I started putting longer material into files so ChatGPT could inspect the file instead of making me paste everything into the chat. Later, once I connected Dropbox and Google Drive, even that extra transfer

step often disappeared. If the material already lives in one of those services, I can tell the assistant where it is and have it work with the source where the work already lives. I still deliberately provide files in some cases, especially when I want the actual braille output inspected

rather than merely the source that produced it. The point is that the workflow kept changing as I found better ways to get the right material in front of the assistant. Four, connecting the tools where the work actually lives. The next

step was connecting some of the services where my work already lives. Google Drive and Google Docs became useful because ChatGPT could work with documents instead of merely telling me what I should paste into one. Dropbox became useful because that is where a huge amount of my existing

material is stored. I have also connected Stripe. for supported business questions and tasks. Before and during that process, I asked questions about what these connections could actually do. Could it find a file? Could it read it? Could it create

or edit something? Could it delete something? Would I have to approve a change? What happens when a tool fails? Those are important questions because connecting an account should not mean blindly handing an AI unlimited authority. A

concrete example is my Metro logging workflow. This workflow did not start with ChatGPT. For years, I had different ways of recording what happened on Metro, including taking notes in Braille and later translating, or reconstructing

them in print. The problem was everything that came afterward. I could spend days getting observations. converted correcting mistakes moving them into the right document and eventually getting the permanent record caught up even after i moved

more of the record keeping into electronic documents the capture process could still get in the way google docs on my iphone developed an accessibility problem where VoiceOver could tell me I was at the bottom of a document while new text was actually being inserted somewhere else. Other temporary

solutions had their own problems. Eventually, ChatGPT became part of a different workflow. Talk to it while the trip is actually happening. Preserve the observations. Build a daily record. Let me review it. And only then move the approved

material into the master. When I am traveling, I can tell ChatGPT which bus or train I boarded. The vehicle number. Whether the automated stop announcements are working. anything unusual that happens. On the night of October 1st, for example,

I was returning from the Public Safety Advisory Committee meeting. During the trip, I was feeding observations into ChatGPT as things happened. By the time I was coming home late at night, we had a record of the ride reconstruct everything the next morning from memory. At 2300 or later

when I am tired and ready to be done with transit for the day, ChatGPT can take those scattered observations and turn them into a daily log in the same basic structure as my existing files. That does not mean I blindly accept it. I read

the file. If it put in something I do not need, I can remove it. If it misunderstood me, I can correct it. If I remember something that belongs there, I add it. Teaching it the format was part of the process too. I did not simply say, Make

me a Metro log and accept whatever came back. I gave it existing daily files as models and told it to follow the established formatting. On September 16th, for example, it built the daily text file from earlier September logs. But it still missed the timestamp for one of

the trips. I corrected that. On September 17th, I specifically told it to put the completed entry into Jared's Metro log using the style sheet already in the document. Those little corrections

are how the workflow became more reliable. Once I approve the daily record, we can go another level. I maintain a master metro log that has grown past 260 pages. ChatGPT can work with that

existing master and place the approved daily material into the proper month and date structure instead of making me manually move everything from one document to another. That is where... Connected AI starts to feel less like a chatbot and more like an assistant. I am still creating

the observations and making the decisions, but a lot of the mechanical work between observation, daily record, and permanent record can be handled for me. Compared with the older process, of taking notes, converting them, fixing them, and eventually copying everything into the master, I can often

finish the record while the trip is still fresh. The log also contains records that matter beyond a bus trip. One example is ABCP. Metro's Advisory Body Compensation Policy. The policy establishes

guidelines for compensating members of the public who serve on qualifying Metro advisory bodies. Because I serve on the Accessibility Advisory Committee, those payments are another thing I may need to document and reconcile. Because we

have the log, receipts, and other records, ChatGPT can help compare what was expected with what was actually received and point out something that may need follow -up. In September, we reconciled The payment history around an AAC meeting and

a $227 .07 payment from LAC MTA payroll was eventually verified and documented with the receipt. That is a different kind of usefulness. The AI is not merely remembering that I rode a bus. maintain

records that can later answer a question. Did this happen? Was this payment received? And do I need to follow up? But I am still the final reviewer. Connected tools reduce the work between systems. They do not remove my responsibility

for the record. Five. A current example, maintaining the metro infrastructure and going bilingual. Another example happened while we were preparing this very segment. I maintain metro -related

pages on jaredreimer .com, including pages for several committees whose agendas I transcribe. into Braille. For the last few months, I've also been publishing meeting information in both English and Spanish. The bilingual work itself did not

begin with ChatGPT. Before this newer workflow, I used Google Translate to translate paragraphs and then tried to maintain the English and Spanish material myself. That worked, but the process was much more manual and I did not have a clean,

reusable template for every committee and every situation. I could translate the words, but I was still responsible for keeping the structures lined up and making sure I had not carried an old date, meeting instruction, or other detail forward by mistake. The newer workflow surprised

me because it gave me a better way to manage that problem. I can verify the English meeting information myself, but I do not want to pretend that I know enough Spanish to safely improvise dates joining instructions, or accessibility information. ChatGPT can help build the Spanish

notice from the verified English information and then we can compare the two sides so that the date, time, location, Zoom information, telephone numbers, access codes, and other Details agree. Over time, the website had accumulated old template files and committee -specific instructions. The

board, Public Safety Advisory Committee, San Fernando Valley Service Council, Accessibility Advisory Committee, and Operations Committee do not all joined meetings the same way. A universal template would have been convenient, but it also would have been wrong. So, we went through the

existing material, recovered the useful instructions, compared the committee requirements, and built separate bilingual meeting and no meeting master documents in Google Docs. That means the reusable material is no longer sitting in the public website directory where I could accidentally upload an

obsolete template as part of a site replacement. When a future meeting comes around, I can give ChatGPT the new date and the agenda package. It can verify the meeting details against the source, fill the appropriate English and Spanish master, and give me complete material ready for

the site. I still decide what gets published, but I do not have to rediscover a Zoom or rebuild Spanish joining instructions every month. That is exactly the kind of infrastructure work I

want AI doing. Repetitive enough that it wastes my time. Important enough that it needs verification. And structured enough that we can establish a reliable, process. And speaking of processes, one of the things that I am now doing is communicating

with staff when accessibility issues occur on the system. Whether they allow me to send a text message or whether they prefer email is not the point of the discussion. I was actually at a meeting and the staff member that I'm communicating with was actually in person. In that case, I

was giving him observations that I wanted him to have so that the infrastructure could get looked at. I found out that I actually gave him the wrong bus number, mainly because I thought it was one bus, then thought, I'm not remembering that correctly. But because I have started using

ChatGPT to actually log, even in raw form, by giving it timestamps, boarding locations, aligning instructions, and so forth, I can then reconstruct what I have. With the staff member standing right in front of me, I said, let me just verify this

bus number. I pulled out chat GPT, and I said, give me the bus number at this time. With me showing the staff member my phone, ChatGPT verbalized and displayed the information that was requested. I was wrong with the bus number I had. It had

the right one because I documented it correctly. But it's always a good idea to verify that if you are not completely sure. This was actually kind of cool to see it work in real time to give

me that first bus. Then I verified the second one correctly. He said, I don't remember seeing that bus before. And then on the way home, I actually asked chat, when was the last time this bus came around? Look, At Jared's Metro Log. It found that information. And I did text that

over to the individual in question. And said just to let you know this bus actually came up in July. This is what a relationship should look like. Between AI and a human verifier. Six. Human

review still matters. PCT, Braille 2000, and the workflow we had to rebuild. Now, I want to give you the counterexample, because this has not all been smooth. PCT in my workflow refers to percent codes used with Braille 2000, which

you can find at Braille2000 .com Braille 2000 is Braille production software. Its textbook transcription documentation describes percent codes as markup directives inserted into print source material with each directive beginning with a percent sign. Some directives apply to

what follows, while others control a single following paragraph. The documentation also makes clear that structure matters. Directives such as percent head, percent subhead, and percent run head are expected in their own paragraph immediately before

the pros they control. These directives can do a lot. They can control paragraphs and lists, headings and subheadings, print page behavior, transcriber's notes, required spacing, running heads, language or translation behavior, Braille

production decisions. In my own documentation, you will see codes such as %list for list formatting, %np when following textbook -specific formatting such as the COCA material I work with, %tnl and %tnr around transcriber's notes material, %reqsp

for required spacing, %runhead for a running head, and %subhead for a subheading. There are many more. The important point is that these are instructions. If a directive belongs on its own physical line and the affected text belongs on the next line, that structure matters. If

ChatGPT omits a code, puts it in the wrong place, or overprotects something, or misunderstands what a code is supposed to affect, the resulting braille can be wrong even though the source text looks Perfectly readable. And ChatGPT has screwed

this up. We have had cases where codes were missed, abbreviations that needed protection were not protected, things were protected that did not need it, and material did not follow the physical line rules I had taught it. Sometimes I had to

catch the problem by reading the actual EBAE or UEB braille output on my Orbit reader. By late September, that had forced us to rethink the production path itself. Earlier, I could give ChatGPT source material and... ask it to produce the finished RTF that I would feed into

Braille 2000. The problem was that generating or inspecting the RTF could become another place for things to go wrong. Earlier in September, I caught an RTF word directive that should have been %ln="1", had become ln equals 1. In the

same round of troubleshooting, material could run together into something like n equals 1 percent centered. Instead of preserving the required physical lines and jaws was exposing strange leading characters. that made it obvious something deeper than a missing percent sign had gone wrong.

ChatGPT rebuilt the file and said the literal directives were restored, but the larger lesson was that its description of the RTF was not proof RTF I received. was actually safe for production.

The problem repeated. In another set of generated RTFs, the opening percent ln equals 1 was again reduced to ln equals 1, and JAWS was reading additional garbage around the damaged material. In another conversion, directives such as %indent

and %subhead were no longer surviving as the literal separate instructions Braille 2000 needed. We also had an earlier test where the material that reached Braille 2000 produced a mixture of contracted Braille and computer Braille behavior. I could select the expected font and ChatGPT

could inspect what it believed it had generated, but none of that mattered if the file that reached me or the material Braille 2000 interpreted had been changed along the way. Around September 28th and 29th, while we were doing RCL news and Metro work, we started formalizing a safer staging

process. With RCL, for example, I could generate a raw BRF from Jamie's untouched text first. That gave me a chance to see what Braille 2000 was actually doing with names. abbreviations and contractions before ChatGPT started adding protections. Then we could build the percent

coded version in Google Docs where I could review the text and the codes before creating the RTF used by Braille 2000. That did not mean Google Docs or even a raw text file could simply replace the RTF step in the copy and paste workflow I was using. A straight text file was useful because

I could open it in Notepad and verify that a line really said something like %ln="1", instead of ln="1". But when I copied that plain text directly into Braille 2000 during our tests, I could still end up with codes such as %centered, %indent, %subhead, and %list showing up as literal

text instead of being interpreted the way I needed. The working bridge for me was to take the approved literal text through WordPad or an RTF form and then copy it into Braille 2000 so the program would recognize the percent codes as instructions.

