In this episode of the Distilled Security Podcast, we break down three converging forces reshaping how organizations manage AI risk - and what you need to do about it now. 🔹 BIPA + AI Notetakers - A class action lawsuit exposes unauthorized biometric data collection, why a single Illinois meeting participant creates liability, the Shopify wiretapping dismissal, and the steps you should take today to audit your AI tools 🔹 GRC Engineering Meets AI - Real AI compliance tools vs. vaporware, using LLMs for policy drafting and control mapping, the hallucination accountability problem, building AI guardrails as code, and the NIST RFI on AI Agent Security (comments due March 9, 2026) 🔹 ISO 42001 Deep Dive - The first AI Management System standard, how it differs from ISO 27001, AI Impact Assessments vs. traditional risk assessments, stakeholder engagement requirements, and why certification is becoming essential for EU AI Act compliance 🥃 Spirit Review: Redbreast 12 Cask Strength https://www.redbreastwhiskey.com/en-us/whiskey-collections/redbreast-cask-strength-whiskey/ ⏱️ Timestamps 0:00 Intro & Episode Overview 2:04 BIPA & AI Notetakers 25:08 GRC Engineering Meets AI 1:07:15 🥃 Spirit Review: Redbreast 12 Cask Strength (Irish Whiskey) 1:11:17 ISO 42001 1:49:30 Outro & wrap-up 🎙️ Hosts Justin Leapline - @justinleapline Joe Wynn - @wynnjoe Rick Yocum - @rickyocum 🌐 Connect with Us Website: distilledsecuritypodcast.com X: @DisSecPod Email: hello@distilledsecuritypodcast.com 👍 Like, comment, and subscribe for weekly security and compliance insights.
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.
Justin LeaplineToday on the episode, AI notetaker is sued for collecting voiceprints without consent. What happens when GRC engineering meets AI? Deep dive into ISO 42,001 and salt typhoon breaches congress. This is Distilled Security Podcast. Somewhere
Speaker 2right now, someone is digging through folders trying to find the right version of an evidence file for the third time this week. Controls scattered across a dozen system, owners who left the company months ago, due dates that already passed, nobody noticed. It's how it's always been done, and somehow, we just accept it. There is a modern, better way. Episki. Visit us at episki.com/dsp for a special offer.
Speaker 3If you're leading a company today, you're not short on ideas. You're buried under them. More tools, more frameworks, more initiatives that were supposed to help and somehow made everything harder. Minus Partners exists for leaders who don't need another thing added. Minus Partners helps leadership teams remove friction that's been normalized over time so execution gets easier, not harder. Less noise, clear
Justin LeaplineAlright. Welcome back to Distilled Security Podcast. If you noticed, we have a couple of new, things going on in the podcast. We're starting to run some ads into it. To start off with, we had some of our own companies sponsoring this podcast, which we're truly grateful for. Thank you, guys.
Speaker 4You know? I appreciate it.
Justin LeaplineYou know? But we are open for other sponsors. We're actually looking for some of them. So if you wanna reach a good GRC cybersecurity leadership podcast with hundreds of people listening each episode, please feel free to reach out to us. We'd be happy to talk to you and get your ads up on the podcast here. To dive into the topics here, we're looking at the first topic, BIPA with AI notetakers. And I think, Joe, correct me if I'm wrong, you use this notetaker. Right?
Joseph WynI use several notetakers. Okay. But, yeah, the one that they're talking about in this article so I was reading one of the JD Super articles. They're always great for what's going on.
Justin LeaplineWe just about them last episode.
Joseph WynYeah. Yeah. I thought so. And BIPA is the biometric information privacy act, and that's from Illinois. So what happened was is these AI tools are moving faster than the law and policy, and they're starting to do things. And what's happening here, it's not just like you're recording somebody, but the recording system actually can take your voice into a voiceprint Mhmm. And understand who it is and then be able to store that information. All of a sudden, they're saying that's biometric information.
Rick YocumSure.
Joseph WynSo that's like the the crux of it. So Illinois is saying if you're doing that, it's more than just consent that you're being recorded or more than just disclosure you're being recorded, but you probably need to be getting consent
Rick YocumRight.
Joseph WynFrom the people being recorded, and it needs to be explicit written like a whole lot of extra things that you may not think to do. Mhmm.
Justin LeaplineYeah. I found it really interesting because me looking at this, and it's always confusing with the law, but, like, who's responsible for it, you know, from an agency? Is it the actual tool responsible to collect that consent, which is who's getting sued into this? Or is it the actual company that employed the tool tool, you know, into that? And, you know, I don't know if they're getting sued and everything.
Rick YocumI actually looked this up a little bit Okay. Because I saw
Joseph Wynfind out?
Rick YocumSo I saw a note about indemnity and specifically some analysis that suggested based on case precedent and based on how liability shields work in practice, it's you as a company on the line.
Justin LeaplineUltimately. And that's what I figured. Yeah.
Joseph WynYou need a company that's pushing the button to say, let the AI notetaker into the call.
Rick YocumYeah. Because there's two things that can go wrong. Right? There can be an abuse of, say, the AI model itself. Right? Data that's your data used to train other models, like those sorts of things. The vendor is liable for abuses on that front. But how you use their tools is basically always on you. Every SaaS platform basically says, if you break the law, like, with our tools,
Justin Leaplinethat's on you. If it's out of the terms and conditions of what the the AI company says, like, okay. We're responsible for this. We're treating your data like this. Yeah. But how you use the tool, make sure you follow applicable laws and, you know, regulations. Yeah. And type of
Rick Yocumthing. And there's a lie there's there's like a library of precedent already because it's like it's all tech stuff. There's been tech lawsuits forever that just being applied to this. And one of the notes in particular I saw was like, oh, that makes sense was around how, you know, companies have errors and omissions insurance for, like, issues if they misuse their client's data, basically, in a bad way. But
Justin LeaplineOr use in a way they don't disclose. Right. Because the FTC has gone after people like that that, you know Yeah. Yeah. Yeah. For whatever data that they're collecting, you know, they're like, oh, you didn't say that.
Rick YocumYou know? But if you're the note take if you're the company that runs the notetaker app and one of your clients doesn't get the appropriate, you know, attestations that it's okay to record people or whatever, like, that's their use of your tool. Right. Typically, your insurance won't pay out for that because it was client's use of your tool. And then you, as a vendor, put in a place where you have to fight your client on who owns that liability. Yeah. So most of the t's and c's just have it written in out of the gate. And, like, all your Microsofts and Google's and Zooms Yeah. Basically, that's how it works.
Joseph WynYeah. And so so it's just, like, thinking of the craziness of it and some of the stuff you wouldn't expect. And so as you think about what you need to do, you know, speaker identity, cadence, emotion, these are things that are the AI can actually talk about. Do you ever use you know, take a transcript and throw it into AI and say, give me my notes, but at the end, also, give me the sentiment of
Speaker 4the customer. Yeah.
Joseph WynMhmm. Of how they felt about us after that. Yeah. And it's like analyzing Mhmm. All the things, and it's trying to take the those pieces. Now, if you're just taking a transcript with no of the none of the other, like, cadence and all that kind of stuff, it's just the words, then it's not really getting that. It's just gonna take it from the words. But I never tried this. You've tried analyzing audio and having it do that? That's something maybe to test out.
Justin LeaplineSo I've worked with it. Customers I mean, so, yes, I've seen that from a customer service perspective. They'll actually see if customers are irate or not and escalate it up to, like, people that are trained to deal with irate customers, you know Okay. Type of thing. So I've seen that in play, but I've also seen where banks have used some type of, like, voice authentication where they'll actually try to determine whether this is the user that's called in before talking about a claim or whatever it is, especially when you're dealing with, like, large money.
Like, I've worked with like retirement places and all that stuff when you're dealing with, you know, $6.07 figure places and they're like, I'd like to cash out. They they use a lot of tools to make sure that
Joseph WynSo every time you call,
Speaker 4it's Basically. Big money. Gotcha.
Justin LeaplineBut yeah. So I mean and I'm not sure they get consent on the end user every time with that because it's you know, they put it into, like, we're allowed to do this per our terms and service
Rick YocumI imagine they would prevent fraud, you know. I imagine they would upfront. Like, hey, you're opening the account. Oh, do you want extra security? Oh, okay. Agree to these things.
Joseph WynYeah. Those 15 pages of things you got assigned.
Justin LeaplineA ton like well, so maybe in maybe in some of the contracts, but if you buy contracts or merchant to something like that, I mean, you think about it, you call up a call center, you're not I mean, they might say, this recorded for quality, know, control and all that stuff. And But
Rick Yocumthat's like the whole point that's the whole point of the b b BIPA stuff is because they're like, well, now it's a biometric. It's not just a recording. Right.
Justin LeaplineYeah. Yeah. But you can turn any recording into a biometric.
Rick YocumWe can now.
Justin LeaplineThat was the joke when we were talking about this topic before. It's like, I got a Sony Walkman. I'm gonna record you. And you're like, well, no no no. It's a voiceprint. But if I have a recording of you, I can get a voiceprint.
Speaker 4You can
Joseph Wynthen take it and have AI. You can do the AI later.
Rick YocumYeah. It it's almost like and I think we might have talked about this on the podcast, but like how all the encrypted data breaches of the past when, like, quantum hits, they all become unencrypted and all those problems that, like, had kind of been, you know, noted, oh, it was encrypted when we lost the drives. Who cares?
Joseph WynYeah. Right.
Rick YocumThis is kind of the same thing. Oh, all these recordings that you have, oh, well, you couldn't do large scale biometric analysis on the phone.
Justin LeaplineOn them. You know?
Rick YocumBut now you can't
Joseph Wynhave them. Now if
Speaker 4you know the name of the
Joseph Wynperson, you can associate it with the recording.
Speaker 4Yeah. Absolutely.
Joseph WynYeah. Most companies are assuming, hey, we disclose this recording or we disclose that we are recording. And they think, like, oh, the vendor, is gonna handle all the compliance stuff. I don't have to worry about that. And it's it's just a transcript. But what it turns into is what are they doing with that in the back end that turns it into this extra stuff. And And now what's happening is that, you know, it's going from theoretical to real. Plaintiff's attorneys are actually targeting companies who are using AI transcriptions and note taking tools. And so, you know, I'm starting to see companies are putting together an extra set of policies for internal that and putting it into some training and putting into their employee handbook a little blurb that says AI note takers may be used. And, whenever we use it, here's what we want you to do.
And one of the things I'm personally getting the habit of now is every time I join a call where the recording is already yeah. In Teams, you can say, oh, I want that call recorded.
Rick YocumYou just
Joseph Wynhit it in advance. It'll just start recording.
Speaker 4Mhmm.
Joseph WynSo now when I jump on, I just becoming a habit.
Justin LeaplineRight.
Joseph WynI'm just asking. I'm like, hey. Everybody good? We're recording.
Justin LeaplineAnd at least for the normal person, because I do this exact same thing. I don't care about the, like, the recording or really the transcript. I like the notes, you know. Like, take notes for me. Give me the bullet points. Give me the summary. Give me what we talked about, and give me action plans.
Joseph WynYeah. So the problem is is that every time I do that with Microsoft Teams, there's, like, 35 tasks on a thing that should have had three.
Speaker 4And I'm like, oh, that's gotta get better. Clean it up anyways. But we're not
Joseph Wyncomplaining about that right now.
Speaker 4We're complaining about a different problem.
Joseph WynLike, I know you you've commented on this. There's the the no harm rule for this too.
Justin LeaplineYeah. What damages were actually taken on this? And I think it's way and you can correct me. I think it's a way BIPA is set up that it they put it into that it can it's all basically civil Yeah. Lawsuits and everything. They've turned it it's not criminal, you know, it's civil. So now, every lawyer and their brother, you know, want to basically sue for any wrongful step. And was there any remediation actually like happened? Because if this participant was totally myth that they were getting a, you know, a voiceprint Mhmm. Wouldn't the first step be, like, call the company, like, can you delete that?
Right. Or And then if they delete it, it'd be like, okay.
Joseph WynYeah. Or I'm being recorded. Right. And one of the things I see Fireflies is doing, and the reason I like the Fireflies tool, although I'll I'll probably just get everything in the Microsoft Teams and stop using that. But the Fireflies tool does several things. One is they have some good attestations and some good third party audits of their practices. So I looked into those and I liked them. They also are doing this thing. Kinda find a little annoying because if you're in a Teams call, it writes to the chat, hey, I'm Fireflies. I'm here.
Here's what you can do. But it makes it really easy. You can do, like, some some keystroke combination, like slash f f for Fireflies and then leave, and it'll just leave. Anybody can do that.
Rick YocumOh, nice.
Joseph WynYou don't have to be the meeting owner to do that.
Rick YocumThat is nice.
Joseph WynIn fact, when we're on some of our INS faculty calls, some folks, their Fireflies are set to automatically join all their calls. Mhmm. Well, when the they do the faculty internal calls, nobody from IONS is on it, so all the faculty can chat. So if you're an IONS faculty member, and I know some of them I think some listen. Two of us are in that right now, and we get on those calls, and then you'll see the recording on there. And it makes it really easy for somebody to control it. Anybody can say, oh, let's get it to leave. They don't have to be a meeting
Joseph WynYeah. The same thing with transcript on that from Microsoft Teams now seems to or facilitator seems to have the ability to kick the recording out if you disagree, I think.
