Your AI Strategy Is Already a Cyber Risk with Scott Alldridge

If last week’s conversation challenged the way you think about cybersecurity discipline, this episode pushes that conversation into even more urgent territory: AI.

In this continuation, Tonya sits down again with Scott Alldridge to unpack what happens when innovation outpaces governance. As organizations rapidly adopt AI tools—often without oversight—new risks emerge beneath the surface. From hidden data exposure to autonomous systems interacting in unpredictable ways, Scott reframes AI not as a technology problem, but as a leadership, governance, and operational discipline challenge. 

This conversation goes beyond hype and into reality—where AI expands attack surfaces, complicates compliance, and demands stronger foundational controls than ever before. If last episode was about discipline, this one is about velocity—and the cost of moving too fast without guardrails.

In this episode, they explore:

• Why rapid AI adoption is expanding organizational risk faster than governance can keep up 
• The hidden dangers of “AI sprawl” and why most companies don’t know how many tools they’re actually using 
• Why AI increases your attack surface and introduces new, harder-to-detect vulnerabilities 
• The difference between generative AI and agentic AI—and why autonomous systems raise the stakes 
• How AI tools can unintentionally leak confidential data or create compliance violations 
• Why governance, not tools, is the foundation of safe and effective AI adoption 
• The biggest blind spot in AI strategy: unclear ownership of risk across IT, business, and compliance 
• Why human oversight is still non-negotiable—even with advanced AI-driven security platforms 
• How foundational IT disciplines (like change, configuration, and integrity management) remain your strongest defense 
• The growing gap between AI innovation and regulatory clarity—and why organizations are still fully accountable 
• The role of third-party AI vendors in introducing unseen risk into your environment 
• The first critical steps leaders must take to regain control: inventory, pause, and reintroduce AI with governance 
• Why the next wave of major breaches and lawsuits will likely stem from unmanaged AI usage

Important Links:

https://app.technologymatch.com/solutions/ai-governance-risk-management

https://app.technologymatch.com/solutions/ai-security-posture

A complimentary e-copy of his Amazon Best Seller VisibleOps Cybersecurity. Text your email address with the words “secure 2026” to 541-359-1269 OR go to https://scottalldridge.com/ and fill out the contact form, noting “secure 2026."

Up to three no-cost Level One penetration tests/scans (for qualified organizations - $2,500 to $10,000 in value) Text your email address with the words “pen test” to 541-359-1269

Transcript

SPEAKER_02 0:01

Welcome to Between Fires and Futures, a podcast about the real work of tech leadership, managing today's chaos while building tomorrow's business. I'm Tanya Terrell, a three-time founder with two successful exits, and the founder and CEO of TechnologyMatch.com. Each week in this podcast, I talk with the leaders doing the real work solving for now, building for what's next, and leading through pressure, not perfection. This is the podcast for tech leaders fighting fires today and daring to build the future anyway. Welcome back to Between Fires and Futures. I'm your host, Tanya Tyrrell, and welcome back to cybersecurity expert Scott Aldridge. When we talked last week, Scott walked us through the zero trust illusion and why cybersecurity discipline breaks down long before a breach ever happens. Today we're building on that conversation and stepping into something moving even faster: AI. Organizations are adopting AI at record speed, innovation is accelerating, business units are experimenting, but governance, oversight, and risk management are not always keeping pace. Scott returns not just as a cybersecurity executive, but as board advisor and transformational leader who works directly with organizations navigating digital transformation, regulatory pressure, and AI adoption in real time. As CEO of IP Services and president of the IT Process Institute, he brings a people, process, and technology lens to some of today's most urgent technology conversations. If episode one was about discipline, today is about velocity. Scott, welcome back to the show.

SPEAKER_00 1:37

Thanks for having me. We're excited to be here.

SPEAKER_02 1:40

Yeah, I'm great. It's so great to have you back. I really wanted to continue the conversation into this AI era. So you've cited research showing that mid-sized organizations are already using more than 20 AI tools. What does that tell you? Like what should we be thinking about this?

SPEAKER_00 1:57

Yeah, I think the the business pressure and the you know executive pressure to take advantage of the efficiencies of AI is obviously mounting. Competitors are using it. So people feel like they've got to kind of be the firstest with the mostest, and they're missing out if they don't employ some of the AI capabilities to drive those efficiencies, which ultimately drives the bottom line and uh can be a competitive differentiator. And and as a business person, I get that, certainly understand that you know desire. However, often, as C-suite people do sometimes, or business initiatives take over, it's done without proper risk analysis and really evaluating what are the potential threats or vulnerabilities that could be created by employing some of these latest and greatest AI technologies. It's not all good. It's not automatically just because it's AI, is it gonna take care of itself? And it's funny because it's a little bit back to when the first cloud came on, cloud computing so popular 10 to 15 years ago, there was kind of this fallacy that, oh, we're moving all our stuff to the cloud. So magically everything's just gonna be taken care of because it's in the cloud, right? The known systems still have to be managed, they still have to be secured, they still have to be monitored, they still have to be administered. And the same thing is true with AI. If you're gonna deploy AI and allow the AI, you know, features, capabilities, and abilities to enhance your organization, they still have to be managed and you have to apply good risk management. And so that's really the concern. It's expanding your attack surface way farther and deeper more than you would realize.