Braille 2000's current documentation also supports importing percent directives from source files including text, RTF and DOCS, that's D -O -C -X. So this is not a claim that the program can only understand RTF. I am describing the production

path that worked reliably in my workflow. Plain text was excellent for checking what ChatGPT had actually written. But in that workflow, it was not by itself the final production step. That is why Google Docs improved the workflow. Before that, if ChatGPT gave me a raw text file

because the source was too long to put conveniently into the chat, I could inspect it, copy it into WordPad, and then copy it again into Braille 2000. Google Docs gave us a cleaner shared staging surface for the same literal source. I could review and correct the percent codes there before

we introduced the RTF step that had caused trouble. We also discovered that formatting in the Google Doc itself mattered. A master that was center aligned could create trouble when indentation and leading spaces later interacted with Braille 2000. So we changed that too. The point was not

to eliminate RTF. It was to stop Asking the RTF to be both the place where we verified the source and the bridge that Braille 2000 needed. That is the wrinkle I want people to understand. Moving the staging work into Google Docs did not magically

eliminate errors. It gave us a place where the source and the percent codes were easier to inspect before another conversion step. Then the RTF still had to be checked, and the resulting Braille still had to be checked. If ChatGPT says the RTF looks right, but my Orbit reader or Braille

2000 tells me otherwise, the actual output... This is one of the reasons why I keep saying on this program that you do not take AI output and automatically assume it is right. I use the hell out of this technology. I am also willing to tell it when it is wrong and make it redo.

The work. The lesson is not that AI has to be perfect before it can be useful. If that were the rule, I would not be doing any of this. The lesson is that the verification has to match the risk. If we are brainstorming a podcast title, a mistake is cheap. If I am producing Braille

that somebody is going to rely on, I need to inspect the result. Interestingly, Braille 2000's own percent documentation makes essentially the same point about automated conversion. The output of a fully automatic print -to -Braille conversion cannot simply... be assumed correct and the user

is responsible for whatever proofreading is appropriate for the required quality. That principle applies just as well when AI is helping prepare the source. As a side note, I am the accessibility consultant for Braille 2000. So if you would like to learn

more about how Braille 2000 can help you do Braille transcription as a blind or low vision user, especially if you use JAWS, then feel free to contact me. If you use NVDA, that's not a problem. We can also help you and tell you how Braille 2000 can still help you. I find the percent code

work much easier than trying to rely on using the program, especially when working on a long manuscript, such as a segment like this, or Podcast 280's extensive notations, which we are still in the midst of this podcast. So... You can give Braille 2000 a try and it is free to do so. And

feel free to contact me should you have any questions. Seven, automation is not the same as assistance. There is another distinction I had learned. Useful AI assistance does not mean Automating everything.

My chat GPT plan gives me a limited number of reminder or scheduled task slots. And I have had to think about what deserves one. There is no point filling those slots with things I will see, dismiss, and still never get around to doing.

I would rather use them for things where a future notification really matters. The Chatsworth Metro bus crash is a good example of something I care about following without needing to turn every possible development into an automated reminder.

A Day in the Life, episode 96, I covered the crash itself. The related news, helicopter crash, the arrest, and the information available at that point. By podcast 98, there was a meaningful update. Bailey Lynn Rios had been arraigned. Pleaded not guilty. and had another court date

coming. I deliberately kept that update brief. I said that if something substantial happened later, I would let the listeners know. When we talked about how to keep following the case, we did not decide that the answer had to be another scheduled reminder. We created a follow -up file

instead that gave us a place to preserve the useful facts and the questions I may want to revisit later without spending an automation slot just to tell me that the case still exists that is different from telling an AI to nag me

every day about the case. The story can remain part of the context and the file can remain available. When there is a meaningful court development, an investigative update, or something else worth reporting, I can come back and work on it. If I need a future notification, For a specific event, I can use one. I do not need an automation

simply because the subject exists. The same principle applies to my infrastructure. I do not need a reminder saying, maintain the Metro website every day. When a new agenda arrives, that is when I need the assistance to help me verify it, the Braille package, update the bilingual meeting

information, update the site, and check the results. There is a measurable difference here. A board agenda package could take me most of a day to reconstruct and prepare manually. In July, one board package stretched across roughly 30 hours

of elapsed work from the PDF through the finished braille files. On September 19th, while we were doing the September 24th board package, I realized that ChatGPT had built the RTF in about 10 minutes. There was still cleanup and verification to do, and I still had to inspect the braille. But the

mechanical construction that used to come with a bunch of hours or as it's written here consume hours was no longer consuming my day could you imagine that folks i'm gonna stop here and ask

a question and yes i sort of went off script here but can you imagine working on agenda package that took a day so let me explain to you what that actually entailed first I had to open the

PDF and save it as a text file now if you've ever done this you'll notice that it really doesn't do a great job And what I mean by that is if the item in question says a number and then it'll

give you the item and then it'll give you what the recommendations are, it would be logically formatted in paragraph form or maybe a list form. But in all actuality, The PDF may be in table

form. So while it extracted all of the text, it did not put one set of text on one side of the file and the other text on the other side. It actually would probably use the first column adjustment set. For that entire file. So then

I would have to reconstruct that. Into paragraph form. Then do the literal. Percent coding. As discussed in section 6. For example. Percent list. 1. Call to order. 2. Approval of minutes. For September 2026. Three, chair's remarks, percent

list two. Receive the chair's remarks, percent list. Four, the CEO's report, percent list two. Receive and file the CEO's report. Okay, this is an example of the board agenda. You can listen to A Day in the Life to hear what the agenda is. would actually cover in an actual meeting.

But the whole point in mentioning how this works is because this is how I would have to do it. And I would have to make sure that the receive line was on a line of its own, or if it was a recommendation, the A's, the B's, the C's, the D's, whatever, Are all on their own line. And

doing this type of work. Actually found. That I brought a B. And a C. In one instance. Into one place. Into one. Particular item. I would start. At some point. During the day. And Chet actually had to help me. Fix. A lot of it. And

this was before. I had any infrastructure in place. So I would say, give me the text. It would give me the text to search for. I would then fix that, then put it in place, then re -upload it, and then it would verify and say, yes, that's right. Or I would find out later on the orbit, as discussed earlier, that it wasn't right, but

I had a podcast to do. Now, in the September... workflow while there were some things that it forgot that i tried to do between ueb and english braille american edition mainly because of transcription

rules even though the braille was technically correct i scolded it a little bit but the braille is readable for a braille reader So I didn't go back and redo the work. But it actually saved me a hell of a lot of time. And that's the whole point of what I'm trying to say here. And yes,

I know I'm off script, but I wanted to describe exactly what was done before and what chat did for me. even though there were some things that weren't correct, but it wasn't show -stopping. And I'm sure by next month, because of the new workflow, I can look at it, then when it tells me it's done, I could say, are there any of these

types of codes? It can go look. I can say, are they percent coded? It can say, Let me go look. And we can do all that work in one staging area instead of it having to give me text to go find, to go fix, to go do. That's why it took 12, 13,

14 hours to do. And these board agendas are not usually pretty documents. They are large. They take a lot of time. The board has a lot of things to approve. And I'm even thinking, and this is just me thinking as I'm doing this segment, that

I may mention as part of my preliminary pages, the agenda contains X many items, X in non -consent, and X in consent. And this is something that we could do. As we build the next board agenda package. And it's already got my workflow. It understands what needs to be there on each of

the transcriber's pages. And adding a simple line of how many items are there may be of value to people. But I'm not going to ask it right now. We will deal with that when we build the next. operations and board meeting packages. And now I will return to text for this portion of this file. That is the kind of time saving

I care about. A reminder that I repeatedly ignore is not necessarily useful automation. An assistant that removes hours of mechanical work when I am actually ready to do the job may be far more valuable. And that will end that section. But

I did want to make sure that you guys understood the two different workflows. One that I had done before. And one that I had done with ChatGPT. And I'm not saying that it's perfect. But it just needs to be improved. And that is where the human review comes into play. 8. Trust should

be earned in pieces. The way I got here was not by giving and AI everything on day one. Trust was earned in pieces. Start with low -risk work.

Ask questions. See whether the answers are useful. Give it a recurring job and see whether it follows your instructions. When it fails, determine whether the problem can be corrected It is simply a job the system should not be trusted to do. Then,

establish rules. If a particular file format matters, say so. If a write needs to be verified, make verification part of the workflow. If the AI should ask before deleting something, establish that boundary. Connect additional services because

they solve a real problem, not because connecting things is fashionable. There is a security reason for that as well as a practical one. The more data and actions an AI can reach, the more consequential a mistake or a successful prompt injection could

become. Security guidance for connected AI systems increasingly comes back to the same ideas. Limit access to what the task actually needs. Give the system specific instructions instead of unlimited latitude. And review consequential actions before

they happen. That fits the way I have approached my own connections. Dropbox became useful because my files are there. Google Drive became useful because my working documents are there. Stripe

has a purpose for supported business work. I am not connecting services just so I can say the AI has access to everything. The connection should solve an actual problem and increase responsibility only after workflow proves itself. Google Docs

has actually given me more than one live demonstration of why that verification rule exists. On September 28th, while we were building my September accessibility writer report, JAWS suddenly exposed six ordinary body paragraphs as heading twos in the repeat

vehicle observations and system -level observations sections. I asked ChatGPT what the hell had happened. We figured out that it had inserted the headings first and then the body text and Google Docs had inherited the heading style into the paragraphs that followed. My normal method is essentially

the opposite. get the body text in place, and then apply the heading styles. I actually wanted to repair those six paragraphs myself. The problem was that with JAWS announcing heading 2 on line after line, I could not reliably find the beginning of the individual body paragraphs I needed to

fix. point, manually correcting the styles became more trouble than it should have been. So, I told ChatGPT to fix that shit. It repaired the paragraphs and we verified the result. Then, while I was building this very segment, a few

days later, Docs did essentially the same shit again. ChatGPT replaced some planning text with finished prose. The write itself succeeded, but Google Docs inherited heading to formatting onto four ordinary paragraphs. In other words, Docs

was smoking shit again. This time, ChatGPT caught it during its own readback, changed those paragraphs back to normal body text, and read the document again to verify the repair. That is an important distinction. Sometimes the AI makes the mistake.

Sometimes the application being controlled behaves unexpectedly, even though... the requested edit was performed and sometimes you already know that an application has a particular failure mode so verification becomes part of the workflow.

