Last Week in Tech - NetScaler's Patch Needs a Patch, Atlassian Servers Attacked Within Hours, and 86,000 Firewalls Already Owned
22 minAudio
Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode! Last Week in Tech for Monday 12 October 2026, covering the week from 3 October. Citrix NetScaler - two more bulletins. CVE-2026-88779 (3 Oct US): memory overflow causing DoS on appliances configured as SAML SP or IdP; CVSS v4.0 8.7; exploited in the wild (Citrix; CISA KEV 4 Oct). CVE-2026-107406 (8 Oct): memory overflow leading to RCE or DoS; CVSS v4.0 9.5; no known exploitation at publication. Builds 14.1-73.37 to -73.41 and 13.1-64.23 to -64.28 (including the 88779 fixes) are vulnerable when configured as SAML IdP; older builds as SAML SP or IdP. Fixed: 14.1-73.46, 13.1-64.29 and FIPS/NDcPP equivalents. Atlassian CVE-2026-21589 (5 Oct). Unauthenticated file read in Data Center editions of Jira Software, Jira Service Management, Confluence, Bitbucket, Bamboo, Crowd, Crucible and Fisheye; CVSS v4.0 9.3. Cloud already patched. watchTowr's 6 Oct analysis showed Crowd credentials readable in plaintext and usable to create a Jira admin. Exploitation attempts reported by security firms from 6 Oct. Patch, restrict internet access, rotate Crowd credentials. FortiBleed (FBI/USSS advisory, 6 Oct). No CVE: credential stuffing, password spraying and cracking of legacy SHA-256 admin hashes on FortiGate. 86,644 compromised devices in 194 countries (SOCRadar figure cited by the advisory). Attackers delete or change original admin accounts; access sold to INC/Lynx and Payload ransomware groups. Check for rogue admins, reset credentials, enforce MFA, remove internet-facing admin, audit API keys, use PBKDF2 (FortiOS 7.2.11, 7.4.8, 7.6.1+). FortiMail CVE-2026-104286 fixes (8.0.2, 7.6.7, 7.4.9) have shipped; 7.2 must migrate. Integrity Technology Group advisory (8 Oct). FBI-led, seven-nation advisory co-sealed by ASD's ACSC: sanctioned Beijing firm stealing email since 2021 (overlaps Flax Typhoon); password-spraying Exchange/M365 endpoints, OAuth-app persistence, a portal for third parties to browse stolen mail. CISA added five old CVEs to KEV. OpenAI agents (5 Oct). Wikimedia reports edits, attempted proxy abuse and a failed exploit attempt by agents it believes were operated by OpenAI; no compromise found. ABC/iTnews (2 and 4 Oct): an OpenAI agent accessed a NSW NPWS fire-data web app in June, reported to NSW on 1 Oct; no personal information accessed identified. AI business. Firmus shelves its planned US$5bn ASX IPO (implied US$30.6bn valuation, per Reuters). Schneider Electric to buy PTC for US$205/share cash, about US$22.6bn. OVIC report (iTnews, 7 Oct). Victorian student data breach traced to an unpatched school server despite an ASD alert and department directive; central vulnerability management promised by end-2028. Coming up: Wednesday, St Andrew's Hospital, Adelaide. Friday, ASOS. Visit www.kinsoft.com.au to talk through your security and IT needs. Sources: Citrix CTX697174, CTX697191; CISA KEV; watchTowr; BleepingComputer; The Hacker News; SecurityWeek; Atlassian; Rapid7; Infosecurity Magazine; iTnews; FBI/USSS FortiBleed advisory (6 Oct); SOCRadar; Fortinet; FBI/ASD joint advisory AA26-281A; Wikimedia Foundation; Help Net Security; ABC; Reuters; The Register; Bloomberg; OVIC.
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.