Rick YocumOne of the one of the practical notes that I think was interesting that I encountered when I was looking into some of this stuff was a suggestion to almost like at the bottom of these emails, like, hey, if you receive this in air, you gotta delete
Justin Leaplineall that crap. Don't tell me the banner thing. Yeah. Come on.
Rick YocumWell, but from a legal
Justin LeaplineThe most perspective worthless security recommendation ever.
Rick YocumWell, because it's not security. It's liability.
Justin LeaplineYeah. It's still Right?
Rick YocumBut so it was interesting because it said, okay. Well, you might want to, if you're at a company, have a note that basically says no AI agents are allowed to be used unless there are AI agents, and those have to follow policy and all that stuff. And so because some of the things
Justin Leaplineto be a corporate. Yeah. Exactly. I mean, that that's tool sprawl that you're, like
Rick YocumIt is. But but I mean, if you're companies have meetings all the time with people that aren't employees of that company, and they might bring their own AI agents.
Justin LeaplineThat's just bringing in, you know, your own software like any anything else.
Joseph WynRight. But so Or if you're like my company and I'm joining a customer call, sometimes my customer will have their other AI agent or their notetaker join.
Rick YocumExactly right. And then it's your call. You're the host, but it's their AI agent who's responsible for the disclosures, the thing, like, what what policies
Justin LeaplineIt's whoever brings that AI agent to bear.
Rick YocumI don't know what the I don't what the legal standing is, but the recommendation
Justin Leaplineever controlling that software that's bringing it to record.
Rick YocumIt it well, they're the ones responsible for doing certain things. They own some of the liability.
Justin LeaplineLike, if they're having fire like, Firefly, as an example, dial in and actually do it, you know, then they're responsible for that. Like, I use a lot of, Notion. You know, Notion, I think a year ago, we start to do Sure. Recording and transcript and all that stuff. You know, the good, I I like it. I can record on any call and nobody knows, you know Sure. Sort of thing. So I don't know if I'm disclosing something. Yeah. Yeah.
Joseph WynWe might need to hey, can we edit this part? Yeah. No.
Speaker 4I was gonna
Justin LeaplineBut but yeah. Yeah. It's nice, like, and the nice thing about it, they integrate with calendar and everything. They're like, hey, you have a meeting in one minute. Do you wanna start up a note and start the recording and launch into Zoom or Teams at the all at the same time with one button.
Joseph WynOh, wow. That's cool.
Justin LeaplineAnd it's nice. And like I said, I've never listened to one recording. I don't even like, they delete the recordings. It's it's the output. It's the output of what I'm looking like. Great. I got all the meeting notes and all that stuff. Yeah.
Rick YocumBut back to the back to the disclosure about no agents that aren't ours in a meeting that you set up. If you're in a three part third three party call or something like that, I agree with you is a matter of practicality, like who should own that. Mhmm. But legal disputes get messy. Yeah. And so the point is it's not it's not a it's it's kind of a fail safe just to demonstrate your intent to follow rules more than anything else. Right.
Joseph WynAnd No. I totally agree with you there. And and the real situation is if I'm on a call with, you know, maybe two of my team members and we have a prospect on and the prospect has their notetaker join, it's my responsibility in my mind since I'm It's
Justin Leaplineyour host of the call.
Joseph WynYeah. Yeah. Senior person on the call. Maybe I'm the host of it. I'm gonna say, oh, okay. But I also have the ability in Teams to not allow their notetaker to join.
Justin LeaplineYeah. Yeah.
Joseph WynYeah. So I leave that sit in the waiting room Mhmm. Along with mine, and I say, hey, is everybody good with
Justin LeaplineRight.
Joseph WynThe notetaker joining?
Rick YocumThat's good way of
Speaker 4doing it.
Joseph WynAnd then I just let them all in at once. Yeah. And what I've done though with my team and I say, well, a lot of times, some of our customers or some of our prospects have their own notetakers join.
Rick YocumMhmm.
Joseph WynSo be aware of that and be careful what you say because we don't have control over the recording. And also be careful because we don't know that they're recording like Justin does. And, maybe they are capturing their own. Yeah. Right. So
Justin LeaplineAnd you should always I mean, from a consultant perspective, you have to be okay and cautious at the same time.
Speaker 4Yep. Yeah.
Justin LeaplineYou know, with both of those. Yeah. You know, from a company perspective, I agree with you. I mean, that's only polite to ask about recording. Like, I like how Zoom does it where, you know, as soon as somebody joins
Speaker 4Yeah.
Justin LeaplineIt's like recording in progress. Yeah. You know, type of thing. Like, there's no doubt. Like, it's like, okay, I know right now before I even start, there's a recording going on, you know. Yeah. And I can be okay with it or not and deal with it, you know, type of thing. Yeah.
Joseph WynSo I've actually just reminded me I I'm not gonna say the name of this person or this company, but it's somebody that I've talked to or respect a lot, and they are an attorney who is an expert in privacy, and they're also in charge of like, the most senior person in charge of maybe not all of InfoSec, but they cover all the GRC stuff. Okay. And they make sure that all of the audit external audits they
Justin Leaplinehave. Yeah.
Joseph WynSo they they really know what they're doing. And some of the customers that they work with will bring their own recording.
Rick YocumMhmm.
Joseph WynAnd that own recording might hear things the wrong way and transcribe Yeah. The wrong amount. So imagine that you're talking about that the the fee for this service is gonna be $50,000, but it records it and hears $30,000.
Rick Yocum5 or 500 are both bad.
Joseph WynYeah. Yeah. Right. And it it here's the wrong number. And when you go to a lawsuit, whoever has the best notes Yeah. Usually has a better case.
Justin LeaplineRight.
Joseph WynAnd so if you had the best notes but they're wrong and the other person doesn't have any notes, then you were gonna have less of an argument.
Justin LeaplineWe talked about this before. Do we? And yeah. I think we're talking about, yeah, the accuracy and whether you'd be hold liable to what some of these note takers were taking. Right. Oh, yeah. Yeah. Yeah. I I I'd like to see the court case for that because, I mean, it's really gonna be circumstantial, you know, with that. Because the next thing after that 50 or 500,000 comment is gonna be, well, show me the contract you signed.
Joseph WynRight.
Justin LeaplineAnd it's like, well, you signed it for 50,000, not 500,000. It's like
Joseph WynRight. The paper will
Speaker 4outweigh all that.
Justin LeaplineYeah. Exactly. It's like, I don't care what they said on the phone. You'd signed an agreement.
Joseph WynYeah. But how many contracts do you have that say, here's the price, but then travel expenses will be a 10% of the some of the consultant firms do it this way. They'll be no they won't exceed 10% of the entire engagement. Yeah. And then and so now you don't have a real number. And Right.
Justin LeaplineBut that, again, should be in a contract, you know, type of thing. Yeah. I guess so if that wasn't in the contract and it came down to what they told me, you know, and that's the that's where it would get a little interesting. It's like, well, you promised to do also this in the engagement. We didn't put it in the engagement, but you said you would also look at these IP addresses
Rick YocumRight. For for Even even just things that require verbal approval. Like, I work I'm working with a bunch of lawyers on a bunch of different things in any given time. And sometimes there's like, oh, okay. So do you want me to take this action? And that action might be notifying someone in a very formal capacity or whatever. And the bill comes later. Right? Because these people, like, you know, individuals are on retainers or whatever. And so I'm I absolutely am sure there's gonna be issues in the future where even if where the notetaker mishears something and the person you're talking to even gets sick, And then someone's covering for them reviews the notes and goes, oh, yeah.
You directed me to do x y z. He's like, no. I directed you not to.
Joseph WynOh, right.
Rick YocumLike that stuff like that's gonna happen.
Joseph WynYeah. Yeah. Well, that's one of the reasons every time I get on a call that has attorneys, the first thing they do is make me kill all the notetakers. Yeah. Yeah. They they won't let out work
Speaker 4for this.
Rick YocumAbsolutely. Yeah.
Joseph WynAnd so I'm like, alright. Hey, team.
Justin LeaplineLike physically kill all the notetakers?
Speaker 4Yeah. Yeah. Yeah. No. Take them out. Yeah. No.
Joseph WynI'm like, hey, team. You're you're back to manual notetaking. Right. Who's typing? Yeah. Get this stuff down.
Rick YocumBut even if not lawyers, like, I mean, even like, oh, well, you told this DBA to delete this table and start over. Like, whatever it is. Right? There's just gonna be things that require verbals typically that aren't like
Speaker 4Yeah.
Rick YocumContraction, and that's gonna be get messy.
Joseph WynYeah. Yeah. So, you know, I I think that people just need to start being aware of this. That's why I thought this was pretty this is I don't know. It's not cybersecurity related. It's more just knowing what's happening. And I found it fascinating just the way it was presented. But you gotta assume that, you know, this AI voice analysis is just gonna trigger biometric laws and
Justin LeaplineOh, yeah. You know? Lawsuits.
Joseph WynLaws yeah.
Justin LeaplineYeah. It's it's gonna be a mess. I I really don't like BIPA because of the civil lawsuit aspect of it. It's everybody hungry for a dime, you know, for
Speaker 4the lawsuits.
Justin LeaplineThere's a
Joseph Wynwhole another topic with the the cookie. What's that?
Justin LeaplineCookies. Again, another worthless thing. Yeah. But there are so many verification for, like, alcohol websites. Yeah. Are you 21? Nowhere, yes. Right.
Joseph WynWell, I'm so glad they stopped asking me to, scroll all the way down to my year of birth.
Justin LeaplineOh, those are the words. I mean, I should actually put in the meeting notes. There's a, like, there's a worse consents test that you have to make it through in, a minute and a half, and you have to answer all affirmatively Oh. Based on the questions. And they're all different and they're all like real world examples of getting like all this stuff, you know, to answer and it's phenomenal. I'll put it in the show notes. I forget what it was. It is a phenomenal website that basically just runs you through, like, trying to figure out how to say yes to whatever. That's funny. Yeah.
Funny. Oh, that is great.
Rick YocumOne other thing on this though. Yeah. And it's not really it's kinda related to this, but not specific. I was thinking there's this interesting outcome that I suspect is gonna start to happen from all these geographically based legislations. Because this is one of those things like because it's Illinois, right? This one?
Joseph WynYeah. Yes.
Rick YocumSo it just and it doesn't matter where you are as a company. Right? And it doesn't matter where the servers are or the data is or anything like that. It's is one of the people that's being recorded, right, in that location. Well, one of the things that's naturally gonna happen is you have things like GDPR that require you to know someone's citizenship or BIPO requires you to know where someone is. Like, you you you compound these on top on top on top. You're gonna end up with a bunch of technology technologists that have a need to collect all sorts of information from a person before they know what rules can and can't be applied. So you almost have to do more monitoring to know what rules legally apply to a person in a given situation. And if this keeps happening with, like, legislation, it's gonna get real messy real fast.
Joseph WynOh, yeah. So you're gonna end up having, at your company, a notification that says, if you're from any of these 49 states, I can't have a meeting with you.
Rick YocumRight. I there's gonna be all sorts of stuff like that. It's like, what's your what's your citizenship and where are you physically located? And what's the and what's this and what's this? Like Well,
Justin Leaplineand then there's I mean, we haven't even talked about it, but there's the, you know, either single party or dual party consent Mhmm. For recordings and everything that nobody really follows, you know, into this, you know? Right. But it is a technically real thing, you know, that if you're in a dual party, like one party is in a dual party, both part well, technically, that one party has a consent, you know, into that. But, yeah, it could be both people have to consent, you know, into it. So Right. Not a single party consent.
Joseph WynAnd I think the big takeaway for me is figure out if you need to have a policy, figure out if
Rick Yocumyou need
Joseph Wynto have training
Justin LeaplineWell, you do need to have a policy.
Joseph WynAnd and and then get your acknowledgments in place. Yeah.
Justin LeaplineYeah. And I would say, like, get the authorized tools in place, tool or tools or whatever it is. And like any other, you know, tool, like, you don't wanna do tool spread, you don't wanna support unsupported tooling, you know. It's a lot easier if you have your single stack and deal with it that way, you know. So, I mean, meeting notes are important and people should be able to do that. So have one that meets the company's needs, you know, and that's it. You know? Definitely.
Rick YocumI'll drink to avoiding tool sprawl.
Justin LeaplineYeah. Yeah. Yeah. Easier said than done.
Rick YocumRight. Agreed.
Justin LeaplineAlright. Moving on to the next topic here. GRC engineering meets AI. And I think we Rick, what did you bring this topic up
Rick Yocumhere? Yeah. So the reason I started thinking about this, and this could probably go in a million different directions
Justin LeaplineOkay.
Rick YocumBut the main thing I was thinking about recently was, if I think about the standard CIA triad, right, confidentiality, integrity, availability Mhmm. And I think about where I'll just lovingly call them, including myself, security and compliance nerds spend their time. Right?