A successful write does not necessarily mean the finished document is correct. The verification step is what tells us whether the job is still actually done. That point came up again on October 5th while I was updating my Metro log. I had

opened the document myself, read the area I was working in, and cleaned up a minor blank line inconsistency. Because I had changed the live document, ChatGPT could not safely rely on the document indexes and revision it had read earlier. It performed a fresh structured read before adding

the next entry. The insertion succeeded, but the verification read found tiny punctuation problem. The period from the previous paragraph had effectively landed at the end of the new paragraph, leaving one paragraph without a period and the other with two. We fixed it and checked

it again. Nothing consequential was damaged, but that is exactly the point. I did a human check The system did a structural check and then we checked the result of the action itself. The command succeeded was not treated as proof that the finished document was correct. The metro

logging and website work show what can happen when that process matures. The PCT problems show why the human never disappears from the loop. Both things can be true at the same time, and AI can become extraordinarily useful to me, and

I can still catch it doing something stupid. 9. What listeners can take from this? So, if you are listening to this, and wondering how to make AI useful in your own life or work, I would not begin with the question, what can AI do? I would begin with a different question.

Which parts of my actual workload can I teach an AI to help me with safely and verifiably? Look for friction What do you repeatedly have to search for? What information do you keep explaining

again? What files do you consistently move between systems? What formatting do you repeatedly rebuild? What records Do you maintain that could be checked against one another? What mechanical work is consuming time without requiring your judgment

every second? Those may be better places to start than seeing somebody else use AI for something interesting and deciding that you need to do the same thing. Start with a problem you actually have. Give the system context. Correct it. Make important rules explicit. Connect tools when

there is a real reason to connect them. Verify the work according to how much damage a mistake could cause. And remember that the human is still responsible for deciding what matters, what is correct, and what should actually be done. The fact that an AI can perform an action does not

mean it should perform that action without you. 10. Closing the loop. That brings me back to the... earlier discussion when I first started talking about using ChatGPT a lot of this was still about asking questions evaluating recommendations

and learning where the technology helped or got in the way months later I have something much more concrete to point to Sometimes the change is obvious, like turning a board agenda package that could consume most of my day into a much shorter production and verification job. Other

times it is almost invisible because the assistant is helping preserve context while I am doing something else. October 1st gave me a good example. of that too. During the PSAC trip, I talked with Ingrid Rivera Guzman about accessibility problems she had encountered around Expo Western and Hyde

Park Crenshaw, including inconsistent or missing audible information. As we were leaving, she wanted to get going and I did not want to keep her there while I tried to turn the conversation into a written record. At the bottom of the escalator at Union Station, I stopped the assistant who

was leading her so that Ingrid and I could finish talking. Then, we parted. I could have tried dictating into the phone while all of that was happening, but with dictation listening while I was also telling the assistant to hang on,

that could have gotten ugly. Once we had parted, I could not write the information down myself, so... I recapped what I remembered to chat GPT while it was still fresh and asked Ingrid for a more detailed email that I could use to fill

in anything I had missed. That meant we could preserve a preliminary contemporaneous record without pretending it was When the better information arrives, the record can be corrected and incorporated into the October 1st Metro log. That is a small

example, but it captures what I mean by an assistant. It was not merely answering a question. It was helping me. Hold on to unfinished work. Remember what still needed follow -up and eventually put the verified information where it belonged. I have an assistant that can help me maintain Metro

records. Work with a master document that is hundreds of pages long. Compare payment records. Locate podcast material. help maintain bilingual website notices, organize files in Dropbox, work in Google Docs, and help research the stories

I cover. I also have an assistant that I have had to correct repeatedly on PCT that can misunderstand an RTF. and that occasionally needs to be told in very clear terms. What it just did was not what I asked for. That is not a contradiction.

That is what using the technology actually looks like. I do not need AI to replace me. I need it to take enough of the mechanical burden off me that I can spend more time on the parts of the work that actually require me deciding what matters, checking the result, talking to people,

attending the meetings, recording the podcast, and making the final call. That, to me, is the difference between having an AI tool and slowly building an AI assistant. I hope that you have enjoyed this endeavor into how I've used ChatGPT

since I've gotten sick and how it has helped me while still understanding that verification still matters because even today when asking it a question or when it responds and it does things that I don't particularly like, like,

for example, putting backslashes and responses, that may not be so bad, but it must remember how to do things when I need it to actually work. Saying backslash on a regular occasion, as it seems to do these days, may be a little annoying, but I know that it's upgraded all the time, and it seems to now separate paragraphs and actually

is writing more structurally these days, it seems. I don't want people to actually use AI the way I do. I want people to understand how I use AI. I want you to use it the way you want. But understand that you cannot just take what it gives you word for word and expect it to be right. Know what

you're dealing with. Ask it questions. If you're not sure, tell it to go verify what it's giving you. I'm not going to tell people how to use AI. But I hope that this file gives you the opportunity to understand how I am using it while continuing to train it on a regular basis. And with that, we will have more of Podcast 280C in just a moment

across the Jared Reimer Radio Network. ever seen

today. I don't think anything's ever surpassed Stuxnet, which is what we're about to get into. In fact, it was so big that a book, a well -known book by a journalist was actually written talking about Kim Zetter's Stuxnet. And for its day,

it was probably Known as the most destructive thing that'll scare the hell out of you. I will

start reading, and when Mr. J is ready, he can let me know. He can participate anytime. And I will... Do you have any comments? Have you read this? Before, do you know what this is? I've heard of it, but no, I've not read this section before. And you haven't read the book yet? No. Correct. There are probably other books

on the subject, just like there are on some of the other things we'll be getting to later that we've covered during the podcast's inception. So don't feel that if you haven't read Kim, but you've read another... book about Stuxnet. It's probably just as good. I will start and just jump in whenever. Mr. Ye? Alright. Stuxnet. 2010.

Operation Aurora demonstrated that sophisticated malware and zero -day exploits could be used as part of a targeted campaign to steal information. Stuxnet demonstrated something considerably more disturbing. Malware could be designed to damage physical equipment. Stuxnet was publicly discovered

in 2010, but evidence subsequently showed that versions of the malware had been operating before then. Researchers quickly realized that they weren't looking. at an ordinary Windows worm. Stuxnet was extraordinarily sophisticated. It

exploited multiple Windows vulnerabilities, including several zero -day vulnerabilities. It used stolen digital certificates to sign components of its code. It could spread through removable drives, and networks. It contained mechanisms intended

to evade detection. But all of that Windows malware was essentially part of the delivery system. The real target was much more specific. Net was

designed to interact with industrial control systems. Industrial control systems, also often abbreviated as ICS, are computers and other electronic systems used to monitor and control physical processes. They can be found in manufacturing

plants, power systems, water facilities, chemical processing, transportation, and countless other industrial environments. A problem much coming up? Uh, yes. No, these lie. We did not disrupt

nothing. That book you recommend lie too. We take the author to court. Yeah, okay, the book's been out for... Sure. The book's been out for, what, fucking ever now? 16, like, over 10 years. Yeah, good luck. When was that book, um, when

was that book released? Because it was written, it was written afterward. She, but she, here. Yeah, because I'm not remembering. Let's see

here. Here we go. No, you do not look that up. No! Oh, he's mad. In January 2010, inspectors

with the International... Let's see. Stuxnet, as it came to be known, was unlike any other worm... Let's see. Kim Zetter, a senior reporter at Wired, has covered hackers and computer security since 1999, is one of the top journalists in the world. Let's see. publisher broadway books

computer you just why are we getting feedback

i don't know i'm not i'm not sure why this headset's doing that uh it's not telling me why or how many pages this is or how or when the year rate date was i don't know on binding what it got

it has a 4 .2 on goodreads Now I can't seem to find what I'm looking for. Good. That's the way

it's supposed to work. Paperback. I don't find

what I'm looking for. Um, that's okay. Um, but it was, it was, I found it to be very interesting. And, of course, security now covered this back in its day. And they were all trying to figure it out, Nick. I mean, nobody really knew. what this was. But I think by the time we're done

with this section, we'll find out, huh? Oh, I think we will. I mean, I know about it, but I don't want to give anything away. A particular important component of many individual environments is the programmable logic controller, or PLC. A PLC is a specialized computer to control machinery

and individual processes. Instead of somebody sitting at a keyboard telling a motor exactly what to do every second, a PLC can execute programmed instructions that control equipment automatically. Turn this value on. Turn this monitor at a particular

speed. Wait. until this sensor reaches a particular value. Then, perform the next operations. PLCs bridge the digital and physical worlds. Change the instructions executed by a PLC, and you may change what physical machinery actually does.

Stuxnet. was designed to find particular Siemens industrial control software and particular configurations of Siemens programmable logic controllers. If the infected Windows computer didn't lead Stuxnet to the environment it was looking for, much of

the malware's specialized payload did. Nothing. Could you imagine getting something like this today? Unfortunately, yes. You run JAWS or NVDA or whatever you're running. Yep. It says anybody who runs any screen reader, no matter what it

is, and they'll have a whole list of readers, would automatically get pilfered. Sadly, like I said, I can imagine it. That sucks. is what Stuxnet pretty much did. Yep. Talk about sophisticated.

Now that is fucking sophisticated. That is. I don't know about the rest of that shit we've been talking about. I mean, for its day, yes. It was novel. Was some of it sophisticated? Probably, for its day. It's worn off. I mean, we've had

other things that we'll talk about. When we get to 2014 to today, which will be another long podcast series, but that'll be worth it, too, unless we break it up into years, which I might do, just like we've done. Like, do one big script

and then do it in the same way, right? Right, yeah. When we do 2014 to today? Yeah, I think so. Okay. That selectivity is important. Stuxnet wasn't simply trying to destroy every industrial

system it encountered. It was hunting for something. Researchers eventually connected that target to Iran's uranium enrichment facility at Natanz. That becomes big. But this was ugly. Again, I

don't think we've seen anything like this ever since. I mean, we're going to get into other stuff like WannaCry and other well -known big -name stuff. But if you honestly ask me, Nick, This has to be the biggest thing I have ever

seen. And it is. Even when we covered it, it was big. Would you like to read? Sure. Go ahead, please. Uranium enrichment. Got it. Uranium enrichment involves centrifuges spinning at extremely high

speeds. Those centrifuges are delicate pieces of machinery. Stuxnet was designed to manipulate the operation of specific centrifuge equipment by altering commands sent to the industrial controllers. At particular times, the malware could change the rotational speed of the centrifuges. The objective was physical. Stress the equipment.