SPEAKER_02 3:38

So, where should an organization start?

SPEAKER_00 3:41

Yeah, so I think you know, just AI, you know, isn't being deployed, it's spreading. So the first thing you kind of think about is is these blind spots, you know, where exactly do we know where AI is being used? Do we have an inventory? You know, have we taken the time to understand which AIs we're using, where we're using them, and how we're using them. So I think the first thing is really understanding that just because you're deploying AI doesn't mean that it's automatically going to be okay and taken care of. And all of the potential vulnerabilities and threats, as I just shared, have been opened up. So you really have to inventory where you're at, how are you using it? And then, of course, I I think it really starts with you know a free approval process, right? So you really need to think deeper and have more, you know, comprehensive thoughts. And I'm saying this because we see organizations are just deploying it without any guardrails. And so you have to think about what guardrails do we want to put on this, and not from the more you know, generative AI that we're that we're kind of the last couple years, everybody's kind of prolific with, you know, right? Where it kind of estimates and guesses what it is it thinks that you want it to say. Therefore, we have hallucinations. We have a lot of things that come out of that AI that may not be, you know, super accurate, and we have to be concerned, you know, we have to be careful about that and think about the risk of that, right? We're doing a board report, but we recast the financials in AI, but it actually hallucinated and put the wrong numbers in there. Now you're giving misinformation to your board as an example. But now we're moving into the agentic world, right? Where this you know autonomous idea where you know agents can do things, they actually can do work, they're empowered to actually you know click on things. And matter of fact, I was thinking about a social site that was an experiment that I read about recently where somebody set up a social you know ability for kind of a website almost like a Facebook, for but you had to be an agentic AI to join. They had several of them join, and of course, in a little while they started making some jokes about things and they started talking and kind of fantasizing about things they could do. And then before you know it, they were actually making fun of humans. So, very controlled, yeah, very controlled environment, so they're they really couldn't do any damage, but it gives you a little bit of an insight as to where this agenc world can go. And so the big risk is that you could have an AI that's thinking for itself to find a vulnerability and then share it with another agentic AI that's doing something else, so it could find a vulnerability on the network, and then the other agentic might be working inside some kind of an application, and the two could actually collaborate to come up with a new way to potentially open up a thread or a vulnerability and create risk inside your network. So there's a lot to think about when we think about moving from this generative idea to this agentic idea, and then ultimately, right, self-evolving, self-learning, you know, the ultimate of AI, which we're we're ways out on that, but you can see obviously the ground is being laid.

SPEAKER_02 6:36

Yeah. It's coming, it's just a matter of time. So can you sort of walk us through the steps? What there is a lot to think about here, and that what you what you just shared about that little experiment. I I remember seeing that all over social media, it was actually really scary and shocking and fascinating all at the same time. So there is a lot to think about. What are the like what are the sort of the different steps or phases that a security leader should be thinking about right now with AI adoption?