Justin LeaplineIt's probably two thousands maybe for the
Speaker 4CIA triad. Well, yeah. But is it I'm not still kidding. The fundamentals. Yeah. Yeah. But,
Rick Yocumlike, where do you spend your time? It's like 95%, you know, confidentiality and available and and availability. Like, that's where people like, integrity very rarely gets a lot of attention. Right? And I was trying to think about, like, well, why is that? It's like, oh, well, was integrity never a big deal? Was it a bigger deal back in the days where, like, nightly batch processing was a huge thing and you could have a whole bunch of data that goes wrong overnight and then the business might not catch it? But now, things have happened in a bit more real time and they've moved on. And regardless now, with the concept of AI and agents more and more being either let loose directly or indirectly, right, indirectly being the business can basically batch process data at will Mhmm. Via certain agents, or you can have autonomous agents doing things, you do run the risk of basically, data drift, essentially, in ways that you didn't before.
Joseph WynWell, yeah, my take on the whole thing is that, you know, when you had systems and you're putting in code and you're going through a a good QA process Yeah. Companies have a dedicated QA team. What they're looking for is to make sure that the data integrity that's coming out of the system is right.
Rick YocumYeah. Rigorous SDLC protects for
Justin Leaplinethat stuff. Especially in financial systems. Oh, yeah.
Rick YocumOr health care or engineering. Like, there's a bunch of places where precision is important.
Joseph WynYeah. But with AI, AI, you know, it it won't crash the system when integrity fails, but it will confidently give you the wrong answer.
Speaker 4Right.
Joseph WynAnd so Yeah. What's happening there? And so now you're using a tool that has a little bit of flexibility Mhmm. In the output it might give And a little prompt engineering, a little confusion, a little something might create it to give you different answers. So no longer you're dealing with that static processing black box of the code that's taking data in and out, and you get the obvious expectations at the end. Now you're having this AI thing do its work, and you may not know that you're always gonna get the same answer. In fact, even prepping for this, I said, give me give me a quick summary of this. Yeah. And I said, oh, I want you to include this and then give me that summary back. Well, it changed the whole thing up when it gave me the second time.
Rick YocumRight. Well and and I think an important point is a lot of users in the business see IT as one big monolithic thing, and they see AI as an IT thing that they can leverage.
Joseph WynYes. And count on and rely on it. It should
Rick Yocumbe accurate. And that's the point because they're used to relying on IT stuff that's gone through rigorous SDLC processes and doesn't have, you know, fuzzy math underpinning a whole bunch of stuff. So now if it gets 80% of the stuff right, well, okay. I'm just gonna use you know, I always use, like, AP as an example here. Nothing against AP people, but it's just an easy example. If you have a bunch of AP analysts that use AI agents to run their stuff over and over and over, like, eventually, you could get this concept of data drift that actually has a material impact to financial statements. But it could
Speaker 4be this
Rick Yocuminsidious change slowly over time. So anyway, my whole point in thinking about this was, I think there's gonna need to be a resurgence or rebalancing of the integrity controls amongst the stack as AI stuff becomes more prevalent. Because you can fight some of it with, like, Guardian agents and, like, this concept of, like, defense in-depth abusing agents, essentially. But even that, like, how much do you rely on that? And frankly, how much do you have of that in place by default if you have just a bunch of a a AP analysts being able to write their own prompts and basically batch process live.
Justin LeaplineYeah. Yeah. Yeah. And I think I was gonna mention Yeah. One quick story. I was just recently doing a sell, and I had AI do a a number of things, including, like, some of the work breakdowns and pricing and everything. And it gave me a little nice little grid, you know, as AI does, it makes it look nice and pretty and broke out. It's like, here's the tables and the columns and format like this, and here's how I want the hours breakdown versus hourly rate and sum it up, you know. Yeah. And it did all that. And then I'm like, I'm looking at the number. I'm like, the total doesn't seem right.
Rick YocumWait a minute.
Justin LeaplineCopy and paste it. I put it in a spreadsheet. Yeah. Absolutely. Like, math was way off, you know, type of thing. And I'm like, yep. Alright.
Rick YocumThere there was a thing we had recently where a team was using some AI to to smash together a bunch of data really quickly so it could be presented to some executives. Mhmm. And it got pretty far along, like almost presented, not presented, but almost presented before someone had realized, oh, this pie chart is a 120%, not a 100%. Right?
Speaker 4Oh, wow.
Rick YocumAnd if you really wanna kill your
Justin Leaplineconfidence That was revenue. Right? Yeah.
Rick YocumIf you really wanna kill your confidence with an executive team Yeah. Right. Like, it's stuff like that. So anyway Yeah. Dangerous.
Justin LeaplineYeah. And that's why I think I mean, I'm not in the a development business. I'm more on the consumption side of it, you know, both for my tooling and personal. But I I see more like Claude and a few other AI tools. They're starting to incorporate using tools more Yeah. To augment what they're doing. You know? So in this case, if they're adding up something, they might get a calculator, you know, and and actually say, add up these numbers and give me the result, and then I'll take that, you know, type of thing, instead of throwing it into the AI model and then trying to figure it out. You know? So I think, yeah, I I was just doing something recently with, Claude.
They have something called, co work, which is basically more interaction with your environment and desktops and a whole bunch of MCP connectors and all that stuff. So you can connect your Salesforce with your Notion, with your intercom, with your HubSpot.
Rick YocumOh, like a integration middleware type basically.
Justin LeaplineAnd then you have one chat prompt and say, do this.
Rick YocumOh, neat.
Justin LeaplineAnd it will actually execute stuff for you Yeah.
Joseph WynIf you
Justin Leaplinewant, which is phenomenal. And it will actually do stuff locally on the desktop too. So you can say, clean up my my desktop because I have files and folders and all that stuff. Anyways, it's going a little off track. Like, it does a whole bunch of stuff. I was really impressed. I was using the coding piece of that. Yeah. And it was debugging an API for me, and it's like, well, like, I don't know what's wrong with your code, so let me try this curl command. And it actually reached out to the API directly Wow. To query it to say, like, okay. I expect this to come back. Okay. That's solved. Now let me try, it different with this.
Oh, yeah. I see the error here. And now let me adjust your code. And it did this, like, recursive thing where it kept test
Speaker 4Yeah.
Justin LeaplineIt tested my code, and then when there were faults, it adjusted and say, okay. I see what it's looking for now. Let me now go back and redo your code and everything. So, like, things of that nature, that's where the agent base is gonna be phenomenal. You know? But also dangerous.
Rick YocumLike, I always say, like Yeah. Yeah. The ability to, like, work at scale allows you the ability to screw up at scale.
Justin LeaplineAbsolutely. Yeah. And that's where it's like, yeah, there should be some gates and, Claude, you know, like, I I I was just playing with them a lot this week, so I'll I'll say their name many times on the podcast here. But they've done a good job of making kind of safe things to do. Like, they put a lot of emphasis on being safe, and there's, like, different modes where it's like ask every time you run something Mhmm. Or these commands are okay, but don't, you know, don't run these commands. Or there's like dangerous mode where it's like, just do it. You know? Right. Right.
You know, which it has checked in code for me. You know? Sure thing. So, yeah, you you basically there's a lot of ways of, you know, kinda doing that. I think yeah. I think it's good. Another story I have on the integrity thing Yeah. You know, with that, we had it was one of the programs I ran. We had a data grip database encryption software on the database. And we're coming to the end of our rock.
And one of the things that we discovered was we rolled out this platform, like, three months ago, four months ago, this database encryption, you know, for I think it was, like, MariaDB and MySQL Mhmm. Separation there. It came to be that it was causing data integrity issues Oh, yeah. Encryption software with it. And once we had discovered, you know, they brought it to me. It was like, okay. We either pass PCI with data integrity issues or we pull it and Right. Have to find another solution. And, you know, I made the hard like, I was like, you gotta pull it. Like, you know, in business, you can't deal with, like, data.
And and it was a back end database and all that
Rick Yocumstuff Right.
Justin LeaplineRight. You know, type of thing. So, yeah, I weighed the cost for it, but I couldn't sign off on our rock to say we encrypt data at rest, you know, for this, you know, this point in time Right. You know, type of thing. And it only took about a month. We got another solution in place and everything, and it worked with that. But, you know, that was one of those cases. Like, yeah, integrity is pretty important when you're dealing with, like, you know, financial data and all that stuff, and we had to restore the database multiple times while they tried to debug. They didn't know why they were getting data integrity issues, you know, and then finally narrowed it down to, like, yep. It's the encryption software.
Joseph WynYeah. So how long did it take? Do you have any idea how long they were running before they actually realized it? It was Yeah. It was months. And was that creating operational issues?
Justin LeaplineLike Yes.
Joseph WynLike, what what what sort of operational issues could a company think would happen when
Justin LeaplineYeah. We were seeing errors coming back on, like, select statements and everything, like, or it was pulling back, like, garbage because the data wouldn't be able to be unencrypted. You know?
Joseph WynI mean Okay.
Justin LeaplineYou know, essentially, it's trying to pull back encrypted data.
Joseph WynDid it go as far as,
Justin Leaplinelike, customers were receiving errored receipts or amounts? Don't know about that. Not that I was aware of, at least, into that, but it could've. You know? I mean, this was one of our primary database that it was both front end and back end that we use for our ecommerce Yeah. You know, into this. So it was front end bringing in customers, and then it also served as kind of a back end to manage those customers as well type of thing.
Rick YocumFinance is one of those places where it's important to have precision. But, like, when it comes to customer facing stuff, they're usually a pretty critical audience. So Yeah. If you're having issues, they'll probably feedback. I the places that I really worry about is, like, engineering applications, healthcare applications, where, like, maybe there are customers downstream, but it's removed far enough that you real damage could occur.
Joseph WynWell, think about, like, environmental. Yeah. So say you have a a water supply system.
Speaker 4Yep.
Joseph WynAnd part of the process is that you're treating the water with a certain amount of chemical mix.
Rick YocumAbsolutely. Yeah.
Joseph WynAnd if you if the AI is supposed to be doing something or somebody intentionally is able to make something happen
Justin LeaplineYeah.
Joseph WynSo that that is no longer safe drinking water that's getting into the reservoir Yeah. Then those are the kind of problems I think we we worry Absolutely.
Rick YocumAnd and in the integration of systems as well. So, like, one example, I I was working a lot with some oil and gas companies a bunch of years ago and, you know, their part of their job is to poke holes in the ground to extract stuff. Right? Well, you wouldn't believe how complicated it is to actually poke holes in the ground to extract stuff. Right? So there's all these topological surveys and this, that, and the other, and land surveys, and someone owns the land, and you might be able to pull it up on one side of the property line. But if you pull it up on the other side of the property line, it's a real problem because someone else owns the material that's being extracted through there. And by the way, it costs like a million dollars or some crazy amount to like poke a hole sometimes depending on various factors. So like It might not
Justin Leaplinebe worth it at that point.
Rick YocumRight. So so if the AI makes an error of half a foot in some cases, it's potentially catastrophic from a profitability perspective and things like that. Like, you could have major commercial projects or even like and and I'm not suggesting this is happening anywhere that I know of today, but like, I think, like, structural integrity of buildings or civil engineering or things like that, like, as people use AI more and more and more, it's a fantastic tool to help people do better. But, like, the the you you need to be able to trust it very thoroughly at the end
Joseph Wynof the day for some these your risk really changes. It's decision quality risk now you're worried about over other countries.
Speaker 4And I
Rick Yocumsay, like, would you let an intern do that themselves? And if not, maybe take a look yourself.
Justin LeaplineYeah. And I think I mean, we're basically talking about guardrails. Yeah. That's right. Point here. You know? So, yeah, I don't think AI should ever like, it will never be the Jarvis running, you know, Stark Tower, you know, like, you know, it's like, oh, yeah. It runs all the day to day operations
Rick YocumRight.
Justin LeaplineThat. There has to be guardrails. Yeah. So you were mentioning, like, the drilling or the the what's mixing in the water. There are ups and like, a a good range where certain elements have to be within that. Absolutely. And that should be vital systems that have nothing to do with AI, you know? Right. Mhmm. In fact, when I first started, like, at high school, I went into the transit industry and, you know, like railways and Yeah. Transit cars and all that stuff. And we literally had vital systems and non systems. Yeah. Vital systems were on the train. So if something exceeded or went wrong on the train, only locally the train brakes brakes went on, you know, kind of thing.
Right. And then we had a whole bunch of things outside like switching and routing Yeah. Absolutely. And all that stuff that were non vital, you know. Yep. But if things went wrong in the train, we had a we're slowing down to stop and that that's it until we correct it, know, type of thing. Yeah. So There's almost like
Rick Yocuma data category that's the analogy there. It's like, is this data that must be precise and correct, or is this data that, yeah, you can have, like, a bit more Right. You know, flex with it.
Justin LeaplineAnd I think that's where yeah. You you need to have those guardrails too. And it's like, okay. Now you two agents go add up the same row, and I want you to come to I mean, like and give me your conclusion blindly, you know. This is where I always
Rick Yocumcome back to, like, you know, I I always, like, treat him as an intern. An intern defense some depth is good. Right? Because if two interns argue, you know, one of them is probably wrong. 10 interns is even better. But at the Or same
Justin Leaplineboth of them is probably wrong.
Speaker 4But at the same time,
Rick Yocumyou know, you as the intern manager, like, determines what they're doing and how they're doing about it in the context they have and thinking about it. So it's weird that there's even this, like, this human error factor that can seriously impact the agentic, like, execution of things. Right?