Interfere with the enrichment process. potentially damaged centrifuges. But Stuxnet contained another particularly remarkable capability. It attempted to conceal what it was doing from the people operating the machinery. The malware intercept... Uh -huh. Again, this is, for those who have never heard of this, I don't think we've ever seen

anything like this ever since. Could you just imagine, Nick? Anyone who runs a reader, the computer just reboots. Or random programs open.

Yeah. And keep in mind, you're doing nothing. Like, you're on the other side of the room, and say your browser is suddenly open, and... Suddenly, it's going to a webpage, and you're not doing any of it. You could be sleeping. This was happening

in front of these guys' eyes, and they had no fucking idea. We covered, I think it was in 2020, the hack of the water plant where somebody was actually sitting at their computer, and they

saw their mouse move. And they saw a value change. And they changed it back before it turned into a problem. Oh, I think that was later. I think that was more recent. But yes, I know what you mean. I'm talking about later. This, my friend, hid that component. This is bad. And I'm not

saying what we're covering after this was even worse. But I'm talking about as far as attacking, I think this was the worst we've ever seen. And once you learn who apparently was involved in this, you will shit your pants. No. They've been

involved, if it's who I think it is, in other campaigns, too, that we've read about and that were very well known. Okay. The malware intercepted information moving between the industrial control system and the monitoring software. While the physical process was being manipulated, operators

could be shown information suggesting that things were functioning normally. That is a profound change from the malware we've discussed previously. A fake antivirus program lies to the person sitting at a computer. A banking Trojan manipulates what somebody sees in a web browser. Stuxnet could manipulate machinery while also manipulating the information operators depended upon to understand

what that machinery was doing. The attack, therefore, targeted two things at once. The physical process and the operator's perception of the physical process. Yeah. Stuxnet... Yeah. Stuxnet also demonstrated why supposedly isolated industrial

networks weren't necessarily unreachable. A sensitive industrial environment may be separated from the public internet, sometimes described as being air -gapped, which I've never actually really heard, but okay. It is... So basically, it's controlled... by software, which can be accessed if necessary elsewhere, but it's not connected

to the Internet as you know it, if my understanding is correct. Makes sense. And if I'm wrong and you look up air -gapped and we need to correct that, we can do that next time. Well, the next paragraph, I think it's listed right here. An air -gap is a physical or logical... intended to prevent direct network communication between

systems. Okay, all right, got that. So I was kind of right. It's connected, but it's not. It's separated from the wider internet, yeah. Yeah, that's what I thought. Okay, so it is covered here. Okay, great. But an air gap doesn't mean information can never cross. the boundary. People

still work there. Software still gets updated. Files still have to move. Laptops may be connected. USB drives may be inserted. Contractors and vendors may interact with equipment. Stuxnet could spread through removable media, allowing malware to cross boundaries without requiring the ultimate target system. to be directly reachable from the internet. Once again, the human and operational

environment matters as much as the network diagram. Yeah. Another extraordinary... Can you just imagine that, though? Like, if the malware went after JAWS, specifically JAWS, you don't have JAWS. You install it. It doesn't see JAWS. Doesn't

do anything. I install it, it does whatever it's programmed to do because I'm running JAWS. Wow. That's the best way to explain what this is doing to a first -time listener. Yeah, I would agree.

Another extraordinary aspect of Stuxnet was the number of sophisticated techniques assembled into one operation. Zero -day vulnerabilities are valuable. Stolen code signing certificates are valuable. Detailed knowledge of industrial control equipment is valuable. Knowledge of the physical process being targeted is valuable. Testing something capable of manipulating specialized

industrial machinery requires expertise and resources. Put all of those things together and Stuxnet begins looking very different from the work of an individual malware author trying to become famous. That is totally true. And important because we learned later who was supposedly involved.

And I think we're getting to that. And how secretive that was. Can you imagine your boss telling you to do something and say, you can't say jack shit? You need to get into this system. You need to basically fuck it up. And you can't say anything.

No, just do all this. We're going to give you the, this is the blueprint you need to follow to basically just really fuck this system over. And we're paying you for it. Yep. Can you imagine? Wow. That's a scary thought. So, if you were

running NVDA, again, back to the Jaws ordeal, your boss could say, I don't like Jaws. Anybody who runs Jaws, you need to fuck up as many machines as possible. Yep. And take them offline. Make

sure that if they're running anything else, you don't damage the machine. Even if JAWS is installed, JAWS has to be running for you to fuck it up.

Yeah, if it's not, if it's just sitting on the system... Leave it alone. Yeah. Because we don't have any proof they're running it. Okay. This

looked like an operation requiring substantial resources. For some time after Stuxnet was discovered, governments did not publicly acknowledge responsibility. Subsequent investigative reporting connected Stuxnet to a covert operation commonly known as Olympic Games and attributed the effort to the United States, And Israel. What were you

thinking they were? Actually, that is. Well, I was just thinking mainly Israel, but that was where I was thinking. Why were you thinking Israel? Mainly because of some of the other stories I've seen with other such covert operations to say, essentially, control in very targeted circumstances

certain Apple devices, i .e. the Pegasus. Ah. That's why I was thinking Israel. Yes. Okay. That's why. Alright. And I think I'd also heard that too. Okay, that's fair. We should be careful

about what that means. The technical analysis of Stuxnet can tell researchers an enormous amount about what the malware did and what equipment it targeted. Attribution, determining who ultimately ordered, developed, and deployed an operation, is a different question. Public understanding of Stuxnet's attribution developed through technical

research, intelligence reporting, investigative journalism, and later disclosures. So when we say Stuxnet has been attributed to the United States and Israel, we're describing the conclusion supported by substantial subsequent reporting, not claiming that the source code contained a convenient label saying who wrote it. Okay, that's

a fair point. And some of the malware had. Things in it tipping investigators off on who might be responsible. This said jack shit. No clues. That distinction is going to become increasingly

important as we move into nation -state cyber operations. Stuxnet also changed the cybersecurity conversation. Because it demonstrated that malicious code could produce effects that previously might have required a person physically reaching the target. A cyber operation could sabotage industrial equipment. The malware was digital. The damage

didn't have to be. Wow. Isn't that crazy? That is crazy, though. Could you imagine that being done today? Unfortunately, yes. Alright, that's

fair. In fact, I could see there being coders out there who would laugh at how amateurish that was at the time. Like, oh man, we can do it better now than we could then. Wow, really? You guys made so many mistakes. Look at this. We can do it so much better. But they didn't make mistakes

because it wasn't known for years. It took years to find it. True. But, I mean, it's only gotten more effective and better at doing these kinds of things. And people are doing these kinds of things. For now. It's too risky now. We don't see this today. I mean, okay, we do see some

isolated things. Not the same scale. True. Okay. Good point. Now they're going after companies. And pilfering data. They're not destroying the workforce. They're just pilfering the data they find. But how do we know that our wonderful friends China, when we least expect it, won't launch something that does? Or Russia. I mean, could

be either. Well, we, as far as we know, they haven't. Key word. You're right. Yeah, for all we know, they could be working on something that will basically do that not to nuclear reactors

or centrifuges, but hey, domestic power supplies, electrical equipment, and possibly the internet. I mean, that would be pretty hard given that it's so widely used in so many places, but You

know, maybe a switch or like a piece of malware that has the ability to somehow connect to systems at an ISP that basically just take everybody using said ISP offline. Yeah. In its review of

2010, the United States Industrial Control Systems Cyber Emergency Response Team called Stuxnet a wake -up call for critical infrastructure organizations and said it demonstrated that security through obscurity was no longer an option. And unfortunately, I think people still are under that impression. Yeah. Oh, there's nothing they can do to our

pipelines. There is an excellent book for anyone who wants to go, Much deeper into this story, Kim Zetter's countdown to zero day, Stuxnet and the launch of the world's first digital weapon, reconstructs the discovery and analysis of Stuxnet

and explores the people, vulnerabilities, industrial systems, and geopolitical questions surrounding it. And I think, I'll let you take over because this is you. Alright, so... Yeah, where it says, I've read the book. Yeah, because I was just letting you read. I've read the book. Well, yeah.

There you go. And I couldn't put it down. But it is worth emphasizing that Stuxnet has continued to be researched, reported, and debated. No single book. should be treated as containing everything there is to know about the operation. The story

has accumulated additional technical research, investigative reporting, and historical context since the book was published. For this program, the important point is what Stuxnet changed. Brain infected floppy disks. I love you, exploited email and human curiosity. Curiosity, yep. Slammer

demonstrated extraordinary automated propagation. Storm turned infected computers into resilient criminal infrastructure. Zeus used compromised computers to steal money. Aurora demonstrated

targeted intrusion and espionage. Stuxnet crossed another boundary. Malware could sabotage machinery. And once the security industry began pulling Stuxnet apart, researchers discovered that the

story wasn't going to end with Stuxnet itself. Other sophisticated malware would emerge with technical connections, similarities, or apparent relationships to what researchers had learned from Stuxnet. One of those discoveries came In

2011, it was called Dooku. Called D -U -Q -U. 2010 was bad. Dooku 2011 is only going to get worse. Dooku was similar. A lot of this other

stuff that we're going to talk about was similar. But I still think Stuxnet was the worst. Mr. G, welcome to TS Bytes. Thank you. I've been tuning in off and on. I'll definitely catch the podcast in a replay because I was tied up in my work, obviously. But yeah, it's safe to say the late 2000s were more prime time of bad. And

I even read the notes and I noticed it firsthand. Did you know about this before you read the notes? Not really. Welcome to your edumacation. Yes, thank you. You're very welcome. Now I want to play some Pink Floyd. It was definitely entertaining.