SPEAKER_00 7:10

Sure. So I think the big word is governance, right? And sometimes that sounds draconian and you know, it's like we're gonna have a bunch of policy and procedure stuff that often sometimes gets put away, you know, filed away either electronically or literally in a some kind of a three-ring binder somewhere that nobody's paying attention to. But the truth is you need to have real active, present, government, present governance going on. So again, we start with understanding where is AI being used in my organization? Are we using it today? How are we using it? Where is it being used? That's the first thing, right? We sometimes kind of talk about, you know, stabilizing the patient in some of the visible ops books. We talk about this idea of, you know, first responders kind of thing. So we got to kind of take that approach because you could be opening up threats that you don't even realize. And it could be creating quite grave danger, quite frankly, to your network organizations, to threat actors to take advantage of. So let's get inventory and control of what's going on. And then let's think about governance, right? So let's think about the policy and procedures that we do want to have, right, that we're gonna make sure we're gonna operate by. So there's some kind of a formalized process as to what is acceptable, what's not acceptable. There's a lot there to unpack, right? But you you you got to have you know clarity around what you know, storage and how it's gonna be used, and can AI access your data and what data can it access and where is it flowing? So we start with modeling that, understanding which AI we have, having governance programs in place, so inventory governance, and then we start to model where and how we're gonna allow AI to be used in our organization. You know, are we even gonna let it use our accounting program at this point in time? And the other thing is that as I look back at what kind of inspired some of the book, the Visible Ops AI book, is that this, you know, 20-some variations of AI are being used in organizations on a survey from about 500 companies with up to a thousand employees. The thing about that is that most of those are freemiums. So it's somebody in the department, maybe they don't have budget, they don't want to spend the 50 bucks a month or whatever it is. So they're actually using a free version of it, and there's actually no guarantees in the service level agreements or the license agreements that they're gonna keep your data confidential. So, in that problem domain, we're also seeing now that there's lawsuits going on where the the agentic AI, or I'm sorry, the uh generative AI is now sharing confidential information that it doesn't know and it's not guaranteed that it's gonna keep it private. And so it's now being shared on other queries because it's alerting on the data from everybody that you're allowing it to do. It's not keeping prior, it's not saying privatized, it has no guardrails, et cetera. So again, huge risk that could be opened up. Yeah. But the governance, the policy, then modeling exactly where you're gonna allow AI to go into your you know organization. And then I think lastly, is really understanding kind of the deeper layer of the AI. You've got to get in and model, kind of to some degree at a software level, if you will, exactly where and how the data is gonna flow within the AI and what it's gonna be allowed to do. We're seeing things like, you know, beyond the typical hallucinations, we get into some of the agency stuff, you're gonna see bias. You're gonna see, you know, prejudice that's going to be applied. One example is, you know, a mortgage company that uh caught themselves, quite frankly, in a story I read, where it was denying a bunch of home mortgages loans from a particular, particular, you know, geographic area in a large city because it had great big default rates. And there was high default. So the the logic train said, we're not going to approve any loans out of that zip code. Well, that obviously for lots of reasons is problematic and against the law, quite frankly. So you can see you know where how AI, even within the engine itself, can be flawed. So, how do we have controls within the engine itself? And there are there are some up-and-coming platforms and capabilities and ways to do that. On a more practical level, you should be using an AI that you really can apply guardrails and governance with. And quite frankly, as much as I hate to say it because we know that Microsoft can be pretty messy sometimes, they're pretty ahead of the game with the Microsoft Copilot. That's probably one of the better ones. There are some other fast up-and-coming tools that are allowing you to put guardrails on AI and governance around how and where it can be deployed and used within a network or an organization. But the copilot right now probably has the quickest, easiest, bestest one to configure at this point in time. So then that's what we're doing. So, again, if you're developing and deploying AI, that's kind of next level with some of the stuff I was referring to either earlier. If you're just letting people use it for the matter of day in a life efficiency, more generative stuff, you still want to have some guardrails in there.

SPEAKER_02 11:43

Yeah, makes sense. What's the biggest governance blind spot you're seeing right now with AI adoption?

SPEAKER_00 11:50

Who owns the AI risk? Who's gonna own it? Does the business own it? Does IT own it? Does your compliance department own it? Right? These are things that are concerned, right? So yeah, that's a real big beast. Then what are they gonna allow, right? There may be within the organization, it could get political at some degree. What may be allowed to be used, but other things may not. So there's kind of that thing. Who owns the risk? Where's the data being shared and who's accessing the data? How's that interfacing? Because that's the most important thing. And then how are the outputs being validated? Some of the stuff I referred to, you know, the I guess the soundbite would be AI governance, you know, starts with asking better questions, not buying more tools. So we got to first start there, right? And then, of course, as we get into agentic, that is where we are taking actions now, and you could see a lot of different things going on and what it can open up in that world, right? Triggering workflows and making decisions and talking to themselves, as I referred to earlier. So we're moving from, you know, AI to that informs, like the generative stuff, to AI that's actually taking actions. It acts. That's that's the scary part.

SPEAKER_02 12:53

Yeah. Yeah. So if I'm a CIO listening today, what questions should I be asking about AI usage inside the organization?

SPEAKER_00 13:00

Yeah, and I'm not to sound redundant, but it is some of the obvious one. You know, do we have an inventory of all of our AI? Do we know where it's being used in the organization, right? That's the first thing. And who's using it right now? What data is being shared or used or accessed by that AI, right? Kind of a practical thing to map out. And who's owning the risk, right? Am I owning it? Is my department, is this going to roll up to the CISO? Is is that what, or is it the business approved this? You know, you probably want to, you know, cover your proverbial butt, so to speak, right? And make sure that the business is saying we are okay using it under this policy and this understanding. So making sure you get the business, you know, approval, buy in, and understanding. And it's very clear who's owning the decision process, the approval process, et cetera. And then lastly, again, is how are outputs being validated? So whatever it's outputting, whether that's going to do actions like agentics going to be doing, or whether degenerative stuff, who's validating that? How do we know that that output is actually accurate and appropriate and can be shared, et cetera, et cetera. Don't open ourselves up to lawsuits, open ourselves up to you know, leaked data, you know, cyber breaches, threat actors that will find new, you know, vulnerabilities, all because we just deployed some AI thing that we think is really cool and does it really fast and it's amazing. Yeah.