Joseph WynDid you guys watch The Pit?
Justin LeaplineYeah. No. Yeah. It's fantastic.
Joseph WynThat's the show that's taking place in supposed to be Allegheny General Yeah. Just North Side Of Pittsburgh. And Noah, I can't remember his last name, was on ER. Yeah. Now is the doctor on this. And this last episode I won't give any spoilers, but this last episode
Rick Yocumnot seen the last episode.
Joseph WynThey had the a different doctor who's coming in is gonna, you know, hang out and be in charge of the ER for a little while. Yeah. And she brought along some new new things. One of the new things was the AI system to talk to instead of writing all your charting. And as they're going through it, this is like what's happening during the episode. And they're saying, watch this. And then they went through and they talked about the, you know, all the stuff for the patient, and it was as they were saying it, it was just writing it down and transcribing it and then it reorganized it and made it whatever. Mhmm. But it got one number wrong and it actually had a problem. Oh, And they made a really good point to like she she made a really good point to say, well, it's not a 100% accurate.
You have to review it. And I'm thinking, alright. Well, you get a busy ER, and people are gonna start relying on this stuff.
Justin LeaplineThat's
Rick Yocumthe problem. Yes.
Joseph WynYeah. You do it 20 times in a row or a thousand times in a row, and it's never makes a mistake. At what point do you just stop, like, reading it closely?
Justin LeaplineYeah. But I guess my counterargument to that would be how oftentimes do the nurses or whoever's taking the notes would actually make a mistake? And is it better than that? You know?
Joseph WynBut the difference is is that the person taking it and, you know, they have personal insurance for that kind of stuff, malpractice insurance. And, you know, you'll still get human error. What what point
Rick Yocumdoes it
Joseph Wynfeel more wrong when the AI took the notes and you didn't proofread them and you missed the thing versus you wrote it wrong because you misheard it? And is one worse than another? I think the point they were making, it's gonna be so much time savings and
Justin LeaplineRight.
Joseph WynSo accurate that it'll be less error prone than a person.
Justin LeaplineYeah. And I think that's probably right. You know? I mean, studies will show and time will tell Yeah. You know, type of thing. But we look at that with, you know, self driving cars now. I mean, you have a Tesla. You know? Like, they are statistically way better in self driving mode from a wreck, you know, standpoint. You know, that they would avoid wrecks, you know, more oftentimes than a human would in the same situation, you know, into that. In fact, I just saw, last week, week before, there's a a popular insurance company that is now saying, like, if you have, like, a self driving car, we're gonna give you a discount.
Joseph WynI need to find that.
Speaker 4I need
Joseph Wynto find that because my insurance is so high.
Rick YocumYou know what's right? I've predicting this for while. You know what's gonna happen next, though. At some point, I this is I'd I'd try not to be in the habit of predictions, but I can absolutely see this in the future. Fast forward ten, fifteen, twenty years, and I absolutely bet that it will cost a premium in your
Justin Leaplineinsurance To actually drive the car yourself?
Rick YocumWant to drive yourself.
Justin LeaplineYeah. Yeah.
Rick YocumLike, once it becomes
Joseph WynThe drivers you ever get the some of the insurance companies will give you this module you plug into the Oh, yeah. Thing, and it will monitor and then send to the insurance company things like, are you going over the speed limit?
Justin LeaplineThose are crazy. Yeah.
Rick YocumYeah. Well, I actually And did
Justin Leaplineyou see what Congress just passed? No. I didn't read into it too much, but, basically, their, car manufacturers are now allowed to put a kill code, like a a kill signal into your car.
Rick YocumI hate everything about that.
Justin LeaplineYeah. Just saw that it kinda snuck through and everything. It. So yeah. I don't know what the ramifications or because they already had a little bit of that. Like, if you had the star network onto it, they could, like, remotely shut down your car.
Rick YocumRight. Yeah. But but you choose whether or not you have that network. Yeah. Like, I mean, there's I don't I choose
Joseph Wynto have a Tesla and I know they can do that. Because if I take it for service, I'm like, oh, I didn't do whatever he's like, don't worry. I got it. And he just pushes some button on this computer behind the thing and now he owns my car.
Rick YocumYeah. The the option to have the manual thing, I think, is is useful. It's crazy. Well, the but about the about the chips And
Justin Leaplineall the privacy people buy, like, old Chevelles and Yeah. Just go old school. Yeah. No electronics. No chips.
Speaker 4I want no chips.
Rick YocumStraight mechanic. Back to cassette players. I had an employee several years ago who put one of those chips in her vehicle, and I remember so vividly, like, fast forward a month after that. She said, oh, yeah. Because discounts are over. And then we're talking and she's like, yeah. I never should have done this because apparently, I make too many left turns on the way to work. And from an actuarial perspective, my rates are now going up because of the number of left turns I have to make on table.
Joseph WynThat's dangerous cutting in front of those other cars. Woah.
Justin LeaplineI was like, so now she just needs to make like, three right turns and go straight. Right? And I was like Right. Right. Right. It was nuts. So That is nuts. Yeah.
Rick YocumAnyway Yeah. Well, so so also so I was thinking about the integrity thing. The other thing that popped into my high head on this topic just really quickly was as agents do more and more and more from a GRC perspective, right, there's gonna be automated testing and evidence collection and this, that, and the other. Right? So from and there could be hallucinations. And I I saw one quote somewhere. It was it was super funny. It was about vendor questionnaires. And like, oh, if something's automatically filling out your vendor questionnaires for you, you absolutely have to check it. And the quote was, hallucinated compliance is just fraud with extra steps.
Speaker 4With extra steps.
Justin LeaplineWhich I thought was pretty funny. Yeah.
Rick YocumBut so one thing that might you know, if if you have a decent program in place and you have some time and energy to start thinking about you should start thinking about your reliance and reperformance, like posture with respect to agentic testing. Right? Like, how much of that stuff do you have to look at yourself? How much of that stuff can you rely upon? How much do have to reperform?
Justin LeaplineAll that. I've said it before, and I'll say it again. Like, we're turning into a, an environment where it's AI talking to AI, you know, type of thing.
Joseph WynRight. Oh, yeah.
Justin LeaplineSo it's like, oh, you send me a questionnaire. I'm gonna have AI just write it, you know, type of thing. And then I'm gonna send it to you, and then I'm gonna ask my AI to summarize what your AI, you know, wrote into it. Yeah. You know? And it's just gonna be like AI just writing
Joseph WynOh, yeah. The the major GRC systems are already doing that. Yeah.
Rick YocumRight. Well, I think that happens with, like, insurance claims too between, like, what brokers and underwriters and stuff like that. Like, there are systems in place already that, like, negotiate amongst each other, like, a lot of the things that just sort of happen automatically. And then they're, like, looked at by the people at the organizations and all that But, yeah,
Speaker 4that happens.
Joseph WynAt what point is AI governance questions gonna start showing up in your your SOC two audit? Absolutely. Kind of stuff.
Justin LeaplineYeah. SOC two will take a while, I think. I don't know. What was our last update? 2017?
Joseph WynWell, it doesn't mean the auditor won't start deciding they wanna ask you about what kind of risks you're
Speaker 4Right.
Joseph WynNow needing to deal with.
Justin LeaplineAnd depending system of what they're doing, I guess it could fall into play.
Rick YocumYeah. The scope and the operations.
Joseph WynMean, you take the integrity part like that Right. Trust service criteria.
Justin LeaplineYeah. And confidentiality and all that stuff.
Joseph WynYeah. And then start finding out, have you considered AI risk? Yeah. We might get that a little bit later in
Speaker 4the show.
Justin LeaplineCurious. If you're listening to the podcast and you're a CPA company doing SOC twos, please let us know if you've done anything with AI. Oh, yeah. I have not talked to any CPA companies that do anything with AI.
Joseph WynI haven't yet either. Yeah. But one of the things that, you know, was as I was reading through this stuff, I started thinking about, well, if we're really concerned with it being in the GRC realm, then at what point does it become policy, and at what point does it need to be audited as part of your program?
Speaker 4Oh,
Joseph Wynyeah. And at some point, you're gonna say, well, this is my design. This is my control. And I I think you said one of the topics later is the ISO 42,000 Yep. One which is all about
Justin LeaplineAnd I'll get into that.
Joseph WynProtecting. But yes. But we'll get into it later. Yeah. But if it's gonna start showing up in those audits, then Yes. It's a matter of time.
Justin LeaplineYeah. And it's interesting. I'm not gonna talk about it now. Yeah. Yeah. Let's hold on. Yeah. And I think so I guess, you know, solutions out of this topic here. What what should we, like, be focusing on to tell, you know, our audience out of this?
Rick YocumI I would think if you haven't looked at your overall control set and you're responsible for that stuff, take another look at it and really think about the integrity controls that you have in place. And if they need to be bolstered, changed, applied slightly differently, as AI stuff Yeah. You know, it continues to accelerate.
Joseph WynI'll expand on that and say I'd say go through your policies, go through the controls, and see which ones can translate into the actual technical guardrails. Yeah. Because there's gonna be things and then was thinking, like, how is this different than administrative controls versus technical controls?
Justin LeaplineMhmm.
Joseph WynLike, thinking from maybe, like, the HIPAA realm. You have your administrative controls. These are the ones that AI create policies, procedures. Yeah. Yeah. Yeah. And but administrative control is something a person kinda does, where a technical control is kind of like a preventative control.
Rick YocumAnd and so AI stuff so I've I always this is a thing that I've thought about for a long time. AI is effectively, like, code at the identity layer. Right? You can sort of treat it that way. So it's this weird blend between again, if I treat it like an intern, it's like an intern robot. So Okay. Are your controls administrative or are your controls technical? It's like, well, it's a little of both. It's actually this new blended middle ground because I can't fully trust it, but at the same time, it processes way faster than a person could.
Justin LeaplineReally hard.
Rick YocumAnd you can point out lot of stuff. So but to your point, Joe, I think that's a really interesting thing to think through from a practical matter. Yeah. Like, think about your administrative and technical roles and which ones need to blend into AI from both sides of that. And I'd also say from, like, an audit and straight GRC perspective, if you're thinking about these tools, your audit teams are thinking about these tools or, you know, all that sort of stuff, if you're responsible for them, think about in advance, like, how much you need to and want to double check the work and codify what that looks like upfront.
Joseph WynYeah. Yeah. Maybe there's a standard that will help you Get that right. To
Rick Yocumthat. Right.
Justin LeaplineYeah. Yeah. And just, I'll add my, 2¢ before we, go into the drinking bit here, and everything. I think from an, GRC engineering, I think we've only scratched the surface. Like, a lot of it has been, like, automation and pulling controls and validating and simplifying, you know, to what are the biggest pain points, which, you know, is what AI is great for. You know? Things are repetitive in in task and simplifying them to only a few seconds or minutes, you know, into that. So, like, in a PISCI, we're doing we were putting a number of AI stuff over the last few months. In fact, just this week, I put in our own model context protocol.
Rick YocumOh, nice.
Justin LeaplineSo Yeah. Now within your AI tool, can you actually hook in a PISCI and say, give me all the tasks that I have in the tool or issues or you know? And then soon, I I have it read only right now just to test it out, but it will soon be able to do, like, recommend good test procedures, you know, for these controls and put them in. Yeah. You know? Or I want good solid tests based on this, this, this, throw it into the tool. And all of sudden, I'm not even coding anymore. You know? It's just I'm giving it availability to another tool to code like, to do the task.
Joseph WynThat wasn't any of the demo stuff you gave me, though.
Justin LeaplineOh, no.
Joseph WynNo. No.
Justin LeaplineYeah. This is yeah. This is brand new this week and yeah. So you'll be able to open up OpenAI or Claude or whatever it may be and basically have a conversation with the Episcay stack on what should I do next? Where should I go? And eventually, we'll actually integrate more of the chat stuff right into the the web tooling of it. But this is kind of a nice shortcut to give that availability. I see that and all the context protocols that are coming out for all these tools. It's gonna be the new API integration. Like, nobody's gonna, like, try to integrate their tool.
It's gonna be like, we're gonna integrate with AI with the CNL. And then all of a sudden, I'm gonna be sitting down and saying, GRC wise, why don't I just ask if our AWS instances are doing this? Is our AWS instances have multifactor access on all regions and endpoints? And it's going through. It'll query AWS. It'll pull back all the stuff. It'll like, it's like, you're missing three, you know, and here are the systems and all that stuff. Or I got a questionnaire from a client. Are they an active customer? Okay.
Let me reach into HubSpot. No. They left six months ago, you know, type of thing. Or, you know, they're not an actual customer.
Joseph WynAnd then instead of having to waste my time asking that question, I would hope that Episcay gets that built in so it just knows to go and say, oh, this customer oh, I have a control. I have to have MFA everywhere. Well, I'm gonna proactively just go and query all the things you already told me I have access to. Right. And I'm gonna come back and tell you what it is. And you wake up in the morning, make your coffee, and there's your list.
Justin LeaplineYeah. And it's it's great for that, but it's not cheap. You know? So every time you're doing that, you're eating up tokens, you know, to whatever service. Yeah. You know? If you're doing that for long term, an API direct connection and having actual, like, context and, you know, drone down to say, query AWS for exactly this is gonna be better in the long term. But, like, I was noticing when I was doing some cloud stuff, it was actually looking up some tickets and other repos. Yeah. I'm like, maybe this is a bug somewhere in the software stack that we're cool.