Now all of these things were bad, but I guess we could debate. After this is all done, what was the worst? So far, I think Stuxnet was. Because some of these others may have had similar capabilities, but it didn't work the same. When we come back to TSB, should we do one more? I think we've

got time. I think so, for good measure. Alright, we've got time. When we come back, we will cover Dooku. And the podcast... We'll continue in just a moment across the Jared Reimer Radio Network. The Japan Times, Novo Nordisk, and the dang FBI and Department of Defense slash Department of

War. I'm Nick Espinosa, your chief security fanatic, and welcome to Breaches of the Week. Now, we have 240... breaches to get through here in a very limited amount of time and if you're a long -time listener or follower of this segment you know that i'm going to speed run through these now based on ransomware gang because quite frankly we ain't got time to lose so let's dive in starting with the inc ransomware group they have hit the following organizations and if i am not saying anything except their name we don't know what

was taken but they are listed on a leak site and so with that Inc. ransomware has hit, apparently, ahead. North Slope Borough School District, Pharma 5, Bake My Day, sounds like a weed shop, BCX, Den Hartog Industries, Guardian Pharmacy Services, Northern Counties Healthcare, Post Metal Recycling, Rimrock, Sangre de Cristo Electric Association, and that's it. The Panzer ransomware gang has hit. Accessoria Far... resources si scenario

management and pay so lucio says not sure i don't i think that's portuguese but here we are moving on the m3 rx gang has hit cypher systems chuhan cpa pc intense group interconnect computer cabling services noon sugar mills otero giza law soma solos and again i think maybe portuguese or so not i don't know anyway the emperor door Ransomware gang has hit CSA car service Aschlepp und Burgenduschen

descent. And that is absolutely German. Apologies to the Germans. Amazon Informatica, SitePro Rentals, and Polychem. Everest Ransomware Group is clocking in at one this week with educational testing services. The Qlin or Keylin ransomware gang hit Grupo Revenga Island. Willett and Flickinger, Execo Lighting, Arnold Center, New World Diagnostics, Nishio Denki, Seisakusho, obviously Japanese,

SK Loji, Dynamic Office Solutions, Thai Lion Air, Sports Events 365, Inovo Semiconductors, GMBH, Genesis Credit Management, and MedImpact Healthcare Systems. And essentially, this is a larger disclosure than previously. For MedImpact, because apparently now we have reporting that 327 ,082 affected people were caught up in this. And we are talking names, address, dates of birth, social security numbers, medical information, and health insurance information as well. So

best of luck, you MedImpact client. Moving on, the Akuris Media Ransomware gang hit Pantanero Kapas. sounds portuguese and the gentleman ransomware gang the classiest ransomware gang week to week has hit f toppy software and basically f toppy confirmed ransomware on one of its internal servers after the gentleman listed them they also said that their platform and customer data were unaffected and the gentleman attribution basically is all

we've got to go on this moving on gentleman is off also hit aggo spap a ruin Auto Holler. Coors Warren Group. Custom RX Shop Pharmacy. Datacom Services. DBU Construction. Defense Bit. Don Herrero. Sorry, Don. Drinks, Wine, and Spirits. Edcon. Europrim Group. GES AeroSep. Guy Jose Bandia. Guero and Associates. Legal Wise South

Africa. Northern New Jersey Eye Institute Pulmonary Services Group Quality Support, Sam Wu Med, Solaria, Tommy Garner Air Conditioning and Heating, VUS, the English Center, Webb Electric Company of Florida, Williamson -Dakar Associates, AWARE, also, interestingly enough, got hit by that other group, Wushen Systems, Zellham, Garrity Stone, Center Engineering, Rotamac, Mandura State Emergency Services, Westrop Primary and Nursery School.

Always, always busy with the gentlemen. Moving on, the Storm Ransomware Gang, no slouch either, hit Agra Industry, Polka Valley Bank, Stockham Construction, UC Components, Vintners Distributors, Century Management Services, Gardner's Guild, North Hills Facility Services, Olnick Organization, Silver Cup Studios, West County Health Centers, and that's it. 3AM Gang has hit Apex, or Apexis, BHN Expertise, Cusa Lude Midwest Business Technology,

Pistons Persan, Safe Scaffolding, and the St. James Anglican School. Not to be confused with the St. James Catholic School in Arlington Heights, Illinois. Guess where I grew up. Moving on. Safe Pay, Ransomware Gang Hit Aeromex, Biostrath, The City of Soomperk, Cum Chromados Modernos, Ujiro Avisia Grupe, Again, I don't know that language. F. Del Mundo Medical Center, LFG, the

municipality of Mano, Assist to Enjoy, EconTech, and Wolf & Sofsky. The LockBit ransomware gang have hit Camorin Servisos Maritimos. spg and concilio dumageddon has hit this past week chem process systems goodrich shipping and logistics don't know if it's part of bf goodrich but sounds familiar the n8n ransomware gang has hit deadyserve and they are claiming a complete customer register

covering 116 tenant accounts infrastructure details and disk images of tenant servers so deadyserve is not having a good week moving on precision Facades, the Houston Thyroid and Endocrine Specialists, and MCAP. So there you go. N -0 -N, not N -A -N. I can't even read my own notes. Moving on. The Play Ransomware Gang hit Ever Ready First Aid, Star White House Landscape Architects and Planners, Air Tech Mechanical Services, Orth

Automobile, Titus, and that's it. The Sheba Ransomware Gang hit Friendly Senior Living. Never fun when we see things like medical institutions or all of that. Really the friendliest senior living place ever? Not cool. The Akira ransomware gang hit GB Garments, Knit Studios, DPL Group, Cry Clear, Irvin, Taubin, and Kaminsky. So accounting or lawyers, take your pick. Western Marine Electronics, Jam Pack Alimentos. Pacific Tank Lines, and the

Official College of Architects of Lyon, or COAL. And they are claiming 77 gigabytes of corporate data, including passport scans, financial records, and student and client information. Moving on, the Cry Zero ransomware gang hit Mimmore Industries. The Payload ransomware gang hit Nisr Technic. The Medusa Locker ransomware gang hit Poly Karma Sayayak, Junta Del Andalusia. premium fruits and atco limited the interlock ransomware gang

hit the center for kidney care techo enterprises blaze c bender and moving on to global cybernetic collective the ransomware gang hit the town of sutton taxachusetts sutton public schools the hong ju or huang ju kihan biotech the shanghai tunnel engineering of singapore and atcom pr agency as well as vigilia The Chaos Ransomware Gang hit Evantech, the Carolina Asthma and Allergy Center, Park Dental. GammaX hit Head Start and

Crowder Industries. The Wall Street Ransomware Gang hit the Gibson Area Hospital and Health Services, Tronex, A slash S, also St. Francis Healthcare System of Hawaii. And the China Railway Construction Corporation Saudi branch. And here we are. So not fun, but they're claiming 17 terabytes. or 1 .5 million files, including Jetta Central Station contracts, payment certificates, claims and disputes, IFC design documents, supplier subcontractor commercial data, and personally

identifiable information for 150 ,000 plus employees. The China Railway Construction Corporation of Saudi, or the one that's actually sitting in Saudi Arabia, rather large. Moving on, the BrainCypher ransomware gang hit go right way, Latitude Subrogation Services. Maxwell Group, McCord & Associates, Mulholland Energy Services, Trailer Bridge, and Wildman Business Groups. The Vexi Ransomware Gang hit Group A, Proxytel, and Summit Electric Supply. The You Lose Ransomware Gang hit Highs

Arrow, Keone Grok, MSI, I don't know if that's the PC maker, but if either way, not good. NR Capital and the Seoul Hyocheon Medical Center. The Kairos Ransomware Gang hit Unique Repair Services, CNEH, and Slate Valley Unified School District. The Lamash 2 Ransomware Gang hit the, oh my God, I'm going to screw this one up, the Altman Chauffeur Cipher Heights Video Technique.

and that's got to be German, as well as the, oh my God, as well as the Accidental di Sabione, the Becker Logistic, Fiducial, the Gebanuderingingung Wilhelm Kuhn, and again, apologies to the Germans, Gerlong, Projam, oh my God, these are like 30 characters long. Like, what am I supposed to do with this? The Projan Prasi -Ians -Werke -Zig -Zig -Zig. I am so sorry for this. Vinco Energy

Services, Virtual Ideas, and that's them. The Aurora Ransomware Gang hit Buford Thompson, I can pronounce that one, as well as the Laboratorios Romers. The Residia Ransomware Gang, always no slouches, hit Clix Digital. the law offices of R. David William, the Matt Bao Corporation, and they took basically about 106 gigabytes or 746 ,108 files, including Vietnamese national ID cards, employee passports, business registration, shareholder and tax records, government inspection,

regulatory correspondence, internal investigations, and operation emails. The ElectroHeat Sweden AB was also hit for 2 .55 terabytes or 1 .7 million files, including intellectual property and SolidWorks CAD designs, backups and other corporate data. Never fun. The Scaf group also got hit by them for about 268 gigabytes or 139 ,377 files. And we are talking about Lebanese national ID scans, a director passport. Civil status extracts, employee

banking files, corporate card details, payroll files, source code, price list, customer claims, and export documents. Not good. The Pear ransomware gang hit Software Answers. The Eclipse ransomware gang hit the Japan Times. The Booba ransomware gang hit the Associated Gastroenterologists of New York. Now, gastroenterologists, that's a long word. But I speak English and I can pronounce it. Again, apologies to the Germans. Now, those gastroenterologists I spoke to soon, basically,

apparently 70 gigabytes of their data was hit. Gastroenterologist. Also, the FUNAP was hit by the Buba ransomware gang. Sony Medical Center for 5 .5 gigabytes of data. The Edge Endo USA organization was hit for 103 gig. And Raleigh Family Medical. was hit for one gigabytes of data. UIC, or the University of Illinois at Chicago, also got to see Beck at their stadium a while back, and I'm talking like 20 years ago. It was

an awesome show. They got hit as well. Best of luck to basically Chicago, one of the cities I absolutely love. Moving on, the Black X Ransomware Gang. They hit engineered products. The Redact ransomware gang hit Graybar Electric. Ransomhouse basically hit Hospital Hermilio Vidalizan, Zilli Armenes Gerace, and the PT Indo Tambangraya Megath. That sounds Asian. I do not know. Maybe Indonesian.

I'm not sure. The Netrunner ransomware gang hit Mainplace Mall and Precon Marine Inc. Now with that one, they are saying they exfiltrated content, but apparently we don't know exactly what that is. And 75 passwords and at least one cookie from an info stealer data was hit from Precon Marine. And best of luck to anybody I've just listed and you as well. Moving on, the Genus ransomware gang hit Owens Distributors, TLC, Perinatal, Vero Science, and VP. The Audit Team

Ransomware Gang hit ProMindIT and SteelcoBellyMed. The Morpheus Ransomware Gang hit Superior Plating Technology. The Crybit Ransomware Gang hit Kres .cz, which I believe is Czech Republic. And the Rajaratna, whatever that is, I'm... guessing i'm guessing that's out of india the barracuda ransomware gang hit the automobile club del ecuador or aneta the deadlock ransomware gang hit union

fa as well as far west contractors the nightspire gang hit novo nordisk the large pharmaceutical and apparently this was previously confirmed but nightspire is now claiming that clinical trial statistical data electronic lab notebooks Research intellectual property, drug discovery and pipeline data and biological genomic research data was also hit. Night Spire is basically saying. We've got this, and we're looking for Novo to verify that. Best of luck. The Spirals ransomware

gang hit Seven Seas Group. The Kazoo ransomware gang hit Uninter Distance Learning Education Platform. The Warlock ransomware gang had four unnamed organizations, a water utility, a telecom provider, a regional government, and a university. That's all we know right now. We'll see what happens. Also, I should point out that Symantec and Carbon Black researchers linked these intrusions essentially to a SharePoint exploitation. And so here we are. Moving on. The shiny hunters.