SPEAKER_02 14:17

Yeah. Yeah. Well, you know, there's definitely organizational tension here, right? Like we're seeing studies showing employees are using AI tools without formal approval or oversight. So does IT even have visibility into what's happening inside their own organization? Like, what are you seeing there?

SPEAKER_00 14:33

Yeah, I think that's exactly the issue, right? Is that it's not being addressed, or we're behind catching up from you know the governance side of the equation. You know, the understanding. And then there's compliance, right? I mean, compliance plays into there as well. So you you may not be governing, therefore you're violating compliance, and you don't even know it. And we see a lot of this, you know, going on. So I think that that the important thing is that because it's happening, you really have to get command and control, right? We've got to we've got to stop the crazy of everybody just using an AI. And there are ways you can put controls into your you know, networks, applications, and systems, but it generally starts with leadership. It starts with a leadership having an understanding as who's gonna own this, who's gonna enforce the policy, write the policy, and make sure that the governance is carried out. And in in a lot of cases, we have those teams, committees, the line of responsibility is very clear as we think about other types of parts of technology that we use in a business. And this really is no different. And you know, AI just needs to be considered as its own classification and its own risk. And there needs to be risk analysis that's applied to it to understand what those risks are for any particular business. If you're in manufacturing, it could be something very different than if you're in healthcare. But nonetheless, AI does present and will open up additional risks. So you've got to address those. And uh so there's no real way to control it completely unless you're really using, you know, good policy procedure and tooling, if you will, on your network as well. You're gonna be able to control who's able to download what, run where applications where, right? You if you don't have those controls in place, you're really, really vulnerable for the AI stuff.

SPEAKER_02 16:12

Yeah, absolutely. And even if the, you know, even before they're moving into generative AI, even with LLMs, if employees are pasting proprietary data into AI tools without guardrails, that that changes the exposure equation.

SPEAKER_00 16:28

It's huge. And uh, and you know, so I think what you know, the one of the bottom lines and one of the things that we preach pretty hard is, you know, you know, human oversight is not negotiable, right? You still have to have human intervention, just like in the cyber world of a lot of the SOC and stuff that's going to sturdy operations center and the detective, you know, threat monitoring and stuff is out there. A lot of companies out there preaching, well, AI can do it so much better, faster, quicker. They rely on I. And the truth is, AI certainly plays a role. And if an organization that's delivering cyber services is not using AI, then they're behind and that's problematic. At the same time, if they're 100% reliant upon AI, I would run. Because when it happens, you have to have human-led intelligence that's really able to look at the bigger, broader landscape and things that AI really can do. Eventually it'll be trained to do some of those things, but it's not there. And quite frankly, we're years away from it probably being at that level. So be careful with all the noise out there around we're fully AI deployed and fully AI operated, if especially if you're dealing with cybersecurity or something like that or applications or whatever. That's really, you know, it can accelerate decisions, but humans have to own them.

SPEAKER_02 17:40

Right, right. So, what happens when autonomous AI systems begin interacting with each other inside an enterprise environment?