And it was using GitHub's API access. So it's actually a command line, g h space API, and then it was querying the repos and doing search parameters. And it returned a JSON string that had then digested and looked to see if, like, if there's other stuff related to the problems that we were having.
Rick YocumDude, there's so much first of all, I fully agree
Speaker 4with you.
Rick YocumYeah. I think, like, that that this hyper integrated context through MCP is gonna be like a thing. Yeah. But now that I'm seeing that vision that you've shared, I am terrified of like the AI spaghetti diagram.
Joseph WynOh my
Rick YocumAnd and the and the the thing I know before is the and the ability to, like, screw up at scale. Yeah. And that's where, like,
Justin Leaplinethe access to, like, identity access to agents and all that stuff. Like, it would be way and that's where it gets it's gonna be a whole platform building just for agents, you know, with this because you look at this is if I'm a GRC person performing a GRC role, and let's say that scenario where I'm saying, are they a real customer, you know, type of thing. I should have some access to basically say yes or no Sure. You know, type of thing. Do I need to know how much they paid over time or, you know, that, you know, that there's other information into there?
Or should I have right access to the CRM, you know, into that? Like
Joseph WynYeah. I would say no.
Justin LeaplineWell, I would say no as well. You know? But If I'm in the customer, you know, service, maybe I would just wanna update a record. Like, hey, I got a new address. Update it, you know, to this customer. I should have right Access, but maybe not full blown. You know? Like, there are so much stuff that it's like, okay. Now I have to, you know, do identity, and it's almost per task or per agent basis. I read a a good paper about it that it's not necessarily all the time it should follow the person. The person is a good indicator, but sometimes the person doesn't mean they should have the access. So if I'm a global admin, does that mean I should see everything in HR or finances share drive?
Rick YocumYeah. You might need to see it, but maybe, like, that's an emergency access Yeah. Only because you're a steward and you're a trusted party.
Justin LeaplineExactly.
Rick YocumYeah. And your query shouldn't look at that by default.
Justin LeaplineRight? Somebody in HR has a full salary spreadsheet, you know, and even though I have access, doesn't mean I should theoretically grab everything back into context.
Joseph WynWell, that's where you know, so this is getting back to, like, zero trust. Identity
Justin LeaplineYeah.
Joseph WynIs the new boundary. Right? And so thinking about this, instead of having one MCP that has the access, you have multiple that are just like users that you give them the access they need, and they can only do that job. And so I was starting to think about, alright. Well, yeah, the the whole thing, AI won't take your jobs. So he knows how to use AI, will take your job.
Justin LeaplineWell, now
Joseph WynI'm starting to reconsider that because now you create all these things and you think about the scenario. You have your GRC system
Speaker 4Mhmm.
Joseph WynWith all that has all your policies and the AI knows what all your rules are.
Speaker 4Mhmm.
Joseph WynAnd it does and it says, oh, I have a rule to make sure MFA is everywhere, and I'm gonna go and check all this stuff. And as soon as it finds it, instead of you waking up and saying I have to go deal with this report it found, instead, you now pass that off to a ticket system Yeah. That creates a ticket for an engineer to go and adjust that AWS system to turn on MFA. But really, the engineer it goes to is some other department's agent who has the access to do that. And so what's happening is in almost real time, does you just tell it, we need to have these controls in place.
Here's my, you know, my config profile. Go
Justin Leaplinedo it. Right.
Joseph WynAnd it just starts interacting and checking Yeah. Opening tickets, logging the fact that it's happening to create that audit trail and allowing the other system to receive that, make the change, and then report back that it's done, and then it can go recheck and then it can close the ticket.
Rick YocumSo this is one of the topics
Justin LeaplineWouldn't that be an awesome environment? Well, is one of the
Rick Yocumtopics I talk about a little bit.
Justin LeaplineIt's It's Yeah.
Rick YocumWell, it's agent granularity. Right? So how how broad or how narrow are the agents that you have? Is your engineering agent an agent that can do 13 different tasks that all come into that ticket queue, or is there 13 different agents and each one performs only one specific task?
Joseph WynYeah. And is it less expensive?
Rick YocumSerious pros and cons
Justin Leaplineto each.
Joseph WynAnd, Justin, when you're saying, like, you're gonna use all your tokens, it's gonna be this expensive. But at the end of the day, is it a $150,000 a year?
Justin LeaplineNo. Absolutely. It's cheaper than a person.
Joseph WynYeah.
Justin LeaplineYou know, doing the test almost every single day, you know, type of thing. But I'm saying for measuring from an API call to going through
Rick YocumAI Yeah.
Justin LeaplineAn API call is gonna be a thousand times cheaper, you know, per day. So, yeah. So there's there's places for it, and it's where you're hap interacting with context, you know, is where AI is gonna play well. If you're just do this thing and you know exactly what it is, you can make AI calls into that. But if you don't know exactly how to do that, AI could probably figure it out and then execute it for you.
Rick YocumBut this is where orchestration layers get
Speaker 4Yeah.
Rick YocumSuper important because that's where like, you need guardrails on agents themselves to not, like, go crazy in their specific focus task, and then you need orchestration layer guardrails so that's like, it's just like attack chains. Right? So certain things don't get chained together. It's like, oh, wait. No. You shouldn't like, this agent shouldn't talk to this agent, shouldn't talk to this agent because that indicates a bad thing's happening. But you look at, like,
Justin Leaplinethe MFA example, like, I think that would be a phenomenal thing if you gave agent or agents enough power into that. And even if it was just read only from a GRC perspective
Rick YocumOh, for free.
Justin LeaplineSay, like, okay. Why is an MFA turned on on this endpoint here? Is it a Terraform misconfiguration that is in GitHub? You know, like or did somebody turn it off? You know, like, does the Terraform say it should be on that we're rolling it out and it should be Yeah. On all the time, but somebody turned it off, you know, and that's where you get to learn what's root cause here, know, type of thing. Or maybe it's all, like, manual setup and it just it goes off and on as per people setting it, you know, type of thing. Or you can even say AI, was it on and it turned off?
Joseph WynYeah. Go do a pull request. Yeah. Adjust the Terraform and
Justin LeaplineOr that. Yeah. And then do a pull request if that's the actual, like, thing. Like, somebody screwed up, you know, the config file and they got through. It's like, okay. Well, you know, that should be it. And then write a new rule for Terraform before when it rolls out to make sure that setting is set. You know? So now you're linting, putting oh, yeah. Update in the lints, you know, to actually make sure that when configs go through, this is a chat.
Rick YocumBut this is like the AI That's
Justin Leaplinewhere they're talking about oh, sorry. No. You're good. Of, like, compound engineering. You know? Yes. Exactly right. If you're doing something and you're like, oh, that's a problem. Go right into the fly file to next time it it won't it'll catch it, you know, type of thing.
Rick YocumBut so one of the things, though, and I talked about, like, the AI spaghetti diagram before. Right? So a thing that I think is gonna be critically important as, like, these worlds begin to develop and you end up with hyper granular agents assigned to specific controls or tasks that are all orchestrated together is gonna be this concept of, well, how are you building this such that changing one agent or one tool, like, doesn't require changes to the thousand things that everyone in your business has built over the
Speaker 4course of
Rick Yocumthe past six months once you've turned this on. Like, you need, like, those abstraction layers and the concept of source control or, like, whatever the various bits and bobs are.
Justin LeaplineA big problem actually with AI right now Absolutely. Is there you know, everybody's teaching it context and memory, you know, into this. But the biggest problem that it has with memory is what happens when it changes. Yep. You know? Like and I just had that the other day. Actually, when I was coding my NCP, I was doing some AI coding with it, and I changed the endpoint. I said, I don't make it NCP dash server. Just make it a dash slash NCP, you know, halfway through me coding. And then ten minutes later, it was trying to go to m c p dash server.
Because I remember that's how I did it. How you it. Start, you know, type of thing. I'm like, no. M c slash m c p. Remember we changed it? Oh, yeah. That's right. But that's actually what I said.
Speaker 4Oh, yeah. Yeah. That tracks me up when it happens.
Rick YocumAnd I'm sure we'll see some clever things like, oh, just have really good enterprise change control and make sure that all of your individual agents can also watch for change control to see things that might impact them.
Joseph WynRight.
Rick YocumBut, like, you're starting to build these ecosystems that are so fast. And it gets back to a thing that you hit on before, Joe, which is kinda like when everyone starts to rely on the speed that all this automation can provide, It's it's just as dangerous to be in a cargo in a 100 miles an hour. If it stops immediately, boy, that's pretty painful. Right? So if you you then you make a change and it breaks a thousand different agents in a bunch of different ways, that that's all that needs to be part of your Doctor plan essentially.
Right? Depending on what you have these things doing.
Joseph WynYeah. Well, I'm gonna change my takeaway from figuring out which policies need to be
Justin LeaplineRecorded. Yeah.
Joseph WynYeah. Well, you can have both. But my more my my new improved takeaway is like, look, if you're gonna learn, if you're gonna train on stuff, and maybe we can get a guest in here who could take us through how does a person who like, how long did it take you to just figure out how you wanted to do this stuff? You've been coding Episcay stuff for years, so you've kinda come up with all this stuff. And so your learning curve was your learning curve though to be able to get to the point of doing this.
Justin LeaplineOh, yeah.
Joseph WynWas, you know, quicker than others. But if you have a person out there who's not really in the middle of it all and needs to start figuring out what is the curriculum I follow? What is the way I figure this out? Right. I would love to have somebody come and just sit down and tell people That'd just like, they're in the industry. They're here for a couple years. They're like, you know what? I don't want AI to take my job. It won't. Somebody who knows how to a do AI will take my job. Well, maybe not. It's somebody who knows how to create MCPs and create all the AI They're the ones that are going to displace. And once they're in place and they just put those bots are in place and they displace these people doing these normal things, it's the person who's gonna have the oversight of that. The person who and and this is where, like, the the auditor role isn't always the most glamorous role. But all in a sudden, if you're the person who can make sure that the system and all of this stuff is running right, you have to understand so much.
Justin LeaplineAbsolutely. Yep.
Joseph WynIn order to do
Rick Yocumthat. Absolutely.
Joseph WynAnd how do you get into that path? That's what I'd love to have a whole episode on career advice for people that
Justin Leaplinewant That'd be an interesting
Rick YocumThat'd good. We should do
Joseph WynSo, hey, reach out to us if you are the person who could come in and speak to that.
Justin LeaplineYeah. I saw a interesting study over the last week, think it was, where the perception of how much AI is saving your company And down kinda at the grunt level, it was like single digits, like 5%, you know, 7%, whatever it was. Up at the executive level, it was like 35, 40%, like AI is saving efficiencies across the company.
Rick YocumThe the gap between expectations and reality.
Justin LeaplineOh, yeah. Absolutely. Was it was a entertaining kinda survey.
Joseph WynYeah. Real example. I I I will instinctively know what I'm gonna do. And then I wanna just validate it. And then I'll ask a question, and then I'm like, if I would've just did what I was gonna do, I would have been done. But instead, now I'm reading like five pages of AI generated advice. Oh, yeah. And I'm like, half of this doesn't even make sense. This is not I would never do that.
Justin LeaplineI did something today where I generate something like that, and I think my next three lines were shorter. Shorter. Shorter. Yeah. But that's, like, shorter.
Rick YocumIt's like the new analysis paralysis is, like, chewing on all the data that AI might give you. That's very interesting.
Joseph WynOne of my favorite takeaways, John Ziola says when he writes prompts, he's like, favor brevity.
Justin LeaplineYeah.
Joseph WynSo just write those two words at the end of your prompt. Mhmm. It's so much better.
Justin LeaplineYeah. Alright. Well, why don't we go into our alcohol section here? Anybody need a refresh?
Rick YocumI would love more.
Justin LeaplineSo we have for this episode, red breast, which if anybody's a Scotch drinker, know red breast very well. They're very popular into it, and we got the twelve year cask strength. It's a dried fruit and lively spice. I think they said with apricots and butterscotch and barley finish. And it is a triple distilled matured in the finest oak cask.
Product of Ireland here. It is delicious.
Rick YocumIt is So good.
Justin LeaplineIt is if you know Scotch, it's not as peaty. It has a very subtle
Joseph WynSo you can have a it doesn't have to be from Scotland?
Rick YocumI think this is a this is a whiskey and not a scotch. Right?
Justin LeaplineOh, is it? I think
Rick Yocumit's an Irish whiskey. I'm
Justin LeaplineOh, you're right. Okay. Good. I would might have got angry comments after the Irish whiskey, not scotch. Yeah. Could've that's all bad.
Joseph WynHey. We we could just bleep over all those
Rick Yocumthose things.
Joseph WynJustin was just swearing. That's what it was. It wasn't saying anything wrong about this.
Rick YocumThis is a really beep.
Speaker 4Yeah.
Justin LeaplineWhen I say, it'd be Irish whiskey. Yes, delicious Irish whiskey.
Rick YocumIt's so good.
Justin LeaplineYou know? And yeah. The red breast, I mean, honest, anything they do. I even had the red breast, the Red Label one Yeah. Which is like twenty five years or something like It's
Speaker 4I don't think I've had that.