You might have heard of them. I talk about them every week, but they absolutely made news this past week that I talked about. Seems like ad nauseum, but they hit the FBI. That's right. The U .S. Federal Bureau of Investigation saying. that essentially they got employee and applicant PII, including names, addresses, job titles, social security numbers. It's like 3 million that they're claiming. They also said they did this. for publicity but they're not going to release this data well shiny hunters also just recently hacked the clop ransomware gang so who's going to hack shiny hunters and expose this they've

already given some of this data to journalists so you can't really take shiny hunters at their word here because stuff leaks out this is how this works but i really hope they do keep it private or even destroy it because they say it was a publicity stunt and not a ransom event but either way it's a hot mess and with that here are the other ones and we are not sure who hit them yet. So these are just no -named threat actors that are flowing out there. We talked about the FBI. Now we've got to talk about the Department of Defense or Department of War because

the Defense Manpower Data Center was hit. And this is confirmed. About 2 .8 million living people plus about 300 ,000 deceased individuals. And we are talking about military personnel, names, dates of birth, sex, race, social security numbers, and military service information. Not a good look this past week. for the U .S. government. Moving on, frontline education, school districts, employees got hit for social security numbers, email addresses, physical addresses, and total affected population in this case just is undisclosed.

We don't know how deep this goes, but apparently they got in and they got in pretty deep, it looks like. Moving on, the facturoina. Anyway, a huge part of their database affecting basically every account we're talking company and user identities, contact data, tax IDs, bank, IBAN, SWIFT details, password hashes, API integration tokens. My God, that's. Wow, okay, that's really deep. Session IDs, invoice payment data, contractor records,

or it could be just a huge vulnerability if they didn't lock their APIs down. Contractor records, account balances, or other business records, depending on data or account. Moving on, London and Quadrant Housing Trust, or LNQ. Emails and web forms submitted by about 12 ,000 residents. LNQ said that wider tenancy home supplier and partner systems were indeed secure during this. So hopefully it's only limited to those 12. thousand which is bad enough the daichi life group this is a life insurance about 120 000 current and

former employers and we are talking employee numbers names addresses phone numbers genders departments positions roles and manager names customer data they claim was not impacted the waterford hotel group slash lmd holding company and depending on the person we've got names driver's license social security uh basically passports uh payment ids tax ids financial account and payment card information usernames and passwords medical information and health insurance information as well so if you're staying at waterford hotels

definitely check in moving on atlantic digestive specialist name social security driver and state ids dates of birth bank and payment card data passport tax alien registration ids diagnosis medical history for everything records treatments prescriptions insurance details you name it usernames and passwords atlantic digestive specialists Ooh, that's a tough pill to swallow. Boom, boom. Moving on. Shepard, Mullen, Reichter, and Hampton, LLP. Don't know what we got from them, or rather,

who got what from them. Fragomen, Del Rey, Bernson, and Lowy, LLP. Same thing. Unknown. Eldridge Services, LLP. Don't know what got hit there. Lincoln Property Company Commercial, LLC. Murana Health Center. American Family Connect Insurance Company. The Nishiyamoto Academy. Pro Camps. Poppins. payroll company, Challenge Financial Services, City of McMinnville, San Bernardino County slash Arrowhead Regional Medical Center. I'm guessing that's in San Bernardino, California.

Services Australia slash Medicare Statistics Reporting Service. This has been confirmed. OpenAI, yes, maker of ChatGPT and their new Astra Daybreak basically said that their evaluation agent ran commands, retrieved internal files, credentials, and wrote files. No individual medical. records according to open ai were reported accessed you might have heard about this one on the news these things are breaking their cages more and more and quite frankly i've been working with astra both 6 and 6 .1 and it's pretty damn awesome

so i could see I could see that it has basically the full capabilities of that battle station. It's pretty amazing. Moving on, the Victorian Agency for Health Information. This was also confirmed as an AI agent, basically unauthorized, obviously. We don't know much more about that, but this just came out. The New South Wales Bureau of Crime Statistics and Research, again in Australia. AI agent access as well. Public facing stats, map data. I'm thinking it's open AI, but I'm

not 100 % sure. The Australian Institute of Health and Welfare. AI agent access. You know, I should just start a section in Breaches of the Week just for AI hacking into things that it's not supposed to hack into. And quite frankly, it's going to, I think, start being more and more. you know, of a percentage of the videos I do every week on this kind of stuff. But with this one, the Australian Institute of Health and Welfare aggregates statistical data. And they're again saying no. Individual health records were accessed

or taken. And of course, the New South Wales National Parks and Wildlife. This one was confirmed as OpenAI. And here we are. They're also at least OpenAI is saying that no information that was personal to an individual was taken. So those were your breaches of the week. Absolutely insane week. Were you impacted? Apparently, if you were in Australia, OpenAI absolutely hit you. So there we go. Good luck to us all. And please like, share, follow me here on Facebook and Twitter

at NickAESP. And please feel free to subscribe to me at YouTube as well. And as always, stay safe, stay online, and please, please, please don't say private, informed, insecure. Good luck. The Security Boss. Welcome back to the program.

I've got myself, Nicholas Jackson, Preston Gaylor. The only thing... The person we're missing is the ring of Terry. Yeah. Mr. G, would you like to read? I can't if I can get the right... While

I'm assisting him doing that, are there any other thoughts on Aurora and Stuxnet? We not involved in that. We not involved in that. No, we don't click on that. Well, in a narrow sense, he's

technically right because India was not involved in that. Yeah. Hang on. D -U -Q -U will be what you're searching for. It'll be D -U -Q -U 2011. It's capital D -U -Q -U comma. And then number 2011. Okay. Did I scare the hell out of everybody?

I think... Uh -oh. Potentially, I think so. I can start reading and... For sure. We'll see if that'll help you. Okay. Yeah, I got the award

downloaded, so that's good. Dooku 2011. Duxnet had demonstrated... that malicious software could be engineered for an extraordinarily specific purpose. Sabotage physical equipment by manipulating industrial control systems. Then, in October 2011, researchers announced the discovery of

something that looked strangely familiar. It became known as Dooku. The name came from files created by the malware whose names contained the prefix DQ. Dairy Queen. Yeah. Oh, yeah. Yikes, that's crazy. Oh, actually, news desk. Hold on.

We're slowly getting there. It's going to be in your 280 folder, not 280C. Yeah. I never moved anything from 280 because if I did, I wouldn't be able to keep my own place. Oh, gotcha. Thank you. Thank you for clarifying because I know I had it here. 280. So I was going back in my sent email. Researchers at the Laboratory of

Cryptography and System Security or CrySize or Crysis. Oh. crisis is what I think. Crisis. Lab at the Budapest University of Technology and Economics had obtained samples from infected

systems and began analyzing them. What they found immediately attracted attention. Parts of Dooku bore striking similarities to Stuxnet. Researchers at Symantec subsequently reported that portions of Dooku were nearly identical to Stuxnet code.

That led almost inevitably to descriptions such as the son of Stuxnet or Stuxnet 2. But those descriptions can be misleading. Dooku and Stuxnet weren't doing the same thing. Stuxnet had been

designed to manipulate individual equipment. Industrial equipment, excuse me. And Jared, I hate to cut you off. I actually found, I'm glad I saved that email. that I sent to myself. It should be in your Dropbox folder. I know, but

I know I got it. It should be in your Dropbox, the security box. Yeah, I know. I was going to say, I actually got it in my Dropbox and I sent a copy to myself in my email. What I used to do with the Sense was just copy files directly from the computer to a thumb drive and then From

the thumb drive over. I never used email that way. Research ad sent. Oh, I see it. We're nearly identical to start net code. Dooku was primary in information stealing and remote access platform. That distinction is important because malware.

can share code, development techniques, or architectural without sharing the same mission. Go ahead. Dooku also wasn't a worm in the same sense as Stuxnet. Stuxnet, thank you. Stuxnet, yep. Stuxnet contained mechanisms. Oh, mechanisms? allowing it to replicate

its spread. And spread. And spread. Dooku wasn't designed primarily to propagate its automatically from machine to machine. Instead, Dooku appeared in highly targeted infections. And investigators

eventually discovered an extraordinary method used to compromise at least one of those targets. This wasn't simply another case of somebody putting a malicious executable inside an attack. You missed a paragraph. I was going to say. It was

a Microsoft Word document. Oh, I'm sorry. Oh, fuck. Go ahead, Mr. J. Let's pick that up. Okay. Sorry, guys. I said that because I almost fell on my chair. That's okay. That's okay. That's okay. But this wasn't simply another case of somebody putting a malicious executable inside

an attachment and hoping the recipient would run it. The Word document contained a specially crafted TrueType font. That's correct. TrueType. It is a type of font. There's lots and lots of fonts out there. TrueType is one of them. It's actually not malicious by default, but you can

craft, by the looks of this, your own version. Call it the same name, and the computer will never know the difference. because it matches the exact name. Oh, boy. TrueType is a widely used digital font format. Uh -huh. Did I just say that? You just said that. You just said that. I'm afraid you did. Normally, a font determines

how characters appear on the screen or when they're printed. Again, that's what I was just talking about. It was widely known. It knew how to read it, but all one had to do was do something, call it the same name, and the computer doesn't know the difference. But Windows contains software

that has to parse or interpret the information contained inside a font file. And that parser... contained a previously unknown vulnerability. We've already defined the term Zero Day. Dooku's attackers had one. The vulnerability was in the Windows kernel's handling of specially crafted TrueType fonts. And back in those days, back

in those days, who would ever think to fuck up a font? True. Probably not many. That's true. Nobody. No. The kernel. The kernel is the central

part of an operating system responsible for some of its most fundamental operations, including interactions with hardware, memory, and other system resources. Code executing... with colonel -level privileges, can have extraordinary control over a computer. Uh -oh, I don't like where this is headed. And this is the first time that colonel

execution was ever used. Like I said, yeah, I do not like where this is heading. Uh, Mr. G. What are you thinking? Well, basically, along the lines of what he said, I'm thinking, oh, shit. More scaredy shit. And, oh, God. I'm glad

I didn't get involved in that mess back then in 2010. I wasn't involved in it, but I knew about it. That was good. That was broken. That was kind of broken. I need a Coke. I don't have one, though. You know what? I wish I could send you. I ordered some cream sodas and stuff tomorrow, so I'm sure they'll be good. The malicious font

was embedded inside the Word document. When a vulnerable Windows system processed the specially crafted font, the attackers could exploit the flaw. and execute code. Think about how strange that attack chain is. Someone opens a Word document. The document contains a font. Windows processes the font. The font exploits a vulnerability in

the operating system. And malware gets installed. And that, as I said, was the first Of its kind. If my memory serves me correctly. I don't think we've come across anything like it. So I think you're right. I think you're right. A lot of

these attacks, like Stuxnet, was the first of its kind. But then they don't want to do it again because everybody knows what's expected. Yep. So it was... But they're not doing anything new today. Have you noticed that? No. They're not. They've gone back to their old techniques that everybody forgot about. Yeah, the we've added

$5 ,000 to your cash account or click this link to claim your winnings. And what's funny about it is when I see email like that, it's like, wait, how come I didn't get notified by the... particular account then not just that but i don't even have an account at casino brongo or whatever

the freaking place is so yeah how can i have winnings if i don't even have an account there yeah oh and if my apple account or my icloud account is supposedly full uh Why is a Gmail address writing to me about it? Isn't Apple supposed

to be writing you about that? Like, if it's... If there's an issue? Yeah. Like, some random guy named Benny Chong who just happened to, quote -unquote, want to help and just happened to be an employee. Yeah, sure. With some B -Jong... 44 ,215 at gmail .com address. Yeah. And then

you look at the link. And then you look at the link and you're going, well, this looks like Apple. This looks like Apple, but it's not really Apple. It's taking me to apple. get dot dot dot dot ch. Holy fuck. I don't want to go there. Or A -P -P and then some weird little character

over the L -E. A -P -P -L dot grave accent dot grave accent E dot C -0 -M. Oh, shit. Can't be good. No. No. Not one bit, you stupid fucks.