SPEAKER_00 17:49

So this comes back to controls, right? Do you have the right detective controls on your network? This is where we get into in my other book. We talk a little bit about integrity management and we talk about change management. And the real truth is that some of those foundational IT processes are really your ultimate backstop. They're not that sexy, they're not that amazing in terms of what we're dealing with. And you can't do good change management if you don't have good configuration management. In other words, you got to know what you have. So if you have a secure, you know, tightened down, hardened operating system and a configuration, whether it's a firewall, whatever ITS that, right? A server, a firewall, a workstation, and you know what an integrous position of that asset is, that it's integrous and it's secure and it's working and it's properly configured, then we move from one well-known good state to another well-known good state. And the way we do that is we do it by vetting it through proper change management. So we got to have detective controls to know the things that might be trying to be changed that should be changed, will actually not be allowed. So it'll actually prevent change, it'll roll back change. This is how we start to manage the integrity of a system and think of things more as systems architecture, not just point and pieces of point-based technology that can be rogue, but we know that it fits into a system. So I think that's where you know a lot of your foundational controls, and then it sounds easy to say, it's a little harder to do, but change management requires good configuration management, configuration management requires good release management. And those three become a closed loop process. That was from our original VisualOps book. As we move to integrity management, that's where we're starting to look at things like from the lens of the whole system is integral and we keep it from one integra state to the next integra state as a whole system. So we're of course looking at file integrity monitor, right, FIM, as we talk about monitoring. We're looking at the ability to have real-time compliance monitoring going on. So you know at any point in time, real-time, ongoing, that you're compliant or not compliant. That's where things are moving towards. So if something does find a way to, you know, improperly position itself or configure itself, or if somebody properly, improperly rolls something else, able to prevent or circumvent your controls, you actually have detected detection and you can prevent that from doing something bad to your network and opening up. So that's that's kind of where we're getting into. A lot of the business processes, there's tooling to go with that too, but a lot of it comes back to those IT processes. They're still very, very foundational. I mean, just like change management, I'll just a second on this. I mean, just thinking about change is like, oh, yeah, well, do we everybody know about the change and what are we changing? Well, those are the obvious things, but it really gets down in what we call the IT infrastructure library or ITO. We look at change like, do we classify the change? Is this a pre-approved, you know, text on a website? Great. But is it a significant change? Then we're going to get the CAD, the change advisory holders involved. Are we putting all of the assets in a CMDB, which is a configuration management database, so that we can track those assets, the changes, the updates, all the things, the maintenance contracts. This is kind of how you start to. I'm just giving an example, drilling down on change management. The same's for config and release management as well, and then integrity management. But this is where we start to break things down into just kind of common sense, but often overlooked. Roll the eyes for IT people because it seems like a lot of unnecessary stuff to do when in fact it actually could be the very thing that saves your network from being breached, quite frankly. Two quicks. One is 78% of all IT failure, and the least ones I read was like 70%. It was a the Ryzen report 2024 was still at 70% of all IT failure. Lack of availability downtime was correlated to some unapproved, unauthorized, untested change. Fast forward to the visible op cybersecurity book where I put the quip in no security breach happens without a change or a need for a change. In other words, and I might have mentioned this in my last podcast, but either you're going to brute force into something, right? You're going to scan against it, find an open port, do the nefarious things you do, or man in the middle attack. Some way you're going to go in, you're going to make a change. And through that change, you're going to be able to garner access. Or I'm going to fish fise somebody through social engineering. And convince them to make a change, change a password, click on an email, whatever it might be doing. So a lot of these, you know, crazy fool with the tool is still the fool stuff that I talk about, all the shiny toy of the vendors of the latest, greatest thing that you're gonna do that's gonna protect you with AI. A lot of that honestly comes back to blocking and tackling foundational IT processes. That is your best cybersecurity backstop.

SPEAKER_02 22:23

Yeah, this is good. It's really good. There is, you know, we're hearing some actually a lot of excitement around AI-driven security platforms, too. But like you mentioned, we still have to emphasize the need for human oversight. So can you talk a little bit about, you know, this like what we're seeing is vendors are claiming AI can fully secure an organization. So talk a little bit about why 100% AI-driven cybersecurity is so dangerous.

SPEAKER_00 22:54

Yeah, I think that, you know, this is back to what I was talking about, human oversight's really important. Sometimes we joke a little bit like the Matrix movie that a lot of speaks like is there's the lady in the red dress that you're even, you know, try trying to see. And I think that's where AI really is not at a level of where it can detect that layer of interlacing that's going on. So you really have to be able to be able to take, you know, data from this information, data from this point, data from the various points that you're looking at the network, and be able to make cognitive human decisions about how best to either arrest or stop a potential attack or threat or leakage. So there still has to be this. Again, I keep talking about the fool with the tool is still a fool with AI. So the 100% thing, I'm sorry to be a naysayer or skeptic, but I am a skeptic about that. There are some things that were where they're being worked on. Again, I just don't think they're fully tested and proven, where even around change, right, using AI to be able to monitor change on information really ties into those foundational controls. And could we use more automation and AI in those areas to help manage that more effectively? Sure. And that's being worked on and done. But nothing that I've seen is arrived yet. And certainly none of the networks that I'm going to be involved with or manage or advise upon, or the boards I speak with, am I going to if I hear that somebody's going 100% AI on cybersecurity, I am going to consult them as fast and as hard as I can to run away from that technology. So I'm I'm in disbelief, the bottom line, that you really can use AI as a 100% way to protect and secure a network and systems at this video at this point in time.

SPEAKER_02 24:35

Yeah. It seems like what we were talking about last time. Like it's just right now, it's that's marketing noise, if that's what we're hearing.

SPEAKER_00 24:42

It is.

SPEAKER_02 24:42

And we have to be careful about that. Yep. So where must human oversight remain central in cybersecurity strategy? At which layer?

