AI Needs Rules: Why Cybersecurity Leaders Are Calling for Real AI Regulation

Welcome to Security Boulevard, the cybersecurity podcast from the Futurum Group. Each episode explores a variety of topics within cybersecurity and the technologies that drive it. com, the Security Boulevard YouTube channel, Techstrong TV, and all of your favorite podcast platforms.

Let's meet today's hosts before we jump into our fun topic, starting with Fernando. Fernando, it's good to see you again. Absolutely.

Hi, everyone. Uh, Fernando Montenegro. Uh, I'm, I'm eager to, to, to, to dive into the topic.

We're-- as, as we're chatting, we're also preparing for the, the, the RSAC conference coming up, so, it's a good thing. I just hope that the cold I have kind of weans off before, but I think it'll be fine. Yeah.

You, you'll be fine. Just, drink some tequila. That seems to be the, the, the general solution to- Oh my God ...

getting rid of whatever ails you. Um, also joining me, of course, is Mitch Ashley. Mitch, it's good to see you as well.

Thanks. Great to be here. And tequila, by the way, just makes you not remember how bad you felt, so it's okay.

So I didn't mean to make you cough there. You are sounding better, Fernando, so it's all right. For the record, this is water, okay?

Just- Okay. so you know. Okay.

Good. It's a family show. Um, good to be here, guys.

Yes, and RSAC is on the doorstep, you know, right around the corner here, so that's top of mind, of course. But there's also plenty going on in the world, in world of security, so I imagine we'll have plenty to talk about. Oh, absolutely there is.

So let's jump into the topic, and I, I get to finally say something I've been waiting years to say in my little Gen X heart. Regulators, mount up. Uh, if, if you, remember what song that lyric's from, you should probably take some ibuprofen 'cause your back's hurting.

Uh, but what I wanted to do today is talk a little bit about regulation around security, because we've seen a lot of things come up in the last couple of weeks that make me kind of wonder, how we should be approaching some of these, massive advances in technology and automation and things like that. Uh, you know, look, we, we've, we've beaten the Claude bot thing to death at this point. Uh, yeah, I think we've referenced it, you know, a couple weeks ago.

Uh, the head of, of Meta's AI safety division had it delete all of her email, because, well, she didn't tell it to stop. Actually, she did. She told it to stop, like, four times.

" Uh, and then we have, people that are proponents of this AI and, and not just, folks like Sam Altman, but people like Marc Andreessen, who not only say we shouldn't regulate it, but that it should be illegal to even attempt to regulate those kinds of things. But as a security person, I no longer believe in the goodness and faith of humanity. Uh, I know that everyone is evil and mean and needs to be told what to do.

Uh, so, you know, throw that one out there for everybody to leave a comment about. Uh, but I, I kinda wanna throw this out to, to the two of you. Um, as we see these things get more and more complicated, as we see more and more inputs causing the decision-making, outputs to happen, should we have some kind of system in place that functions almost w- you know, we say guardrails, but it needs to be, like, an inviolate rule.

Things like, oh, I don't know, you can't kill people. Or convince c- people into killing people. Uh, w- right?

Like, well, and, and this... There's been a lot of articles that have come out recently about things around, like, it's called AI psychosis now. We'll probably come up with a DSM version of it later.

But, like, the AI just goes right down this rabbit hole of feeding into whatever a person says that causes them to do it themselves sometimes. Should there be a system in place and not just a, "Oh, we'll, we'll make sure that doesn't happen again," but, like, if something along those lines occurs, you know, kind of like Asimov's rules of robotics, like, no, under no circumstances can you cau- allow that to happen. Well, see, we're kind of operating under the same, similar but not regulation, rules as the internet has been, which is, if I'm a carrier, I'm an ISP, I'm just the network, right?

What you do on top of this is your business. I'm not responsible for whatever it is, illegal or not illegal. Um, w- we've kind of taken the same approach, although, albeit it's not through an act of Congress or regulation, which is we don't wanna, we don't wanna inhibit anything about AI, so let it be what it's gonna be, and then we'll deal with the fallout, with the problems that happen.

Now, if it happens to be someone commits suicide or commits a crime or whatever because of psychosis, that the AI, you know, the AI did it sounds like Son of Sam, but that's a whole 'nother issue. The AI did it, told me to do it. Um, you know, we're in this mode of let's just let, let happen what's gonna happen, we'll-- and then we'll deal with the fallout.

And I, while, you know, I think that's all good from a let businesses do whatever the heck they want, or individuals, there also are consequences to doing that, and there will be some very negative, and we've seen some already, consequences from it. And it's, it's a safety issue, right? It's more than just a regulation issue.