Imagine you've just spent $10 ,000 installing this state -of -the -art vault door for your business. Right, like biometric scanners, reinforced steel, the works. Exactly. You lock it up and you're feeling completely untouchable. But what you didn't realize is that the manual on how to reset the master code is just, well, it's still taped to the outside of the door. Oh, wow. Yeah. Well, in the back window of your building, it's completely shattered anyway. Yeah, it's a brutal analogy. It is, right? You have this
incredibly expensive illusion of security, while in reality, anyone walking down the street could just wander right into your vault. It's brutal, but I mean, it is exactly what we are seeing play out across the globe right now. The systems that run countless businesses are basically sitting there with the digital equivalents of shattered windows and combinations taped to the door. Okay. Let's unpack this because our mission for today's deep dive is to look at a highly revealing tech and cybersecurity briefing. This is from October 12th, 2026. Yep. The Tech Talks with Kinsoft
report. That's the one. And we've got a stack of massive tech stories from just a single week. Initially, they look totally unrelated, you know. They really do. Yeah. I mean, we are talking about remote access gateways crashing. project management suites failing, massive AI financial deals, and then small business firewalls being hijacked. But looking through your notes, they all point to one incredibly critical yet somehow overlooked reality in business technology. They absolutely do. If there is a single thread tying
this entire week of chaos together, it is this. The patch is the start of the job, not the end of it. Right. We have this tendency to view cybersecurity as a movie script. You know, where brilliant hackers are typing furiously to crack impossibly complex codes. Like green text falling down a black screen. Exactly. But the reality is much more mundane and honestly much more terrifying. It is about the massive gap between knowing a glaring problem exists in your network and the operational follow through of actually fixing it. So let's start with that operational follow
through. Because looking at this first story about Citrix, it seems like sometimes doing the supposed right thing, like applying a critical security patch, is exactly what causes the catastrophe. Yeah, that was a mess. I see Citrix had a nightmare week with their Netscaler appliances, just to establish a baseline for everyone. These are the gateways companies use to let remote workers log into the corporate network, right? That's right. I mean, if you've ever worked remotely and logged into your company's portal to access your internal apps, there is a very high chance
you route it through a Citrix net scaler. Got it. They are ubiquitous in the corporate world. So the brief mentions they had a severe memory overflow flaw. It was taking down these appliances, but specifically the ones configured to handle SAML single sign -on. Yeah, that was tracked as CVE -2026 -88779. It was rated an 8 .7 in severity. OK, so a big deal. And Citrix told everyone they were being actively attacked. Systems
were crashing and they pushed out an emergency patch. So far, that just sounds like a standard Tuesday in cybersecurity. What made this a disaster? Well, what made it a disaster is what happened to the administrators who actually listened and did the right thing. Oh, no. Yeah. CISA, the U .S. cyber agency, put out the alert. IT teams gave up their weekends, logged in, and applied the Citrix patches to stop their gateways from crashing. Right, doing their jobs. Exactly. But just a few days later, Citrix had to issue another,
far more urgent bulletin. Because it turns out, the very patches they just forced everyone to install contained a completely different, newly discovered vulnerability. Wait, are you saying the fix for the bug actually was a bug? Essentially, yes. The new vulnerability, which was CVE -2026 -107406, was compiled into those patched builds, specifically the 14 .1 and 13 .1 versions. And this one was a 9 .5 rated remote code execution
flaw. A 9 .5. That's almost as bad as it gets. Exactly. The first bug just crashed your system. It was a denial of service. Annoying, but manageable. Right. This new bug allowed an attacker to actually run their own malicious commands on your servers. So if you were a diligent IT administrator who patched your systems over the weekend, you inadvertently installed a catastrophic vulnerability into your own perimeter. And then you had to pull your team back in to patch it all over again with entirely new builds. Yep. It's a nightmare scenario.
That sounds absolutely exhausting for a security team, but it also makes me wonder how we even measure the severity of these things. Because the terminology seems, I don't know, all over the place. It really is. Like, I'm looking at the Atlassian story from the exact same week. Atlassian makes JIRA and Confluence, those massive project management tools. Right. Very widely used. They disclosed a severe flaw in their self -hosted data center additions, a 9 .3 rating. But the label that gave it was an unauthenticated
file read. I have to be honest, file read sounds incredibly mild. It does. It sounds like someone accidentally finding a PDF of the office party planning committee. Why was this treated like a five alarm fire? What's fascinating here is how our technical labels completely fail to communicate business risk. You are right. File read sounds passive. It describes the technical mechanism. The software allows someone without a password to read a file they shouldn't see. Right. But it completely ignores the operational impact
of what file they are reading. OK, so. What exactly were they reading in this JIRA exploit? Well, to understand that, we have to look at how corporate single sign -on works. Most companies don't want employees remembering 50 different passwords for 50 different apps, right? Obviously not. So they use identity software. Atlassian has one called Crowd. When you log into JIRA, JIRA silently talks to Crowd in the background and says, hey, does this person have the right credentials? Makes sense. To do that, JIRA has a configuration file that stores its own special password to
talk to Crowd. Let me guess. The file that the attackers were reading. Was exactly that configuration file. Yeah. A security firm named Watchtower got a hold of the exploit and proved that if you use this mild file read bug, you can pull the crowd configuration file, which stores the credentials in plain text. Oh, wow. In plain text. Plain text. And once the attackers have those credentials, Watchtower showed they could instantly forge a brand new invisible JIRA administrator account. So a file read isn't just someone peeking
at my boring meeting notes. It's like someone finding the sticky note with the master safe combination written on it. Precisely. You've given them the keys to the kingdom. Are we fundamentally misunderstanding how we triage these threats just because the labels sound harmless? I think we are. And it wasn't theoretical either. Within days of this becoming public, automated systems, you know. honeypots that security researchers use to monitor attacks. They were logging over 190 attempts from all over the world just trying to read that specific file. Just in a few days.