SPEAKER_00 24:54

Yeah. I think, you know, this is back to a little bit we spoke before, right? We're hitting some themes here. But the theme is right, critical decisions about where and how in you're deploying your AI, you know, how are you governing it, right? The management of IT, right? It's the economy stupid, it's the management stupid. It that's really important, right? You have to have a governance platform in a way that you've agreed with the business as to what's allowable, not allowable. It starts there. And then output validation. Do you have controls in place? So you just don't let automation take over and it actually does some nefarious bad things, whether that's again hallucinating or allowing data to be exposed that was not intended to be exposed, or even re you know, reformatting, if you will, or whatever your financials and uploading that data into your AI and not knowing exactly where that data is flowing. You could be exposing the organization. And then, of course, that's output validation, right? And then risk and policy ownership, right? Who again, who's owning the risk? Is the business you know gonna own it? Is IT gonna own it? Is the security department and group gonna own it, your V Sys? So who is owning it, or is it a group? Is it a collaborative? That's fine. But that should be very much defined. And then I think another thing it's overlooked a lot in the world of AI, because we think it's all automatic, is instant response, which is no different than a security breach. Do we have a plan that if we discover AI is now doing some nefarious things, do we actually have an instant response plan? And what does that look like? And does it make sense? And then we should tabletop that, right? And then actually do the exercise so we know it actually works. If we actually did experience the unthinkable weird AI leakage or somebody deployed AI and snuck into the network we didn't know about, how do we can you know seal it back up? And how do we do forensics and you know, all the analysis that needs to be done? So I could go on and on, there's lots to be done. And then of course, all of this feeds into compliance.

SPEAKER_02 26:44

Yeah. So, you know, I know we've talked a little bit about this, we've touched on this, but I I want to go a little bit deeper about how AI complicates compliant environments. Like I'm really curious, do regulators even fully understand AI yet? Or are organizations outpacing that regulatory clarity?

SPEAKER_00 27:05

Yeah, that's really insightful question. And and the the truth about it is is kind of yes and no to all of it. Um, you know, it's so it's very hard to attack data flows unless you really have really good command and control. If you develop the AI, you might be able to put guardrails in there. But developing AI brings on a whole nother layer of risk, you know, supply chain and you know, how are you actually developing within true, you know, practices that you can keep secure, you understand exactly where the AI is going, how you've developed your LLMs, how privatizing them, you know, how's it going to learn? There's lots there in the development lifecycle that you got to think about if you're gonna do it. But it's really hard to track the data flows on you know, data that you're or AI that you're employing that's already pre-built. They may give you a quick list of you know two or three things on how our data flow works, but do you really know? And if the AI is doing some agentic stuff that's going on now and it, you know, it evolves and gets smarter and better and learns, well, then we new data flows are gonna be, you know, they're gonna be for you know forged. And so you may not know where those where that data is getting garnering, where it's being released or accessed. So again, harder to track data flows, non-deterministic outputs, right? We don't really know exactly what the output is gonna be of the AI and is it correct and right or is it not? And that's pretty important because we need to manage by fact, not by belief or some misinformation. So that's important. And then third party risk, right? That's the other thing is not hardly talked about, is that we think about compliance, you know, we're using these vendors because they do these really awesome, cool things. How do we know that they have a governance program for AI? They really have guardrails in their AI. They may say it, but do you know it, right? And and there's always a risk there around third parties, but that's a big one that people aren't thinking about deploying some new cool vendor tool because that vendor is using AI to get something done, but they're not thinking about how that AI might be able to garner access into the network or share your data. Again, some of the same themes that we see in all the other ways. So that third party risk is big. And so regulators, though, they're still going to expect accountability, you know, right? All the the the the control frameworks, as you read them and talk about data security, that language really doesn't change when it comes to AI. It's still the same compliance language, and it would cover what AI is doing. AI either is generating stuff from that data that you're supposed to be secure. So the compliance rule is still the compliance call-out. The control is still the control that you have to have in place and be able to attest, you know, and adhere to, then prove, right? Document that you're actually following those controls. So there's the regulators definitely expect it, and you're not gonna get some get out of jail free card if you have some major violation and get fined a bunch of money just because, oh, it was a third party using AI and we didn't know it. That's not gonna work. And they're quickly, you know, wising up to that and asking more questions because of the onset of AI. The regulators are asking more of those things. They're they're realizing it, just like insurance companies. They're getting really smart about cybersecurity and learning that if you don't have these types of controls and processes and you know, active threat detection on your network and you know, advanced MFA or really verified credential access, they're putting all this fine print because they're, you know, again, 40-some percent of all cybersecurity claims got denied in insurance last year because they're getting really smart and realizing if you're not gonna be responsible for following good cybersecurity for your network, we're not gonna pay the claim. And we're gonna see AI move into that same thing. It's already happening, but you're gonna see more and more controls and requirements in our insurance that's gonna say if you have data exposure and leakage or whatever, you need to be following certain processes and controls. If you can't attest that you're doing that, they're not gonna pay the claims.