It's a safety issue. And that's a lot of what happened with the kerfuffle with, Anthropic and, and the Defense Department of, you know, well, you aren't reg- regulating, but we're gonna kind of self-regulate ourselves with our EULA, with our use, what you can do with our technology, like we don't want you to blow things up with it. Um, there...

But there's, that's, that's a different kind of self-regulation. Yeah. I'm, I'm, uh...

This is a deep, deep topic, and- ... I, I can bring economics into it the very first second of the conversation in terms of externalities, right? When somebody makes a decision that doesn't take into account the whole cost of something, right, they are more, they will do more of it, right?

So, uh-This is not cybersecurity, this is not IT, this is economics, right? The way that we resolve externalities is, amongst many other things, regulation, right? You bring the cost back into the, to somebody who is doing this as a way to minimize that.

And I, as much as I agree that... L-like I'm a, I'm a Gen X kid and, and I grew up under, I mean, South America, but like, like, Reagan, right? And, and, the, the, the most dangerous words, right?

Um, "I'm from the government, I'm here to help you," kind of stuff. But it's not quite like that, right? I think that we have, as a society, deployed technology widely enough that we start hitting limits of what technology...

Uh, we start hitting societal limits. And it's funny you bring up Andreessen. I mean, he wrote the, he wrote back in 2001, "Software is eating the world," right?

Uh, 2011, I think. And he was right, right? Software did eat the world.

Here we are. But the, the, the consequence of that is, look, in a society, we need some guardrails, we need some limits, right? Because that's how societies function, and here we are.

So do we need regulation? Absolutely, we do, right? Are we getting the right type of regulation?

We're working towards it, but I'm a, I'm a, a, a, I'm a free market as much as the next guy, but we do need something. And, and the, the... I think that there is so much going on that we are losing track.

So let-let's, let's split things apart, right? So the whole Anthropic and, and Department of Defense, or Department of Wars, they, they now want to be called, is, is, is one thing, right? I would argue that the indiscriminate use of AI in other things, I mean, this is where, on, on, on civil society is something that we need to be more aware of what's going on and where we can use it, and where there are, I mean, 50 states in the US and probably 50 different regulations, or, or if not more, on the use of AI.

But the call to arms, I would say, for security practitioners, is get familiar with what's going on and don't dismiss it, right? This is, uh... I, I know we talk about cybersecurity, and I'm sorry for the rant, guys.

But I know we talk about cybersecurity, but when I first came into this industry, it was called information security, right? And, well, it was always broader than just, quote-unquote, "cyber," right? Uh, we were chatting just before, the, the, we started the recording, like reminiscing about old certifications.

I'm an old school CISSP, and I remember having to, like, there, there was a, there is a module on law regulation and ethics, right? It's here. We should be aware of it.

Sorry, rant, rant over for now. Hmm. Oh, and don't worry, the law versus ethics episode will be coming eventually.

Like, that, that one's penciled in on the calendar. That, that will happen. But, but the reason why I think it's important that we have regulations is because the systems that we're building are learning from our behaviors, and that's creating a challenge.

Because what do you do when someone says, "Oh, well, that's the law. You have to follow it," and then they don't? Like, what are we teaching the, the, the learning systems of the law is really just a set of guidelines?

Because one of the things that you guys brought up was the, most people are should be aware of this by now, Anthropic was a, provider of AI systems for the US federal government, specifically, the department of something, whatever they wanna be called today. " And they said, "Okay, fine. " Um, like you're re- you're relying on a company to self-police.

That's like saying, "Oh, look, insider trading shouldn't be illegal because companies will just do what they're supposed to do," right? They, they'll do the right thing. And if you look at the law, the right thing for them is to make as much money as possible and damn the consequences.

You know, and I'm sorry, to make the most money for their shareholders. And see, that's where you're getting into the nuance of the law, right? And that's why regulation not only has to be explicit, but specific, because when you let people have, creativity within regulation, that's where we start running into specific kinds of problems.

"Well, you never said that it had to be active. " And it's like, well, what are, what are you trying to argue there? And so when, when we, when...

And, and those are the kinds of things that the system will learn, and then eventually it will parrot back to us, what we've seen because that, those are the stories of the earliest chatbots, right? You expose them to the internet for 20 minutes and they turn into, not nice things, we'll put it that way. So- Well, Tom, I think, you know, there, there's different kind of regulations.

There are protective things like Gabe is referring to, Section 230, right? Right. The Communications Act that protects ISPs.

That, that's very pro. The- Mm-hmm ... I-ISPs, obviously, you know, who lobbied for that.

It wasn't, it wasn't the consumer worrying about whatever. Um, but there's also... Y-you can't regulate yourself out of every edge condition, right?