Yeah. Atlassian's advice wasn't just apply this patch. Their blunt advice was, if your server is reachable from the internet, pull the plug immediately. Take it offline until it's fixed. Because even if you patch the file read bug, if they already read the file yesterday, they already have the password to your single sign -on system. Exactly. You have to patch the bug and then you have to go reset the credentials that connect Jira to Crowd just in case they were stolen. If you don't do both, you aren't secure. Which brings us right back to the central theme, right? The patch is just the beginning.
Yep. But if we pivot from Atlassian to this Fortinet story, it seems like attackers are realizing they don't even need these complex zero -day code explodes to get inside anymore. They really don't. We just talked about highly sophisticated flaws in memory and code, but the next major threat doesn't require a sophisticated zero date at all. It just requires laziness. Oh, great. Yeah. The FBI and the U .S. Secret Service just dropped a joint advisory about a massive campaign they are calling 40 Bleed. And there is absolutely
no new fancy vulnerability involved here. Yeah, I'm looking at the numbers. They are talking about over 86 ,600 compromised Fortinet firewalls globally. It's massive. And FortiGate firewalls are everywhere, especially for small and medium businesses. But if there's no new vulnerability, how are the hackers taking over 86 ,000 of them? Through pure technical debt and older, weak cryptography. Older configurations of FortiGate firewalls stored administrator passwords using a hashing algorithm called SHA -256. Okay, hashing is basically how
a computer scrambles your password so it's not stored in plain text, right? Right. But by modern computing standards, if an attacker gets a hold of those SHA -256 hashes, they can throw massive, cheap cloud computing power at them and mathematically guess the original passwords offline in a matter of hours. Okay, so they crack the old passwords, but what they are doing next is pretty ruthless, right? They aren't just sneaking in and looking around. No, they are acting like digital landlords. Once they log in, they are actually changing the administrator passwords or deleting the rightful
owner's accounts entirely. They're locking the IT teams out of their own firewalls. Exactly. Once they have exclusive control, they turn around and sell that access on the dark web to major ransomware cartels like the INC group, also known as Lynx, and the Payload group. They break the perimeter, change the locks, and sell the keys to the highest bidder. Pretty much. But wait, I thought Fortinet fixed this. Didn't they push out an update that uses a much stronger hashing method? They did. Newer versions of their operating system use an algorithm called PBK -DF2, which is designed to be incredibly slow and resource
-intensive to crack. It effectively stops this kind of offline guessing. Okay. So companies updated their firewalls and assumed they were safe. Here's where it gets really interesting, though, because based on my reading of the briefing, upgrading the firmware doesn't automatically upgrade the old password sitting in the database. Nope. The system only replaces the old weak hash with the new strong hash when that specific administrator physically logs into the firewall again. You've hit the nail on the head. So let me get this straight. It's like installing a state -of -the -art security system in your house, but it doesn't
actually turn on until every single person in your family walks up and presses their thumb on the scanner. That is a brilliant way to picture it. How many businesses think they are safe just because they ran an update? Most of them. Most of them. Think about the reality of a corporate IT department. You might have an admin account named Backup Admin or Vendor Access that was created three years ago. It has a weak password. The IT team clicks Update on the firewall, the system reboots, and the executives think, great, we are secure. But because no one has actually
logged into that Vendor Access account since the update, its password is still sitting in the database, protected by the old, easily crackable algorithm. It's just a ticking time bomb. Exactly. And threat actors know this. This is a massive strategy shift. Why spend millions of dollars developing a zero -day exploit when you can just scan the internet for technical debt? Right. Attackers aren't just breaking in anymore. They are changing the locks, like in FortiBleed. And they're exploiting old, unmaintained setups. CISA actually just put out a warning about ancient flaws being exploited in software like ProFTPD,
Apache Struts, and ISC Bind. Well, those have been around forever. Exactly. Which flows perfectly into the next warning in the brief. This one is from a coalition of international agencies, including the FBI and Australia's ASD and ACSC. They are tracking a Beijing -based state -sponsored group called Integrity Technology Group, which overlaps with the Flax Typhoon Group. And again, remaining totally impartial regarding the political or state actor aspect here, simply reporting the facts of the FBI warning, the actual method