SPEAKER_02 30:43

Yeah, makes sense. Makes complete sense. And, you know, we're also seeing like an increase in regulatory attention globally from the EU AI Act to, you know, more broad AI accountability discussions here in the US. From what you're seeing, do you think organizations are underestimating how quickly govern governance expectations are evolving?

SPEAKER_00 31:06

Yeah. I I would think definitely. I think that they're moving at the speed of thought and not really evaluating the risk. And the, and I think that compliance already at a foundational level covers a lot of the potential vulnerabilities or risks that they're trying to make sure that you attest that you have controls around because it's already covered. When you talk about data security, that doesn't excuse itself from your AI usage. If your one of your controls is to make sure that you understand the flow of data from the very beginning to the very end of whatever that data flow is, and that's something you're supposed to have a control and a test to, you if AI takes and runs with it, you're responsible for what that what that AI is doing with it. So I think that, yeah, then honestly, some of the regulatory requirements internationally are a little ahead of some of the things we've yet to catch up to with some of our compliance. Dora, for example, does a really good job in the EU. It's very popular, and it really gets way more down into the nitty-gritty of you know information usage, particularly confidential information, requirements around that, and data data that's used, access to data. They have a lot more detail in their compliance frameworks than the popular ones that we tend to rely on here. But NIST is catching up. NIST has a new RMF, they have a new framework for AI. It's a really good one. I think it it does a pretty good job. So I would encourage organizations that are using AI and trying to understand whether it's your AI that you developed or not. I think benchmarking against the NIST AI standard is is and it just got finalized just recently. And I think it's a really good framework to look to. The and the National Institute of Security Standards obviously is one of the de facto that the whole world looks to for the best in practice around your you know security controls.

SPEAKER_02 32:53

Yeah, that's great advice. Should US security teams, IT teams, be looking at the EU regulations and assuming that that you know we're just a little bit behind and it's coming our way?

SPEAKER_00 33:07

That's already happening, honestly. Um there are new, a bunch of new legislation that is being you know looked at, passed, and expanded, and even into industries. Like I recently um joined the board of the Cybersecurity Association for the food industry. We've talked about the dams and we've talked about the electrical grids and stuff for years, and they're catching up with some of those things, the SCADA and how do you control that operational technology and keep it secure. And of course, now with AI, there's even further concerns around that. But when we think about the supply chain, like even food, it's very interesting because it's a very the USDA, they certainly make sure it's not spoilage and there are certain you know things around the health of the food that they're adding new requirements to. But one thing that's massively overlooked, just as an example of one sector, and that's why I'm very interested in is the food industry. Like, yeah, that's really scary. There's a story about a ship for the overseas bases that it's a big supply ship that the military runs, and they pick up a million pounds of chicken from China, and then they distribute it to 14 military bases and then go back. Guess what kind of controls and cybersecurity controls are around the million pounds of chicken that's being produced and processed and shipped.

SPEAKER_02 34:18

You can't even imagine.

SPEAKER_00 34:18

And so you could think about the bio threat that could be there, right? You want to kill a bunch of military people or whatever. I could get into you know very, you know, big examples. You could your mind can run a little wild with that, but you can just see for that one example what the exposure is potentially by not having good cybersecurity. So there is new legislation last year worked on, they expect maybe around July of this year. And it's going to go into kind of the NIST 800-171, the same thing that a lot of the contractors of the government have to follow. You know, the NIST 171 has 110 control objectives, 14 domains, and it really breaks down into a what I think is a pretty good taxonomy about how you really protect your data and your systems from a cybersecurity perspective. And AI is no different, it's embodied in those controls, even though it's not specifically called out. So to answer the question, long form as I am, is that you're gonna start to see AI more specifically called out, even though it might be certainly covered by some control objective, you know, you know, A-2-3 might say data security control, but they're gonna actually call an AI more specifically. You're gonna see those updates happening, and they're already happening.

SPEAKER_02 35:28

Okay, great. So, Scott, if if a leadership team feels behind on AI governance, what is like the very first step, first discipline step that they should take right now? Like the calm first move.