Especially, and I think that's the fear with AI, is we don't know enough about it that we should regulate it, but we don't know enough about it that we don't know what to regulate it. And you combine that with we're in an era ofI don't know what the right term of it, of it is, term of art is, sort of not trusting the expertise of people in this, in the, the domain that they're an expert in. So if you're gonna write regulation, but you aren't gonna invite people to the table who either know AI really well but have less of a bias influenced by financial means or driven, also that not only know AI, but also know the security side of it as well.

So it takes a combination of things to do it well. So, but we don't have that today. That's not an environment that we're working in, that we live in, at least not in the United States.

Um, it's, it's more true, I think, in, in Europe, but it's not perfect there either. The, the, the fallback for that is either all the way back of, to your point, Fernando, let's just rely on what the companies wanna do, self-policing, right? Which is, who knows why Anthropic decided they wanna stand behind a couple of rules, and I have a-- and I have some thoughts about it, but the, the middle ground or somewhat middle ground is the whole define your policy and then prove that you met it.

That's what so much of certifications are about with companies, right? So if you wanna be certified for this trustworthy computing or trust, trust online trusting for websites, they don't tell you everything that you should do. They say, "Define what you think you should do, and then prove that you did that," so that the people who are doing business with you could say, "What's your policy, and are you-- is that what you're doing?

" So there, that's sort of the trust but verify. Okay, if I'm willing to buy into what your, what your level of what you're, doing today, and that I have some way of knowing that you've done it, then I can accept those terms, lack of a formal regulation. Yeah.

I, I, I'm, I'm latching on to something that, sorry, that, that, that, that Tom mentioned about, shareholders and whatnot, right? So this is, Milton Friedman and, and, and the, the, the, the theory of corporate governance of shareholder primacy, right? That, it wasn't always like this, and there...

I, I'm, I'm, I'm not going to-- I don't know enough to argue both sides of the, the, the, the, the argument, but there is something to be said, people are questioning that now. Now, but the thing I wanted to bring this back to, to, to, to, to our audience is, is, is what role do you play in this, right? And the role that you play is absolutely critical because from a, as a cybersecurity professional, right, you are being looked upon to help navigate the fundamental question of, do we trust what we are building, right?

And what does that trust entail, right? And there are elements of that trust that are extremely technical in nature. Hey, are we preventing prompting injection?

Are we, do we, do we have the proper level of data security so that RAG doesn't, doesn't find information or doesn't, doesn't spill information it's, it, that it's not supposed to, and so on. But there is an element of governance, and there is an element of regulation as well that, again, I urge people to get familiar with, right? So, ISO 42001 on AI governance, how well do you understand it, right?

How well does it sit in your environment? And even if you're not going to go down the regula- the regulatory part, I think that you have a voice as a security, practitioner in how should we think about this inside of our organizations, right? Who is accountable, right, for AI failures, right?

You should have a voice in that conversation. And my perspective is that as sophisticated as it is, right, AI is tooling, right? Agentic AI is tooling at scale.

Yes, there is a level of autonomy. Yes, but ultimately, that has, that, that accountability is still human, right? That accountability may be the chief AI officer who decided to deploy it.

It may be the, the, the project manager or whatever. It may be the CEO, but, but navigating that conversation and where regulation ties into that accountability is something we should all be doing. So let me give you an example of something that's not a regulation, but it might as well be, and, and the unintended, side effects that I can see happening from that.

So, hopefully you guys are familiar with lockout/tagout on equipment where, engineers and contractors and people carry a little padlock with them, and they lo- use it to lock out a piece of equipment if they're working on it. And the fact that if there's a lock on that equipment, you don't, you don't mess with the lock. It's not yours.

And, and there are stories, right, of people who have left locks on equipment for hours and left the site, and, like, you don't cut that lock off because you don't know what's going on. They've had to recall contractors to cut them off, because a- as people will tell you, that was written in blood a long time ago because people have died because you, you bypassed a lockout. We know that, right?

That's not a, that's not, like, a hard and fast regulation. I mean, OSHA might have one, but, but by and large, people know if there is a locked out piece of equipment, you don't touch it. So what happens if there's no regulation saying you absolutely can't do that, and you can convince someone, "Well, the, the line's down and it's a million dollars an hour if it, if it's not running, and the line's been down for an hour.

" What, what kind of calculus goes in there without a hard and fast regulation that says we don't do this? Because I... And, and Mitch, to your point, yes, I absolutely agree.

But what you have to have is a set of guiding principles. And, and I come back to Anthropic, at least they've tried, right? They have enshrined things in the, the baseline of, you know, we're gonna do this for the good, we're gonna not hurt people.