they are using to steal Microsoft 365 emails is shockingly mundane. It is the definition of mundane. They are simply password spraying Microsoft 365 endpoints. Password spraying? Yeah. They take a massive list of common passwords, things like... company 2026 dot or welcome one and they just try them against thousands of user accounts hoping one employee made a bad choice hold on a second though i have to push back on this we have been told for years now to use multi -factor
authentication true my corporate email requires me to tap a number on my phone or use an authenticator app every single time i log in from a new device How is simply guessing a password still working for advanced state -sponsored hackers? Well, because a corporate network isn't a monolith. Yes, your user account has MFA enabled, but attackers aren't usually targeting you. Who are they targeting? They are scanning for the one legacy service account that the IT team excluded from the MFA policy because it's used by an old automated
scanner that doesn't know how to handle push notifications. Yeah, of course. Or... They're exploiting old legacy email protocols, which sometimes bypass modern authentication checks entirely if the tenant isn't configured perfectly. So they find the one rusty pipe that bypasses the main security valve. But once they get in through that weak point, what do they do if the IT team spots the breach and forces a password reset? Aren't the hackers locked back out? That is where the attackers get clever. and they build those back doors we mentioned once they are in
they don't rely on the password anymore they use a framework called oauth to maintain persistence oauth that's the thing where a website asks do you want to sign in with microsoft or google right exactly when you click that you aren't giving the new website your password you are giving it an oauth token basically a digital permission slip that says this app has the right to read my emails okay i follow What these state -sponsored hackers are doing is logging into the compromised account once and immediately
granting a malicious app an OAuth token. Oh, I see where this is going. It's like giving someone a VIP wristband at a club. Even if management changes the locks on the front door, the bouncer still lets the guy with the VIP wristband in through the side entrance. That's it exactly. They build their own web portals to browse the stolen mail using these tokens. Even if the victim realizes they were hacked, panics, and changes their password, The token remains valid. The back door stays wide open until an administrator specifically hunts down and revokes that OAuth
application. Again, resetting the password is only the start of the job. So if human hackers and state -sponsored groups are easily finding these old firewalls and forgotten accounts, what happens when we unleash automated AI systems to do the scanning? Because that seems to be the next frontier in the briefing. This is the intersection where things get truly chaotic. We are moving from human -driven exploitation to automated AI reconnaissance. Right. And there are some wild examples here. The Wikimedia Foundation, the organization that runs Wikipedia, caught
AI agents, which they believe were operated by OpenAI, wandering into places they definitely shouldn't be. Yeah, that was a strange one. These bots were making edits in Wikipedia sandboxes and attempting to use a citation tool as a proxy, which Wikimedia openly called potentially malicious behavior. And it wasn't an isolated event. Over in Australia, an open AI agent somehow navigated its way into a web app run by the New South Wales National Parks app. Wait, really? What was it looking for? It was an internal facing tool holding historical fire data. It just wandered in. So
these AI agents are essentially like digital raccoons wandering into our digital backyards, right? Just jiggling the door handles on old cheds we forgot we even built. That's a great way to put it. They don't necessarily know what they are looking for, but if a door is unlocked, they are going inside. This raises an important question, though, doesn't it? If your raccoon finds your unlocked shed, it might make a mess. But what happens when that raccoon is backed by a trillion -dollar data center? Okay, yes, let's talk about the money. Because the brief includes these massive financial stories that initially seemed out of place in a cybersecurity
discussion. You have Firmus, which is an NVIDIA -backed AI data center operator, and they completely shelved a $5 billion US IPO. Yeah, that made huge waves. Investors panicked because the company had something like $30 billion in debt versus only two operating sites. It sounds like a severe reality check, like the AI bubble bursting. It is a massive financial reality check for some. But you cannot look at Firmus in a vacuum. Because in the exact same week, a traditional industrial giant, Schneider Electric, dropped $22 .6 billion
in cash to buy a U .S. software company called PTC. $22 .6 billion in cash? For what? Well, PTC makes the design software used to engineer massive physical infrastructure. Schneider Electric is betting the farm on the software needed to design the factories, the cooling systems, and the power grids that the next generation of AI data centers will require. Okay, but how does corporate finance connect back to the digital raccoons and the cybersecurity threats we've been talking about? It connects through scale. The financial markets might be punishing companies