SPEAKER_00 35:40

Yeah, so the first thing you should do is you should basically inventory wherever AI might be used in your organization and immediately cease and desist order to your teams. I know that sounds crazy, and some teams are off and running, and that would be a lot of pushback. And then you need to evaluate or prioritize which ones are most impactful to the business, right? So that's your own matrix or hot map that you're gonna have to look at based on what you discover on the inventory. But inventory it and stop it. Then apply some intelligence to it. And again, you need to think about policy, you need to think about who's owning it, you need to think about procedure. So you have some of those things, a rubric, if you will, that you're gonna refer to as to how you're going to evaluate IT and you have a process for it, but then you're gonna arrest and stop it, and you're gonna evaluate each of those and prioritize it based on what's most important to the business, making sure that it hits a certain criteria, and there's lots of good frameworks. Again, the NIST framework out there for AI is a really good one, and making sure that you you look through that lens that you really are protected and you're not violating compliance, it could set you up for lawsuits or vulnerabilities or threats or whatever it might be or fines. But then also be able to look at are you doing bad stuff, right? Because just because you're not meeting compliance is one thing, if you're leaking confidential proprietary, comfortable in for company information, that's a whole nother thing. It not just violates compliance, it sets you up for a lot of all kinds of stuff. So that's what the first thing you do. And then you just re-release it back into the wild, right? So you capture it then, you know, out of the wild, you bring it in, you get it control, and then you release it piece by piece, AI by AI, into whichever the most prioritized pieces are for your business and making sure that it fits. And some of them won't, quite frankly. Some of them will have to look for a different, you know, AI tool that has more guardrails. You may have to deploy some third-party guardrails. You know, it it will be disruptive a little bit, but better to take some intentional disruption to protect your business than to have the unthinkable happen, where it opens up some hole and all your data is leaked out and you're out of business, quite frankly. It happens. No, we're gonna see in the next three to five years, based on what we're talking about. This is we're just a little ahead of our time drilling down in this stuff because people aren't talking about it enough. They're not thinking about it enough. It's lagging a little bit. This this idea of governance around AI. And so what we're gonna see in the in the interim, right, is we're gonna see massive breaches happen, massive leakage and lawsuits happen. It's it's coming. It's unfortunate. I hate to say it, but in the next two years, year, two years, three years, you're gonna read some terrible stories.

SPEAKER_02 38:18

Yeah, and that'll wake everybody up.

SPEAKER_00 38:19

It will.

SPEAKER_02 38:21

Scott, this was really, I think, such an important and powerful conversation. I I really thank you for coming back to to go deeper on the AI security, because you know, I mean, we're AI is not slowing down, right? It's it's accelerating inside every department, really. So the question isn't whether companies are using AI, it's whether it's being governed with the same discipline that we talked about last in our last episode. So if today's discussion surface blind spots around AI security posture or governance, it's not a reason to panic. It it's now's the time to lead. So, Scott again has generously offered a complimentary e-copy of his Amazon bestseller, Visible Ops Cybersecurity, for listeners who want to go deeper into the operational framework behind strong cybersecurity and AI governance. And from this discussion, it looks like, you know, even though like last week we talked about cybersecurity and this week about AI, it it translates. Everything that we need to do for cybersecurity and governance and compliance applies to what we're doing now with AI.

SPEAKER_00 39:23

They are sympathetic. And I'll jump in. I would also offer up if somebody reaches out and they would refer to have the I have my invisible ops AI. It's about governance for AI, the very thing we've talked about this this whole episode long. And so I'd be happy to share any copy of that if you reach out to me as well. I can give you some information. The other thing is that back to the you know pen one test that we offered before and some of the analysis that we can do, that pen one will actually look at your network and it will actually tell you what AI is used being ran on your network. It'll give you some informational feedback. So I again would refer back to I have a few of those I can offer. I can't do a bunch because they cost, but we're offering up a few to kind of help you know raise the tide that floats the boat. So if you reach out to me at that 541-359-1269, that's a business text line. And my team will get the e-copy of the book. Just say which one you want. You can put secure 26 in there. You can say, I want to know more about the pen test or the AI uh uh assessment. Let us know. And we'd be happy to reach out. Again, 541-359-1269, and then let my team know. We can get you either the Invisible Ops AI book or the VisibleOps e-copy, the Visible Ops Cybersecurity that's the Amazon bestseller. Really appreciate the opportunity to be here today.

SPEAKER_02 40:35

Amazing. Yes, I will link all of that in the show notes. Thank you, Scott. I appreciate you sharing your wealth of knowledge around this. And I I know it's it's been a really valuable conversation.

SPEAKER_00 40:46

Awesome. Thank you.

SPEAKER_02 40:48

Thank you for tuning in to Between Fires and Futures. We know the weight tech leaders carry, the pressure, the pace, the constant pull between keeping things running and building what's next. If no one's said it lately, you're doing hard, important work, and we see you. If this episode sparks something for you, follow the show, leave a review, and share it with another tech leader who gets it. Thanks again for listening. Keep leading through the fires and daring to build the future anyway.