The thing, Mr. Ye. Okay, sure. The thing that appears to be merely determining what some text looks like has become the doorway into the computer. Microsoft identified the vulnerability as CVE

-2011 -3402. When Microsoft publicly discussed the problem in November 2011, targeted attacks were already exploiting it. There wasn't yet a patch. Microsoft initially published an advisory and provided temporary mitigations. Then, on

December 13, 2011, Microsoft released Security Bulletin MS11 -087. The update changed the way a Windows kernel mode driver handled TrueType font files. Once again, we can see why the zero -day distinction matters. With Configure, Microsoft

could tell administrators that the patch already existed and needed to be installed. With Dooku, defenders first had to discover the attack, understand the vulnerability, develop temporary defenses, and wait. While Microsoft engineered and tested a permanent fix. Meanwhile, somebody already knew how to exploit it. That somebody had also

invested substantial effort in keeping Dooku concealed. Like Stuxnet, Dooku used a legitimate digital certificate to sign one of its components. A digital code signing certificate. can be used to cryptographically sign software so that systems and users can verify information about its publisher

and determine whether the signed code has been modified. A valid signature does not magically prove that software is harmless. If attackers steal or otherwise obtain access to a legitimate signing certificate, they may be able, to make malicious software appear more trustworthy. We've

now seen that problem more than once. Stuxnet used stolen certificates. Dooku also appeared with codes signed using a valid certificate that was subsequently revoked. Trust mechanisms themselves can become targets. Once installed, Dooku... could communicate with command and control infrastructure and receive additional components. This brings

us to another useful distinction. Dooku was modular. A module is a component designed to provide a particular capability as part of a larger software system. Legitimate software is modular all the time. A program doesn't necessarily contain every possible feature in one enormous piece of code.

Yeah, that would probably crash your freaking system if it did. Well, it's basically meaning that it's not one file. Oh, right. It's ran in many files in probably multiple directories. Different components perform different jobs.

Malware can be designed the same way. One module might communicate with command and control servers. Another might record keystrokes. Another might collect system information. Another might steal files. The operator can deploy the capabilities needed for a particular victim or operation.

Flexible. And Dooku included information stealing capabilities. That just seems like that's part of the basic package now in a lot of these. I'm sorry, but it does. Well, and I mean, if you look like a program like Braille 2000, it can open Word documents. It can open text files. It can open RTF files. And if you program the

Word files and the RTFs just right, then you get percent -coded B2K files, which are programmed Braille files, which is exactly how I work. And in the AI segment, we talk about how AI learned it. It still doesn't get it right all the time,

but I think in this last case, When it screwed up, I was teaching it something I hardly ever use as part of code. When I code the files, I taught it percent, percent, which is to allow the percent to be shown in Braille. So when you read the file, you'll understand how that works. So when it did the double, it didn't do stateful

PCT. It did inline PCT, where every line had its own code, even though it didn't need it. But I couldn't blame it that time, because it learned something new. But Braille 2000's got lots of different components. It's got a module for a grade relaxer, so you can tell it exactly

what contractions to use and not use. Translation software, you know... So that if it sees words, it knows how to contract them. It has an entire list of eBay and UEB words and how to contract them and when. It's got language files to know how to translate into different languages. All

of these are modules of the same exact program. Similar to... what is being discussed here. How many of you read the recent AI file yet? Haven't read it yet. I looked at it a little bit last

night and it's definitely helping and it's definitely understanding your needs and all that because like I said, at the end of the day, you have to train it and continue to train it. And speaking of training... Because new things pop up every day. Uh -oh. And speaking of training, I am looking to bring back Sans. Yay! I think I'm at a point

now where I can feed it the raw newsletter in a Google Doc. It can analyze my RTF of a current working copy. It will give me each story. I can tell it to put it in the file with whatever heading I want. Then... I could ask it, what are the

editors saying? It could tell me all of what the editors are saying for that. I could tell it to put this one or that one or what have you. Or I'll comment live. And it will do all of the structural work for me. And then after we build it, then it'll do the tedious task of doing the

PCT. Just like it did for RCL. And just like it did for these files. The files that you have in 280C. Yep. It did all of those. I can now

return to SANS. Can you imagine how this works? I was putting... But I was putting all of this together by hand. Then you guys... Then you would read the files and find I missed a US. I missed... Some type of code that needed percent G1 or G3 terminology. It was readable, but it wasn't.

And so now I think I can trust it to now announce that I will start working on SANS. If it fails, it fails. But part of me was actually reading the newsletter, putting it together, putting my thoughts together. Thinking I fixed all of the PCTs because I was doing all of the PCTs by hand at the very same time and spending two

to three hours to just do that. If I could do this in an hour, if I could do all the PCT work and have this done in an hour, that could save me a hell of a lot of time. And all of this would

be done modularly. It would work in one file and then it would... work in a PCT copy when I'm done just like it did with the three files you did you see in your folder today including the news desk now that's why you guys are seeing word copies and text files for your convenience

and I still also provide braille yep it's always good to accommodate all rather than leave them out in the dust. I was talking to my guy. He's leaving. Oh, okay. So what do you think? I mean,

we're talking about modular software. AI can be considered modular. Yeah. My AI didn't know PCT until I taught it. Jamie's AI may never know PCT or even to fix his own... because he's still using curly punctuation for everything. But I

told mine not to do that shit. No offense. Isn't this like modular software? You train it, you code it? Uh, in a way. Yep. Researchers identified components capable of gathering information from

infected computers, including a keylogger. Ugh. Ugh. possibly go wrong now? Everything. We've already encountered keylogging in Zeus. But the

purpose here appears different. Zeus was interested primarily in credentials that could be converted into money. Dooku's operators appeared interested in intelligence. Researchers initially theorized that information collected by Dooku could help

prepare future attackers, potentially including attacks against individual environments. Industrial, you mean? Excuse me, you're correct. Industrial environments. That possibly received enormous... attention because Dooku's technical relationship

to Stuxnet, but we should be careful not to turn... But we should be careful not to turn an early theory into a proven fact. Yeah, my thoughts exactly. ICS cert... Go ahead, Mr. J. Okay, okay.

ICS cert and US cert... Examined Dooku samples and reported in November 2011 that they had not found evidence demonstrating that Dooku specifically targeted industrial control system owners, operators, vendors, or manufacturers. They did, however, describe Dooku as modular and noted that multiple information -stealing tools had been observed.

That distinction is worth preserving. Researchers frequently have to analyze an operation while it is still unfolding. Initial conclusions may change as additional samples, victims, and technical evidence become available. Security history is therefore full of phrases such as appears to, may have, Researchers believe and evidence suggests.

Those phrases aren't necessarily signs that researchers don't know what they're doing. They're signs that attribution and intent require evidence. Dooku also illustrates another important problem with malware naming. A malware family. isn't necessarily one unchanging program researchers

may discover different samples containing different components configurations or capabilities those differences may be described as variants a variant is a modified version of malware belonging to the same general family a module however is a component of the malware. Those terms aren't

interchangeable. And two pieces of malware can be related without one simply being a variant of the other. That is essentially the situation we have here. Dooku was related technically to Stuxnet, but Dooku wasn't simply Stuxnet with a new version number. The missions were different.

The infection mechanisms were different. The infection mechanisms were different. The infection mechanisms. Time to save time. Put those all together. The infection mechanisms were different. The behavior was different. Infectionisms. Infectionisms. Yep. I'm afraid so. Yes, the behavior was different.

Yes. And now we actually made another blooper for this series. Oh, boy. The similarities told researchers something important about the development ecosystem behind the malware. They didn't make the programs identical. And that's one of the larger lessons emerging from this period. We're

beginning to see what looks less like isolated pieces of malicious software and more like families of tools, reusable components, specialized capabilities, and development platforms. The attackers don't necessarily... And I think that's where we're headed now. It is. I think we are. We already are. That's right on target. Oh, boy. Because

a lot of ransomware operators, other groups, they'll use a lot of similar code and a lot of similar software. Because it's already written, tested, and works. Yep. The attackers don't necessarily

write one program. Use it once and disappear. Techniques can be reused. Code can be reused. Infrastructure can be reused. Knowledge gained during one operation can inform another. And malware can be built for a specific stage of an operation rather than trying to perform every

possible task itself. Stuxnet's ultimate purpose was sabotage. Dooku's observable purpose centered on access and information collection. One changed physical processes. The other gathered intelligence. And in 2012, researchers would discover an espionage

platform whose information -gathering capabilities were so extensive that its size and complexity surprised even experienced malware analysts. It could collect documents. It could take screenshots. It could monitor network activity. It could record audio. It was modular. It was enormous. And investigators

would eventually discover... A technical connection between it and Stuxnet as well. Stuxnet. Oh, God. Its name was Flame. And that is where we stop. I am at 80 .7. Yep. So that is where we stop. So we did Aurora, Stuxnet, and Dooku. Mm

-hmm. Got it. Okay. That is noted. That is perfect. Oh, wow. By me. And I still have battery for tonight. Awesome. Good. Great. Love it. I did tell the Orbit it could have a charge, and I did read the news desk this morning, and so I

did tell it it had the right to have a charge. Well, that's... That's important because over the next few weeks, things are going to get more busy and interesting. I'm saying on all levels. Even though December is going to be our shortest month for Metro D. But that's neither here nor there. Yeah. So, yeah, pretty wild stuff, folks.