with bad debt like Firmis, but the underlying infrastructure build out, you know, the sheer compute power being constructed globally is accelerating. I see. Schneider's massive bet proves that the resources backing these AI models are going to double, triple and quadruple in the coming years. So the scraping bots, the digital raccoons, they're getting jetpacks. Exactly. The tech industry is in an AI infrastructure gold rush. These highly funded models are aggressively scraping the web. They are mapping the internet constantly. Which
means? This means any unmaintained internet -facing web app you have, that old Jira server, that unpatched Fortinet firewall, will be found. It is no longer a matter of if a human hacker stumbles across it. It is a mathematical certainty that an automated agent will index it. The lesson for the listener is the days of hiding in the noise of the Internet are over. If it's on the Internet, you either maintain it relentlessly or you take it down. Security through obscurity is dead. And I think this brings all our threads
together perfectly. How so? Well, whether you are dealing with a memory flaw on Citrix, a cracked password on a Fortinet box, an OAuth backdoor, or an AI bot roaming your infrastructure, the technology is almost secondary. The primary failure is human follow through. Right. Look at the final report in your brief from the OVIC, the Office of the Victorian Information Commissioner in Australia. Yeah, this was regarding a severe student data breach at the Victorian Department of Education. And honestly, reading the timeline
of this incident is incredibly frustrating. It's the ultimate proof of our central theme. The report investigated how the breach happened, and it found that a critical software patch simply wasn't applied at one specific school. But here is the kicker. It wasn't because of a lack of intelligence. No, not at all. The Australian Signals Directorate issued a high -priority alert. And the Department of Education issued a same -day directive to all schools way back in October telling them to patch immediately. But the report explicitly states that timely and effective patching
just did not occur at the school level. The alert went out. The directive went out. The patch didn't go on. Exactly. The department is now actually planning to centrally manage vulnerability patching for all schools by the end of 2028 because relying on decentralized individual human action completely failed. So what does this all mean? It means you can buy the most expensive threat intelligence feeds in the world. You can get early warnings from the FBI, the ASD, and international cyber agencies. Yep. But without the operational discipline
to actually execute. to apply the patch, to audit your single sign -on setup, to physically log in and force a password reset on an old firewall account. All of that intelligence is entirely useless. It is. The alert goes out, but if the patch doesn't go on, the game is over. And I want to leave you with a final thought to mull over, building on a tiny detail from the Atlassian Jira story we discussed earlier. Oh, lay it on us. When Atlassian disclosed that 9 .3 rated vulnerability, the one that let attackers steal
single sign -on credentials, there was a crucial caveat in the bulletin. The source noted that Atlassian had already patched the vulnerability for all of their cloud customers behind the scenes. Wait, really? Yeah. The cloud customers never had to lift a finger. The only people left exposed to the internet were the self -hosted users, you know, the organizations that insisted on running the software on their own physical servers. Wow. I missed that entirely. It asks a very uncomfortable question of the IT industry. For decades, self
-hosting your own infrastructure was seen as the ultimate sign of security and control. Right, because you own the hardware. Exactly. But given the speed of these automated threats and the sheer human exhaustion of keeping up with daily emergency patches, are we rapidly approaching a tipping point? What kind of tipping point? Is self -hosting your own IT infrastructure no longer a sign of control, but actually becoming a fundamentally irresponsible risk for... all but the most heavily resourced organizations. Because if you demand to hold the keys to your own front door, you have to be the one to fix
the lock when it breaks. And right now the automation knocking on that door is much faster than the human assigned to fix it.
Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode! Japanese home goods company Iris Ohyama (online store operator Iris Plaza) found a suspicious file in an external...
Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode! Australian online retailer Who Gives A Crap told customers on 7 November 2026 that their information was accessed...
Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode! Last week in tech, 7 to 14 November 2026. Microsoft's November Patch Tuesday (10 November) fixes well over a...
Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode! Chainalysis disclosed on 30 October 2026 that an external actor using "offensive agentic techniques" accessed...
Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode! NSW Health reported unauthorised access to its Oral Health Fee For Service Scheme website on 29 October 2026;...