Joseph WynWas it better than this?
Justin LeaplineYou know, it's probably, but this is excellent.
Rick YocumYou know? This is so good. I like pretty I don't think I've had anything from them. Not that I've had, like, hundreds of bottles. I've had a decent amount.
Justin LeaplineNever have I.
Rick YocumI've I've never had anything from them I don't like. Yeah. Like, it's all very good for my opinion.
Speaker 4So
Justin LeaplineThey're good one. From an Irish whiskey perspective, if you ever wanna venture into that, definitely Red Brass. I think their standard is eight, if I'm not mistaken. Somebody will probably correct me on the tubes here. But, yeah, this is a little bit more of their kinda mid tier, I would say. And, yeah, it's excellent. You get a little bit of smoky flavor, I get Mhmm. Onto the kinda middle tasting palette of it. Finished really smooth. What's the ABV on it?
Rick YocumIt's, like, 58 and change?
Justin LeaplineYeah. 58. So yeah.
Rick YocumWhich it gives you gives you a little bit of the the heat, but it's not overbearing.
Joseph WynNo. It's really Actually, for one
Justin Leaplinesixteen proof, it's it doesn't seem like a one sixteen proof. No.
Joseph WynIt doesn't feel that hot.
Rick YocumYeah. I'm a sucker for vanilla, and I get a ton of vanilla from
Justin Leaplinethat. So Gotcha. Alright, guys. Well, Cheers. Cheers.
Rick YocumClink. Hope we're doing it.
Justin LeaplineNeed do it.
Speaker 7Quick break to hear from one of our sponsors. If you own security, compliance, or risk, and it feels like you're always pushing a boulder uphill, I want you to know about CISO. CISO helps growing companies get audit ready, reduce risk, and stay resilient without drowning in tools, endless checklists, or one time reports that quietly rot the moment the audit ends. This isn't shelfware. It's not drive by consulting. With CISO, you don't just get advice. You get hands on support from real security engineers, GRC specialists, and former CISOs who help you build, operate, and continuously improve your security program over time. Whether you're chasing SOC two, ISO 27,001, CMMC, HIPAA, or you're simply trying to get security under control so the business can move faster, CISO meets you where you are. Their managed VGRC model gives you enterprise level expertise without hiring a full internal team or reinventing the wheel. The focus is simple.
Clear priorities, practical controls, and measurable progress leadership can actually understand. Visit cisollc.com and start the conversation. Security you can trust, compliance you can prove, and people you can depend on.
Justin LeaplineAlright. Welcome back to Distilled Security Podcast. Why
Speaker 4don't
Justin Leaplinewe finish up here, gentlemen, with talking about we kinda segued into it with a little break.
Joseph WynThis is the AI episode.
Justin LeaplineYeah. Exactly. So ISO 42,001. So if those that are not familiar with it, this is ISO's answer to AI governance. And they basically if you're familiar with the twenty seven thousand one, they have their ISMS. This is their AI m s I I I s AIMS. AI management system. Yeah. Yeah. Yeah.
Rick YocumWhen it's really tied to the EU AI act. Right?
Justin LeaplineNot it's tighter
Rick YocumIt's really close.
Justin LeaplineYeah. So it has a lot of those attributes and everything out of it, but I think it would be arguably tighter to 27,000. You know?
Rick YocumOkay. Yeah. I I think it's, yeah, I think it bridges the gap pretty well.
Justin LeaplineYeah. So it does align to the EUAI act Yeah. With that. And I think there's, like, it's even accepted into that. Have you guys either you guys worked much with this yet? Yeah. Not worked with it.
Joseph WynOkay. Yeah. I picked it up right through it. Okay. I mean, the the main thing is, like, you think we've been talking about AI risks this whole time. Mhmm. And AI introduces risk that goes beyond your traditional cybersecurity. So whereas, like, CIA, confidentiality, integrity, availability are what you're worried about. Yep. With forty two thousand one, you're really worried about harm. You're really worried about AI doing things it shouldn't. And you wanna think about, like, what are the consequences of of it? We're we're, like, 27,001 is asking, is information protected? Are risks, the CIA managed? Forty two thousand one is thinking more about what what decisions does AI influence or automate, who is affected by those decisions, and if they're wrong, and then what harm could occur if those systems, you know, if that happens.
Rick YocumYeah. Really really tied to, like, reasonability and fairness in a lot of ways. Like, in that way, I see it a lot closer to a lot of the privacy legislation that exists than some of the security legislation. Yeah.
Joseph WynYeah. Yeah. Because twenty seven thousand one is protecting information
Speaker 4Yep.
Joseph WynWhere forty two thousand one is looking to govern behavior.
Justin LeaplineYeah. Yep. Yeah. And, yeah, I think a lot of people you can extend out your twenty seven thousand to also get 42,000.
Joseph WynYeah. You need 27,001, and then you can come along and add on the controls.
Justin LeaplineThat's not anymore. I thought you can actually now get a standalone 42,000.
Joseph WynI know you can do that with the twenty seven seven zero one for privacy, but I haven't really looked
Justin Leaplineinto one for privacy. I thought it was 42,000. Maybe. I'm gonna ask AI about AI governance here.
Rick YocumBut, yeah, I thought one of the things that, you know, was interesting is it's it's hyper focused on the risk assessments. And like you said well, I shouldn't say hyper focused, but a major component of it is risk assessments. And like you said, Joe, really about, you know, fairness and ethics and those sorts of things. And so I guess one of the takeaways right out of the gate, given that framing, is if you are cybersecurity practitioner, certainly there are cybersecurity elements that are that you know, because the security part underpins same way as privacy, like a lot of being able to do things in a safe and fair fashion. If you're a security team that doesn't typically interact all that much with your privacy teams or legal teams and stuff like that, and you feel like you need to get this or start to walk down this road, you're gonna need to start interacting with your privacy and legal teams from my perspective.
Joseph WynOh, yeah. Yeah. Absolutely.
Rick YocumLike, this is this is gonna be a joint effort. This is not like a sole focus, just cybersecurity person thing.
Joseph WynYeah. And I'll blame it on them. Cold medicine. I think you're absolutely right. The 42,001 is a standalone
Rick YocumYes.
Joseph WynCert you can get.
Justin LeaplineYes. Yeah. Yeah. AI said so.
Speaker 4Yeah. And I told it in
Justin Leaplinethe prompt to agree with me. So it's absolutely
Joseph Wynright. A 100% that's right.
Rick YocumWhat a good intern.
Justin LeaplineYeah. Yeah. Yeah.
Joseph WynSo but but with yeah. And with the 42,001, you're you wanna really just look at the governance across design, training, deployment, monitoring, change management, decommissioning. Like, these are things that are different than maybe ISO 27,001, which is a little bit more control centric.
Justin LeaplineYep. Yeah. Yeah. Twenty seven thousand one, we have a breakdown in the notes. It's 93 control areas, four themes, organizational, people, physical, techno technological. 42,000 has 38 controls across nine domains, and it's AI life cycle focused into that.
Rick YocumSo So one of the things when I was reading through it that I think is gonna be the hardest for organizations, again, is the AI Is the concept like AI, the just the definition of AI. Right? It's not we know. It's not monolithic. It's not just one thing. Yeah. Right? It's a bunch of different things. So how do you define and scope AI? And obviously, there's some baked in ones that you can use. But Joe, I'd I'd rely on your ISO expertise here. I'd imagine, like many ISO things, you can kind of define your scope specifically. If you wanna veer from the definitions, you probably could. Is that true to some extent? Like, you wanna define AI a certain way, could you scope and certify to, well, we're just gonna do AI that, know, works in this way or this specific product as a Well,
Joseph Wynyou you do need to pick your scope Yeah. And you need to define, like, what those and this is, like, very similar across a lot of the ISO standards. And you need to look at what risks you're gonna manage.
Rick YocumRight.
Joseph WynAnd here, you're looking at an AI impact assessment. So all those things
Justin LeaplineTo Rick's thing, you don't have to include all the AI you're doing. No. No. You gotta subset.
Joseph WynYou gotta include the scope of your what what you want your certification to cover. Yeah. And it may be, you know, the the processes for AI that are happening out of this business unit and not the other one.
Justin LeaplineYeah. Or it could be like if I have a tool that has an LLM or AI interactions, maybe I just do the tool because that's what the customer cares about. Right. And that's why I'm getting the certifications. The customer's gonna ask
Joseph WynRight. Your customer's gonna go to your website and say, wanna buy this product. And then you're gonna get a certification not for that product being AI secured, but you're gonna get a certification around your AI governance, your AI management system
Justin LeaplineWithin your product.
Joseph WynThat that Yeah. Governs that product.
Rick YocumYeah. Yeah. But I think that's such a great distinction. Governs that product is potentially different or likely different than governs the organization as a whole. Right. And and with this stuff, I think it's gonna
Justin Leaplinebe because they could have AI meeting notes that they're not considering. Exactly. Oh, right. Yeah. Might not be part it. Hey.
Rick YocumWe just got a pissy, and it has a bunch of AI stuff in it. But I think that's it's such a critical distinction because even, the definition of AI and then another element in here that I think is hard. So once AI is defined, you have to do an AI inventory. Yep. And so, again, if your scope's tight, you know what you're, like, attacking, that's not the end of the world. But if you're gonna be like, oh, we want our organization to be certified to this, oh, you're gonna be opening some closets and turning over some stones if, in many cases, you have marketing teams doing various AI bits and bobs and finance teams and all these different things.
Joseph WynYou probably wouldn't scope those in because your customer may not be they're not buying
Rick Yocumthe product. And so, yeah, I think from a a a practical matter, I guess, like anything, sort of start small. Start with a very specific scope and then potentially expand that.
Justin LeaplineInteresting. I haven't thought about this. But then, like, if you're integrating a vendor into, like, your product or something like that and they claim, like, AI, but it's not really AI, you know, type of thing. You're like, all of a sudden, you're like, now have to flush out the the marketing only AI versus This
Rick Yocumis why the definition matters so much.
Justin LeaplineYep. Yeah. Because I think the the first thing is defining AI systems in scope, you know. So what counts as AI, you know, into that. I mean, do you count, you know, like, just something crunching some numbers and doing kind of a natural process lang like language processing, you know, into that? That's not making kind of deterministic or indeterministic like like choices. Mhmm. It's just like kind of compiling like what the sentence is mostly like that's not AI technically. Sure. You know,
Joseph Wynbut let's use your example from earlier. Yeah. You talked about taking a you're building a quote, you're building a statement of work, and the AI did a lot of good stuff. I use it for this too, it generates a list of tasks, and then you can estimate see, I would estimate hours. Yep. And then you hope it adds it all up. And you think about it, and you're like, how are you using that? So if you're gonna get your process for so maybe you're a large company, and you wanna still use the AI to, at scale, build SOWs to make all your salespeople a whole lot faster. Yep. So it can happen at one one person or it can happen at, you know, a 100 person company, and you're gonna use AI to do that.
So you think about this. What is the harm? So the harm that could've happened is you could've just relied on that. Your intelligence didn't jump out and say, oh, I'm gonna double check these numbers because that doesn't feel look right. And the harm then becomes that. So if you do your AI impact assessment, you might say, well, I'm gonna build an SOW. You I'm gonna let the AI build the SOW. Well, the the impact could be, well, the task could be wrong compared to what I need to do to implement this for the customer.
Justin LeaplineYeah.
Joseph WynOr this hours could be wrong or the math just taking your billable rate times the hours is doing that wrong or just tolling it wrong. So these are all your your risks. These are what what could go wrong. Right. And then what is the harm? Well, at this point, the harm is well, you could've you could've had a reputation damage because you sold them something and you sent them this and they're like, Justin, this doesn't add up. Yeah. What what are you crazy? Right.
Justin LeaplineOr And if it's less, they're gonna sign immediately.
Joseph WynRight. They'll be right. And then, well, now now you're getting the other harm. Yeah. Maybe maybe you've just committed in this contract to spend the next three months doing this thing and you're gonna get four days worth of
Speaker 2pay. Right.
Joseph WynWell, who in society is being harmed? Your family, because now you're not bringing home the same amount of money. Yeah. And so now you're thinking about all these things.
Rick YocumWell, and and from an ethics and fairness perspective too, like, what context are you feeding it? Right? Or if you're feeding it your history of sows in the past and maybe those haven't been maybe your strategy has changed significantly in the past couple years or whatever.
Joseph WynMaybe you raised your prices.
Rick YocumYou raised your prices.
Justin LeaplineWe added a Pima charge into a customer and then it goes Yeah.
Rick YocumYeah. The Pita Pita tax.
Speaker 4Yeah. But
Justin LeaplineAnd then all of sudden, it actually puts it as a light on him, and the customer is like, what's that?
Speaker 4Well, it it could it could be that.
Rick YocumBut it could also be like
Justin LeaplineFor pita chips.
Rick YocumIt could also be kind of insidious where it's like, well, for various legitimate reasons, contracts in this part of the world seem to cost more like this much more. Well, now if it starts applying that is a surcharge. Oh, well, that's, like, from an ethics and fairness perspective, like, oh, well, now there is, like, weird potential
Justin LeaplineMaybe that's even a brand thing. You just start losing business because you're way, you know, out quote, like, you know, over quote you're over quoting in other regions that you're not appropriately
Rick YocumOr there could be regulatory issues or all sorts of various things Or discrimination. That's exactly right. So I think Discrimination.