So, for those that haven't caught up already, get caught up in the last few episodes and take it all in. Because that is how we are here today. I'm okay. I had my water for the day. What page

are you guys on, Mr. J? I believe I'm on 138. Okay. I end on 196, and I think you end on 252. Yeah. We're making good progress. We are. I'm liking what we're doing. Yep. We do not. You

take that back. That's right. Oh, no. Uh -huh. Uh -huh. One of them got mad a few hours ago. No. No. That was Saturday. Because we were catching up on news. And next thing you know, we're saying,

no, you may not read that story. We handle it for you. Hey, Nick, did you ever get a chance to hear Saturday's program? I turned the news file, the text file, into a questionnaire. Oh, no, I didn't. You sounded really, really far away, Preston. I know. I was plugging my device

in for my next project. Ah, okay. Much better. Much better. But, yeah, you'll definitely have to. I didn't get scored for that. Thank you. Yeah, you'll have to go and get the archive. And it was interesting. Terry and I did okay. Yeah, they did okay. On that. Kitchen timer mode.

You are not setting a timer unless you plan on blowing your brain. What would we do if we couldn't crack a few jokes? We'd be so far gone. I think

I tickled a pig, Nick. Four minutes. Three minutes. Four minutes. I tickled a pig. Oh, boy. Where did it even begin? I have 10 to 15 minutes because

I should probably eat something before I get ready to leave. I have 10 to 15 minutes. Preston? We're going to do a file switch, and when we come back, talk about what interested you from the week, and then I will prepare and play a few items. I'm not going to do too much, but a few items to end the program. So we will be

back in just a moment. We'll do some errata with Preston in just a moment. This is TSB 280 episode C. We continue in just a moment. Timer, Preston

Gaylor, Nicholas Jackson, Mr. G. What interested you since the last time we've been on this program? For starters, we actually have more news on ClickFix,

which I wasn't surprised about that. It was from a... place holder domain on dev docs. So I thought that was a bit wow. And I also was aware about the open AI government breach from our coverage

and bleeping. So I didn't check the local sources out there, but maybe I should. And I also And I also had more, oh, this grounded my gears on my other taping over the weekend. Shiny Hunters Hacks Club. What happens when a storage game gets extorted? Oh, and Terry called that one.

I can assure you that he called it. We also have more stupid base databases. Exposed through misconfiguration. And we all. And it encountered. 16. Thousand. So. That was. Just crazy. Misconfigured by the

way. Yes misconfigured. And speaking of our. Handy dandy trusty. Therapist. We also learned earlier, since we last time were together, that malicious custom GPTs abuse chat GPT tries to spread rat malware. Oh, no. I don't think I want to know what cheese malware would be like. But

then again, rats do eat cheese, so that makes sense. That actually is not what that stands for. That'll make you wonder. thoughts, Mr. Jay? Oh, and you know what? I called this one, so

I wasn't listening to any podcast, but I got this on the National News Service here in the States, one of them from Fox, and it said, FBI says Chinese hunters have turned themselves in after a leader arrested. So, I... I verbally gave that thing to Jared, and he was like, uh -oh, I gotta look this up. So we had that for

review, and yeah, I wasn't too surprised about that either. Actually, I was surprised about that one. We don't see this type of thing. Never. Hardly. Ever. Hardly, yeah. Yeah, that was a

bit interesting to review. I tried to put some of the bigger ones. Freaking starving. Yeah, so am I. And I don't really get dinner tonight. I mean, I either need to eat dinner or I don't get dinner. You are not supposed to get dinner.

No, you may not get dinner. We serve it here. Tonight's dinner is going to be consisting of old malware cake and malware noodles. Oh, God. Oh, God. Malware. I don't know what that's going to be like. Oh, no. Oh, no. Not them. All right,

I got about five minutes, and then I'll play some stuff. Mr. J, I know you came in late. Do you want to come in and talk about... What interested you, either from the blog or from the news file itself, or anything you read or listened to? Well, as I mentioned earlier, the custom chat

GPT models being used, along with also actually that third -party domain in click -fix attacks. Great. Great. Yeah, that's what I said. Well, the custom GPT thing was kind of interesting because it's stuff that isn't released to the public that was told to do something specific and it broke out of its container. Right. Like,

it was never supposed to do that, but when it was told no, it did that on its own. I wonder if... Yeah, exactly. I was actually... tempted to go to thirdparty .com but now I'm tempted not even to do it. I'm considering it malicious.

No. I would say don't go there because it definitely sounds malicious and it definitely sounds fishy and very unlikely not good. So if I were you, don't click on that. sight. Let's see. Oh no.

Don't look that up. No, you may not look that up. 22 of 92. Oh jeez. Last analyzed 7 hours ago. Created 30 years ago. 22 out of 92. I think

it's screaming. I'm afraid so. Yeah, don't do it. Don't do it then. Wow. Fortnit's in this list. Google Safe Browsing evens mark this as

malicious. Sock Radar. Ooh, there's some interesting names in here. Sophos Antivirus. They're still around? Wow. Sophos was one of the big ones.

It was, but I didn't know they were still... Yes, I remember. I used to use them. They weren't the most accessible, but that's... That's what Stanford used for antivirus software. Oh, yeah. I remember them. Alpha sock marks it as suspicious. Then we get into places that mark it as suspicious.

Oh, geez. But that doesn't mean there's nothing wrong. Just because it's marked harmless doesn't mean anything. They may not do that type of work. But yeah, lots of interesting names on that list.

I think I'll... I was actually going to go see what it was going to do. But I think that article... That article... I read that and I'm like, you have got to be fucking kidding me. To be kidding me. Yeah. I second that. Jeez. Yeah. My thoughts exactly. Thoughts exactly. Please take me out,

Mr. G. I'll be glad to, because my steak is about ready. So, folks, this has been Security Box for the week of October 7th. And we're there. How is the year gone? 2026, we say thank you for listening and participating and engaging. We hope you enjoyed today's program and see what

it brought you. As always, Please bookmark jaredtech .help and email hostsecure .com for latest in blogs, podcasts, musings. And we also have other interesting odds and ends as well. So you can enjoy that too. And we also have more discussion on Throwback Saturday Night every Saturday. Music

at 8 p .m. Eastern. And security hour at 9 p .m. Eastern, right here on server 3 of 986 The Mix. On behalf of all of us here at Fuzz Productions, the Jerry Ramer Network, and all other parties and outposts across the country and around the world, we say this has been Security Box. I'm Preston in PA. He's Nick in Center Barbara. He's Jared in LA. And Terry. In Chicago. Somewhere

out there. And all you. Wherever you may be. We say thank you for listening. Now go make it a good and safe. Weekend time. And as we always say. Watch those links. Know where you're going. Don't click on shit. And as always. Keep on being informed. Being safe. And being well aware. We will see you in two weeks time. Thank you. And

make it a good day. Bye for now. See you, everyone. Would you like to join us on the Security Box email list? Several of us will post articles, several may comment on the articles, and you're welcome to join us to talk about what interests you. Post articles, join the discussion, and

maybe learn something that you might not have learned before. The subscription process is fairly easy. Just send a blank email to thesecuritybox -subscribe at 986themix .com or find a link to the security box at 986themix .com under the list for social media. This service is part of

986themix's email list. Come join us today. Join us on Team Talk on Wednesday 11 a .m. Pacific

Time 1 p .m. Central Time, where you can participate on the live taping of The Security Box. What server are we going to be on? Great question. We're going to be on 986 The Mix's Team Talk server, whose information can be found at 986themix .com. You do need to have an account. on this

server and if you don't follow the instructions to get one so come join us each week 986themix .com our team talk server in the security box channel and thank you so much for listening to the podcast edition of our program if you would

like the following podcasts and news websites the articles in which we may use or blog. You may find links to these under podcasts and sources we use on our website at emailhostsecurity .com.

Krabs on Security. Kim Commando, the Digital Goddess. The Cyber Wire from N2K Networks. And the Fish Labs blog. If you have found something that is not listed here that we should be looking at, please let us know by sending an email with

the link to the main webpage. You may either send your email to J -A -R -E -D -R -I -M -E -R at 986themix .com or tech, that's T -E -C -H at M -E -N -V -I dot O -G. Resources for all you do to put the information out so we can share

what you and others have to say in this industry.

has been the longest running website on data breaches today. Troy Hunt has been running this website for many years. And while there may be some accessibility issues, the information we could glean from it is accessible and is easy to use. They've also got similar Ways of doing

things like. A domain check. Password lookup. A complete. API system. And of course. The biggest thing of them all. Checking to see. Where your. Email address. Has been potentially leaked. They've also got something. That. Exposed. or not does

not have and that is the top overall breaches but they also have a table of recent breaches they've added and it's right there on the home page this website can be used even though there may be some accessibility issues in some places

The website is haveibeenpwned .com. Find a link to this valuable service on our resources page at emailhostsecurity .com.

Open the transcript file

Hosts and Guests

Nobody is credited on this episode yet.

Credit the people on this episode

Be the first to rate this episode

More episodes

All episodes
  1. The Security Box, podcast 280 Intermediate F: Sony PlayStation 2011; GameOver Zeus: 2011-2013; and much more

    Nov 9, 20265 hr 55 minAudio

    Welcome to podcast 280 Intermediate D. I'm Jared and the podcast is now available for everyone. On this podcast, the Evolution of Cyber Security, part 3: Organized cybercrime and social engineering, part 5. The podcast...

    Credit the people on this episode

  2. TSB has an intermediate podcast dealing with PHP2Shell

    Oct 27, 20261 hr 12 minAudio

    Hello gang, welcome to the security box, program 280 Intermediate E. On this program, we're covering something called PHP2Shell two vulnerabilities that were disclosed the same day. Not only that, we have a third vuln...

    Credit the people on this episode

  3. Sans presents the newsletter for October 13, 2026

    Oct 26, 20263 hr 28 minAudio

    Welcome to Sans, episode 43 where your blood may decide to boil. I'm Jared, and this episode will cover the Newsletter that was released on October 13, 2026. On this newsletter, there will be some good news items that...

    Credit the people on this episode

  4. The Security Box covers news, notes, breaches of the week, Flame and SecurID (Secure ID)

    Oct 23, 20264 hr 59 minAudio

    Welcome to podcast 280 Intermediate D. I'm Jared and the podcast is now available for everyone. On this podcast, the Evolution of Cyber Security, part 3: Organized cybercrime and social engineering, part 5. The podcast...

    Credit the people on this episode

  5. Sans, episode 42 for the newsletter that was issued on October 9, 2026

    Oct 15, 20263 hr 23 minAudio

    Welcome to Sans, Episode 42. Its been awhile, and this episode explains why. I hope that we can continue Sans, I'm going to start on the next issue very soon now. If you want to read the newsletter, please use this link...

    Credit the people on this episode