Joseph WynWell, imagine that you are selling business to consumer, b to c
Justin LeaplineYeah.
Joseph WynAnd you now are adjusting your price based on a certain model and maybe that the person puts in their zip code and it's a particular area where it
Justin Leaplineis With mortgages and all that stuff.
Rick YocumYeah. I get back like the buyer's context there.
Joseph WynYeah. Yeah.
Justin LeaplineThat's a great impact and all that stuff and everything. Even though yeah. I mean, the it gets in a little bit of political, you know, stuff of that nature because, you know, you see from the bank side, they're like, well, it's basically in the slums. So anybody in the slums in that ZIP code, like Is higher risk or what? Risk. Yeah. Exactly. It's like, it's historically cost more. There's more claims coming from that zip code. I can prove it. But then you get, you know, whoever is on the DEI, you know, side of it. They're like, well, but that's 80% black. So are you looking at 80%, you know, and that's why you're charging more into that area? Well, know And
Rick YocumI think that's just the thing that it you you have it the defensibility of it matters. Right. Right? Are you making this are are you doing a calculation and the calculation is fair and ethical and all that stuff? Or are you just saying, well, this is an easy and and I would suggest, like, AI in this case. Right? Does AI then take this constant text and say, oh, well, an easy shortcut is if someone lives in this area or if someone looks like this thing or whatever, like, I can just make this assumption. Then then there's a difference, obviously.
Justin LeaplineI I mean, it's interesting. Like, I I I actually this is how boring of a life I do. In some of those like, I actually saw on C SPAN, like, some of those congressional hearings where they brought some of the bank executives. They're like, we don't even collect race as a, like, a determining factor. Like, we don't know what race is. So, like, well, you could tell by their name. You know? It's like, okay. Like like like, come on. You know? And again, you know, they're trying to do a point. The banks are bad, blah blah blah
Rick Yocumblah. Right.
Justin LeaplineEverybody's greedy. You know? I get it. The political game, all
Rick YocumBut but from an ethics and fairness risk perspective back to '42.
Justin LeaplineIt's just it's justified. Like, if I could show, like, claims in a certain geographical area, and this is why my rates and I don't care if you're what your race is. Right. It's like anybody in that zone gets this,
Speaker 4you know.
Rick YocumBut yeah. But I'm I guess I'm saying from like an AI perspective
Justin LeaplineOh, yeah.
Rick YocumYeah. Like if the risks are ethics and fairness, like, okay, well, how are you addressing those risks? And then also, even just the concept of legal defensibility.
Justin LeaplineYeah. I don't know.
Rick YocumI throw it into the AI black box and I put in all this context that spits something back out is definitely different than, well, here's specifically how we've instructed the AI and here's how we're mitigating risks that could point out and here's how we do
Justin Leaplinehuman rule processes. You know, for positive or negative, you know, type of thing. Absolutely. In fact, there was a big pit thing where, like, they're trying to make it more diverse on, like, some of the image generations. You know?
Speaker 4Oh,
Justin Leaplineyeah. And I was seeing, like, some of the people were like, well, give me a picture of, like, the French Revolution and some of the soldiers. And they'd have, like, a Chinese person in there and a black person. I was like, that's not historical.
Rick YocumBut That's such an interesting
Justin Leaplinetrying to put in more diverse, but obviously, it wasn't coming out as historically, you know, accurate.
Rick YocumThat's such an interesting point too, like injected model bias
Justin LeaplineRight. Exactly.
Rick YocumWhich might have unintended consequences in any in any kind of thing. So anyway, any decision making capacity. Right? Like, again, ethics and fairness and defensibility and
Justin Leaplineall think there's arguments on both sides because I I really like I really like Elon taking over Twitter and doing Grok. If nothing else, do an counterpoint to the rest of the AI industry. You know? Because he was big on AI is OpenAI is kinda ruining it because they were putting a lot of that bias in, you know, to try to make it kinda cleaned up. But then you look at, you know, Grock was, you know, like, pretty open, you know, with a lot of the stuff. But then it turned into an angry teenage Nazi girl, you know Right. Within, like, a few weeks of the inner, oops. You're like, alright. Yeah. We need more guardrails.
You know? Type of thing. You know? So, like, there's arguments on both sides that I I like, you know, into that, and you can go too crazy on either, you know, into where you put the right guardrails and where you put, you know, biases on either side. It's all biases at the end of the day.
Rick YocumYeah. It's just yeah. Managing bias. And and and making sure that you have a consistent and defensible approach. Yeah. Right? But anyway, yeah, think I think that defining what is AI and then the inventory of AI stuff in broad scopes is gonna be hyper painful. So, like, narrowing that scope will be a useful thing if you're actually going after this. I also thought one of the things that was interesting was this concept of, you know, you're supposed to redo these AI risk assessments periodically, and that periodically can be time based, but also, like, when major things change. Right.
It just put me in the mind of, like, pen tests and, like, let's be real from a security perspective. Like, how many people actually redo their pen testing when something material changes? It's I mean, that, like, practically never happens.
Justin LeaplineI can probably count on one hand the amount of times I've seen that through PCI. Right. Because they have the same thing in there. Right. Do this at least annually or upon Or
Rick Yocumwhen a major thing changes.
Justin LeaplineA significant change and everything.
Rick YocumI said, well, we define significant as, and then, like, you never have to do it.
Justin LeaplineOr the QSA never gets it or never determines that it was a significant change.
Joseph WynAnd how much control do you have over if the model that you're relying on, if they're making updates?
Rick YocumYeah. Right. That I think that's a fantastic point. And then are you and then you're obligated to re risk assess because I mean, I I mean, I I understand the chain of logic. Yeah. I mean, when when when something could materially impact the decision making processes related to the
Justin Leaplinerisk You're saying any new model that comes out, you need to redo the impact assessment?
Rick YocumI think it depends on your scoping and the risks that you've defined. But if there's the potential for the new model to impact those risks in a material way, you my understanding is you would be obligated to rerun.
Joseph WynSo what what what I think you
Justin Leaplineneed I think to that's kind of crazy on the timing.
Joseph WynSo I think you need to have is in your own procedure. Mhmm. Your own governance that says how you make that decision. Yeah. That's what you have to document. Yeah.
Rick YocumAnd it can be small yeah, to your point, like, medium, large. Like, you know, is this is this such a major change that we need, like, a full on net new risk review across the board? Or is this small enough potentially that a senior enough person can look at it, think about it, maybe get a second opinion as well and go, oh, yeah. We're good.
Justin LeaplineYeah.
Joseph WynYeah. So, Justin, when you're writing some Mepisky like MCPs, are you able to peg that to a particular Yes. Version of Claude that is gonna use?
Justin LeaplineCan actually we use Vercel's AI gateway. Mhmm. So we have, like, over a 100 models that we can actually don't, like, within twenty seconds switch to.
Joseph WynSo if they so if Claude does a new rev, you can make the decision. So your 42,001 governance can say, well, if we switch from a major rev of quad, like like OpenAI Right. From four to five
Justin LeaplineYep.
Joseph WynAnd that and and soon as it went, like, I my results were just going crazy for a little bit and using your words, bonkers. Right?
Speaker 4Yeah.
Joseph WynYeah. And then, but if you were doing that, you could say, oh, you know what? We're gonna we're gonna evaluate and you might just do this naturally. You would probably, knowing you, would be like, oh, I'm not gonna just like switch it to that new one. I'm gonna do some testing and do that. And then to get 42,001 around it, you're going to, like, codify in some Yeah. Rules or some governance. This is my procedure to decide how I switch from that to that.
Rick YocumYeah. This is standard change control and best practice stuff. Right?
Justin LeaplineYeah. And I think so I guess there should be a process. I guess a full blown impact assessment is where I'm like, okay. That that's not warranted. Like, if the process is exactly the same, you're just testing input output at that point. You know? Oh, but that that could be
Rick Yocumyour risk assessment process, like ensuring that input output remains.
Justin LeaplineRight. Exactly. But that's not the full blown impact assessment. Like, you're not you already know the potential ramifications
Joseph WynYeah. You're doing the delta.
Justin LeaplineYeah. Exactly. Like, is the same inputs getting the same or better outputs? So, you know, type of thing is really what you're
Joseph Wynlooking for. You're building that process, you're gonna
Justin LeaplineAnd it should be for any change. Like, it's less of a concern for minor model updates, like 5.1 to 5.2 on OpenAI. Mhmm. But if it's, like, to a whole new model like Opus 4.5, which is Claude or Gemini or whatever, you know, Copilot or something like that, You're switching that. You're testing full blown regression testing on how
Joseph WynNo. Absolutely. And so does it gonna affect people, the organization, or society? Right. And so if you're taking those things into account and your governance based on the risk that your company can tolerate, do you make a decision that you're evaluating that? And as top leadership, you were like, you know what? That I did evaluate the risk of that. And so, you know, with with your GRC tool, you're not going to negatively affect teenagers who are Right. Online Mhmm. Doing stuff that AI might be generating whatever for.
Mhmm. So, like, these these tools that are out there, these AI friends that you can get and or these these toys now that are coming with, like, AI responses and that kind of stuff. And they're Yeah. They're storing and they're reaching out and they're that kind of stuff. Those companies who are building these toys and these products that are gonna use AI
Rick YocumLike Meta has, like, AI people, right, that interact or, like, I think Instagram and all that. Like, I think there's legit Yeah. Accounts that are fully AI run, like, intentionally from
Joseph Wyncompany that's funny. About a toy that was actually being out there that was recording stuff and then sending it out and then the results would come back and it would speak these results and it was an AI toy for kids.
Rick YocumOh, I remember that. Oh, what's what what are the BIPA implications? No.
Joseph WynNo. Never mind. We're
Speaker 4past that. Yeah.
Rick YocumWhat were
Joseph Wynyou gonna say?
Justin LeaplineI was gonna say, like, in the mid two thousands, I was setting up, like, fake bots on AIM and all that stuff that responded to people. Oh, yeah. But those were, like, just taking, like, bits and pieces of the the yeah. The the sentence and turning it around, like, understand what the subject and the verb and all that stuff was, and then it turned it around into a question or, you know, it would actually interact weirdly with people.
Rick YocumIf you're a full time marketing exec, you had AI back then. Exactly. But but yeah. No. But now, yeah, there's there's full on
Justin LeaplineConverse yeah. And all that stuff. And it's good and bad. Again, it it depends on the guardrails and everything. So we have a little bit of AI. We use a a tool called Intercom for our main kinda customer interaction support and all that stuff.
Rick YocumShowed me that a little bit. That is a very cool
Justin Leaplinetool. Very powerful tool, and we don't use it to its full potential. But some of the stuff that's really nice is, like, we can put interactions into, like, Stripe for your payments. So, like, hey. I'd like to make a a payment for this even though you can go through the, you know, the the website If you're interacting and like, oh, yeah. I'd like to upgrade. You're like, oh, okay. Here. You know, right in the tool, we'll give you, like, click on this and go pay for it.
Rick YocumThat's like an agent that can just help you.
Justin LeaplineYeah. That's what it is. You know? And it basically has connectors that understand your context and then basically makes the decisions and offer up the tooling that may help or may not. Yeah. So yeah.
Rick YocumThat's very cool.
Justin LeaplineYeah. It's it's pretty good. And then a lot of the AI in fact, I just had it this past week here. There were some potential customers that had a little bit of trouble getting in.
Rick YocumYeah.
Justin LeaplineAnd communicating over AI. I was like, oh, yeah. Let me take a look at it. And I had Intercom create a linear ticket, which is our development tickets. Yeah. It wrote up the ticket based on what they said, you know, into it, filed it into it. We worked on it. We closed it same day, you know, and it communicated back to the customer that this ticket was closed on their issue Yeah. Yeah. Into that. And all of a sudden, like and you can actually look in your intercom and actually see, oh, yeah. I opened up something and it was closed and, you know, and it should be all fixed now. And now you got this communication path, you know, with the customer. They're happy that their problem got solved and fixed. You're happy that, you know, you got basically a workflow to do that and not another, like, email and email.
How much time
Joseph Wyndid that save you?
Justin LeaplineOh, a ton. Yeah. Yeah. I didn't have to hop to one tool to another tool to write something, to write it, to file it, to all that stuff. It was just in it. You did the coding
Speaker 4Yeah.
Justin LeaplineWhich I think was AI too, you know, that type of thing. Yeah. And then, finish it up. That's cool. Same day, you know, I've got it fixed.
Rick YocumWell well, Joe, so if if there's an organization that maybe they have a 27,001, maybe they don't. But, like, what if an organization thinks they should start to wade into the AI, you know, ISO certification, what should they start thinking about right now?
Joseph WynWell, they probably already have so if you have a twenty seven thousand one, you already have a lot of things in place. So you have your policies. You already you're doing your normal processes for risk assessment, management review, internal audit. Like, all these things are very consistent across a lot of or almost all of the standard standardized ISO standards now.
Speaker 4Mhmm.
Joseph WynAnd so, you know, you wanna think about what is your governance. So the the nine areas that you mentioned, I believe it was or nine or 10 areas. And there are things like your AI governance and accountability. So the clear ownership. So you wanna define these things. Define the roles and responsibilities is very important. AI risk management is really talking about, the identification of the AI specific risks.
Justin LeaplineYep.
Joseph WynAnd then that's what leads you into an impact assessment. So if this risk came to fruition, what's that gonna be? What's the harm? What's what's happening there? You need to understand your AI life cycle management. So that's a whole, area. You have things like data governance and then human oversight, things like that, transparency Yeah. And explainability. So you wanna be able to have all that stuff documented. And then some of the normal stuff that's, like, the same a lot, on a lot of ISO standards, competence, training, awareness, incident management, and response, like, do you do?
Continuous monitoring and improvement. So these are all the same. So if you're already building a program and you're already a ISO certification in some areas Yeah. And you wanted to extend it, you likely can go find an auditor that can do both. So your your certification auditor probably can do 42,001 on top of the ISO twenty seven thousand one and twenty seven seven zero one.
Rick YocumI was gonna say What's that? That that's interesting.
Justin LeaplineThey have to get certified.
Rick YocumDo they?
Justin LeaplineI was
Rick Yocumgonna ask that. Yeah.
Justin LeaplineWhat's that? They have to be certified.
Joseph WynOh, yeah. Yeah. So there's a lot of them out there that are I was just I was just talking to there's a couple different certification bodies, like, ANAP is one of them that certifies auditors. And I was talking to the auditor who actually was the first, I think, anywhere or at least in the country to have the 42,001 Cool. You know, ability.
Justin LeaplineSo I'm curious because I don't think they can. With 27,001, I know this because I got a provisional auditor at one point with it. One of the requirements to become a full blown bleed auditor was you needed three audits underneath your belt, you know, into that.
Joseph WynOh, that's a very good, I don't know the answer to that.
Rick YocumFor new certs, I think they, like, waive that.
Justin LeaplineI mean, they have to For for a minute. There's no leading auditors out there that can do that, you know, like, you know, the existence of that will be nil or maybe they'd say you have to be twenty seventh. I I don't know. I I I never looked into
Rick YocumI think you can get your stripes on I think, and this is based on some certifications that I had a long time ago. So it could be lapsed at this point. Yeah. But I think you can get your stripes on a different cert. It could be. To your point, like, you've done enough 27,001
Justin Leaplinecan auditor over here
Rick YocumYou can as long
Justin Leaplineas you pass the test. Like grandfather the experiential clause. Over there. That could be. I don't know ISO well enough because all I did was, do the provisional audit on twenty seven thousand.
Rick YocumI was a lead auditor at one point, but it's definitely lapsed.
Justin LeaplineLike, '27? Or provisional lead auditor?
Rick YocumI'm full. I did enough. You did three audits?
Justin LeaplineYeah. Okay. Yeah. Yeah. Okay. So anyway Yeah. I it's just it's hard to do that without we did that as a security company in the mid two thousands. A bunch of us got trained, but I don't think anybody read the fine print on it. It was like, oh. We can't be like like so and we never did pull the trigger to actually hire a lead auditor. Yeah. Yeah. Yeah. Yeah. So we're just basically a bunch of provisional auditors, you know, provisional lead auditors. But
Rick YocumYeah. The training, but not the
Justin LeaplineWe we couldn't do anything.
Speaker 4Yeah. Yeah.
Justin LeaplineLike You can't sign that. The test. We're all that. We're provisional. Oh, okay. Great. Yeah. You can't do anything. Yeah. Like
Joseph WynYeah. So I guess the the big takeaway here is that if you're going to be running systems, and again, any any audit you're gonna get, let me back up a little bit, is somebody comes and says, I get a SOC two audit? Should I get an ISO certification? I'm like, well, let's talk about why. How's that gonna help you? How's that gonna actually make your company generate more money?
Rick YocumYep.
Joseph WynHow's that gonna be a benefit to your organization? So we figure out, like like, somebody comes to me and says, oh, I wanna get ISO certified. Like, alright. Tell me, what are all your customers asking for? Every day they ask for a SOC two. I'm like, why are we talking about ISO? Let's talk about SOC two. Right? So let's just get the right thing figured out first. So whatever it is. So if you're a company who is looking to get a product that's gonna heavily rely on AI as part of it, and you know your customers are gonna be asking you for some kind of third party vendor risk management because if they're talking to any of us, we're probably working both sides of that conversation.
Justin LeaplineI gotta ask from a Piscay if I had a SOC two. Yeah. Or ISO.
Joseph WynYep. I was on that call.
Speaker 4Oh, yeah. And and
Joseph Wynso if I'm guiding a customer Yeah. Who's gonna go buy stuff, I'm telling them, look, you need to have a solid third party risk management program. You need to go ask all these questions.
Rick YocumWell, I think there's a of questions. Product, but also like a service that's A supported by service.
Justin LeaplineWell, it could
Rick Yocumbe a product that's AI enabled or a service that's supported by AI. So either way, but the clients could come calling.
Joseph WynAnd then I'll go at the same time have a conversation with a company who's building these tools, a software tool, whatever it is. And they're gonna say, I keep getting asked all these questions. I keep getting asked if I have a SOC two or how I'm you know, what am I what kind of external audits do I have? And so, you know, I'm I can help both sides. Yeah. And as we do that, we're we're saying, well, what do you need? What risks do you need to manage that if you explain this is gonna make the customer sign the line Yeah. Sign the line.
Rick YocumRight.
Joseph WynAnd if it's and and if you're an AI company and that customer is really worried about that, you know, so at what point do we start telling our our own customers, hey, when you do your TPRM, your third party vendor risk management reviews, when are you gonna start advising your customers to go and you know what? You might wanna start asking if they are 42,001 certified. Like, has that come up in any of your advice that you've given anybody yet?
Rick YocumIt hasn't yet. Although, one of the programs that I'm particularly close to at a client, we're gonna start doing more specific vendor risk management stuff in the AI space specifically. And so I imagine that'll be one of the shortcuts that can be used. Right? Hey. Do you have this, or do we have to ask you this bucket of questions?
Joseph WynRight. And if you're if if one if they're actually going in writing their own AI tools for their customers and they're relying on somebody else's Exactly. Now
Rick Yocumthe Fourth party risk, all that stuff.
Joseph WynYeah. So you wanna, you wanna understand how it works both ways on that. So, my advice is as soon as you think you're gonna need to get ahead of this, start looking at something. I mean, NIST has the NIST AI framework. Yep. You could use that. You can't get certified on it because NIST doesn't have specific certification.
Justin LeaplineHas an AI framework. You could
Joseph Wyndo that. You can get some certifications from places for certain c s cloud CSA is cloud security alliance if Yeah.
Justin LeaplineEverybody's wondering. And Their certification, I think, is coming later. Yeah. They're just rolling out their, their AI stuff now. Yeah.
Joseph WynLike, they have the star and other sorts now.
Justin LeaplineAnd I think it's planned that this will eventually be certifiable.
Joseph WynYeah. Okay. And so you have these, but then you could always ask a CPA firm just similar to a SOC two. You can a CPA firm come in and follow the AICPA standards and write an attestation Right. Their opinion on it's not a certification. It's somebody's opinion of whether your design is good and the controls are effective.
Rick YocumThat's a fantastic note. You're absolutely right.
Joseph WynAnd you can you can do that for anything. You can make your own framework. You can have the Rick and Justin framework for Sure. How to drink whiskey at a podcast, and you could have an auditor come Yeah. You can you can have an auditor come in and say, here's all of my design and here is, our evidence.
Rick YocumIf you're a third party auditor,
Speaker 4we'll do
Joseph WynDo some evidence
Rick Yocumtesting. Supporting this.
Justin LeaplineYeah. Yeah. And so certification once a month?
Speaker 4Is that what?
Joseph WynYeah. Not not a certification.
Justin LeaplineNot a station. At a station. Yeah. So
Joseph Wynyeah. People get all hung up on like a SOC two, all this stuff. At the end of the day,
Rick Yocumyou They just want a third party opinion.
Joseph WynYou really need a third party opinion usually gets you to the point of where you need to be. So whatever it is you're doing, if you're doing it the right way and your design's solid, you have a framework that you can pull from that that somebody can look at and say, that makes sense. A SOC two is nothing more than just some some specs for what that design ought to look like. It's just a structure. Same thing. Yeah. NIST CSF is somebody's opinion of what is a good security framework, and there's no reason that the Still Security Podcast couldn't go out and just audit some third party some company, and we could write an opinion. Joe Justin No deviations noted. Yeah. Yeah.
And so, like, the three of us could just go and write that, give them the paper, and they could say, hey, we got this. Well, great. It's the Distilled Security logo right there on it. And if that closes the deal 20 customers, that's legit.
Justin LeaplineYeah. It's good enough. I had a customer So if
Rick Yocumyou're interested in sponsoring Yeah.
Justin LeaplineI had a customer one time that they got a SOC two from a data center or something like that, and we were reviewing it, I think, PCI or something. And I opened it up and start going through executive summary, blah blah blah blah blah, and I'm going down to the controls, you know, and nothing there. They deleted Oh. The entire bottom section of the SOC two.
Rick YocumI love I love the SOC two that should have been a SOC three.
Speaker 4Yeah. And I
Justin Leaplineand I'm looking at this. I'm like, where are the controls? They're like, what are you talking about? I'm like, there should be controls here. And we reach out to the data center company. They're like, we consider that confidential. Yeah. I'm like, just be honest. You had a deviation. It got listed in there. You deleted it. Come on. Like Yeah. If if there if those were
Rick Yocumall green checks, would you deleted it?
Justin LeaplineYeah. Exactly. Yeah. Yeah.
Joseph WynYeah. And that's that's another thing. You know what? I we've been talking about this for a while. Stay tuned for a future episode. We're gonna do one where we talk about SOC two.
Rick YocumIs it true? Yeah.
Joseph WynHow to evaluate it, what's important, what doesn't doesn't matter Yeah. When to use it.
Justin LeaplineSo Yeah. I might have to go through my hit list of awful references for SOC two and other reports and everything. I think that would be a a fun, like I won't name names, you know, if I might get
Speaker 4in trouble.
Joseph WynWell, the best thing is, like, some of the notes that we started to put together already, the first line of it was the SOC two certification.
Justin LeaplineOh, no. Yeah.
Speaker 4I'm like
Joseph Wynand and AI, that was obviously AI generated.
Speaker 4So we gotta fix that AI.
Rick YocumWell, just one quick takeaway for me on the '40 2000 '1 is because the risk assessments are so closely aligned with more ethics and fairness as opposed to security, super strongly recommend, as I noted before, if you can lever the processes and or tools that already exist in your privacy or legal teams, these risk assessments are gonna mirror those a lot more clearly than they're gonna mirror your typical cybersecurity assessments. So strongly recommend
Justin LeaplineIt's you basically risk. You're identifying risk in the AI, you know, process.
Rick YocumAnd it's very like human risk to people kinda more centric. Like, it's AI, but it's really targeted towards
Justin Leaplinecontrols and putting guardrails in Absolutely. For test, for validation, or whatever it is. Yeah.
Joseph WynYeah. Twenty seven thousand one secures data and Yeah. Forty two thousand one governs your the AI decisions.
Rick YocumYeah. But so anyway, I highly recommend if you can reuse the tools and processes more on the legal and privacy side, probably gonna be an easier route than trying to reuse if you have a separate risk assessment process on the security side. You'll kinda be shoehorning it in. Yeah. Gotcha.
Justin LeaplineYeah. Any final thoughts?
Joseph WynNo. I think we hit on a lot.
Rick YocumYeah. This is great.
Justin LeaplineIt was a good time and and everything. So alright, everyone. Thank you for joining us. Don't forget to like, comment, and subscribe. We'd love to hear from you. So don't forget to comment on all that. Don't beat me up on the Irish versus Irish whiskey versus
Rick YocumWe caught
Justin Leaplineit. Yeah. Yeah. We caught it.
Joseph WynSelf correct. We'll bleep it out. Bleep it all out. All those are swear words. He was totally swearing.
Justin LeaplineEverybody over with AI voice, you know, stuff. It will all be correct and probably funnier with that. So
Rick YocumAI. Remake this podcast.
Justin LeaplineYeah. Exactly. Alright, everyone. Thank you so much, and we'll see you next time. Bye.
In Episode 30, Rick shares the biggest takeaways from two invite only executive conferences, Palo Alto Networks in Napa and CDW in Arizona, where the agenda was 85% AI. We cover scaling laws, frontier vulnerability...
In Episode 29, Doug Salah returns as DSP's first-ever repeat guest. We preview TRISS Pittsburgh 2026 (October 22nd, $118K+ in scholarships over 9 years), break down the real implications of the CMMC pause and class...
This episode covers the growing controversy around Flock cameras: what they are, how their centralized license plate reader network enables powerful searches across jurisdictions, the documented cases of abuse by law...
In this episode of the Distilled Security Podcast, we dive into AI-driven vulnerability management, a massive breach hitting 9,000 educational institutions, and the future of GRC engineering, including a live preview of...
In this episode, we celebrate our 2nd anniversary and Episode 24 of Distilled Security! We cover the Vercel breach, how a Roblox script led to compromised Google Workspace credentials via an unauthorized OAuth...