Like, like you would think that that should be assumed, right? But, you know, the, like the movie "I, Robot" is literally a corner case in how to get around the three laws, right? Like, like what happens when the robots just decide the best way to save humanity is to enslave them.

But, but that's not the problem. The problem is, is that without explicit definitions, things become very murky. Like, well, you never said that a...

You know, a good example of this is kidnapping. Is it illegal to kidnap somebody? Yeah, it's a crime.

What's kidnapping? Well, it's crossing state lines with somebody. " Yeah.

Like that, that's how, like fine the regulation has to be. And so I feel like one of the things that, that we're, we're running into is that without good regulation, without explicit rules, writing it down, that's just inviting a disaster, because if you just expect people to, quote-unquote, "do the right thing" or listen to the experts, the people in Grafton, New Hampshire would love to talk to you as soon as they're not getting eaten by the bears that invaded the town five years ago. Because, oh, people should take their garbage out.

Like, that's a societal norm. Turns out that's not, 'cause under the right circumstances, people can be convinced to just not take their garbage out. Let me ask you, Tom, do you consider like NIST and ISO st- standards, or whatever you wanna refer to them, do you consider that regulation when you use the term regulation?

Because my, my belief is the, the chance we're gonna get anything useful out of the federal government in terms of regulation is, is next to zero, if it's even non-zero. So I think the real, the real effort is around what ISO's doing, 'cause there are some efforts around AI management systems and risk management. I don't know all the numbers for it, but NIST has its own framework for AI.

I think that's where we're gonna get the guidance and the- Yeah ... kind of agreed upon rules of the road, right? They're not gonna come from federal laws.

They're gonna come from that. At least that's my opinion. I, I mean, the only way that that would be, be actually valid is if, the EU said that if you don't follow ISO, we're gonna fine you like 4% of your global revenue every year kind of thing.

The, the-- to me, the difference in a regulation and a guideline is the enforcement policy, is like if, you know, if it's a guideline, you can do it if you really want to, but if it's a regulation, you have to do it, otherwise we punish you. Again, going back to the whole law versus ethics things. Um, you know, regulations are laws, guidelines are ethics.

And, and we come back to, well, who's gonna tell me to do that? I agree with you. A- a- but to be honest with you, Mitch, as low as your opinion is of the, of the US federal government implementing regulations, my opinion of corporations regulating themselves is slightly lower than that, because- Oh, I agree with that ...

if it does not produce a profit for the shareholders, they're not going to do it, and we've seen that time and time again. Like, all you gotta do is go read Upton Sinclair, and you'll know why we are where we are today. A- and, and I just, I feel like somebody who's not the company and not the shareholder needs to be the one making the, the, the call on that.

It needs to be some kind of a neutral third party, and ISO's about as neutral as you're gonna get. Yeah, the, so here's my argument to what you're saying, is that yes, I agree with the... Trust me.

I know the financial motive of companies, startups or large corporations. Don't disagree with that at all. That ultimately is, even when you're following standards or certifications or whatever, that's where those compromise, that com- compromise line is drawn.

But I don't think NIST and ISO are ethics. I think that what they are are a shared set of agreements of how we're gonna do things, but the enforcement part of it you're saying is, is missing, it actually is there. It's there in a much different way.

If the people, if people who are making financial decisions to buy things or work with companies, whatever agreements that they have for their companies to operate with other companies, whether they're vendors or partners, if those things are part of what they require and they provide evidence of that, certifications, different things, that's how that elevates into the, yeah, you're not gonna go to jail if you don't follow it, but you might get sued or you might get whatever, right? There could be some consequences to it. I don't think we're ever gonna have any, you know, oh, if you, if you made a self-guided, self-controlled autonomous missile, and you violated Anthropic's policy, you know, now you're suddenly going to jail.

That's not gonna happen. Not in today's climate anyway. So I, and I'm gonna give Fernando a chance to, to respond to that, but I just wanna say that you hit on the third leg of that stool, which is not legal and not, like, self-reporting, but community enforcement.

Right. Yeah. And, and I'll, let's l- I'm gonna let Fernando respond to that before I jump into where I'm gonna pivot to on that.

No, I, I, I, I, we can, we can, we can move quickly. The, the point that was, I, that came to my mind is that what are we expecting to happen here? I do not want, we want to balance innovation with safety, period.

Right? And I think that I, I, I am a big proponent of, of things like NIST and, and ISO and, and, and IETF, right? In the context of these are agreements that we have made to make it possible.

Here we are talking, the three of us, remotely with our audience listening across the world, thanks in, in large part to the fact that everybody at some point in time, follows RFC 791 for, for IP, right? Um, and the very fact that we have, and, and, and, and I think that we're going to h- and that we have, different enforcement mechanisms throughout, includingA amorphous market-driven, mechanism of not buying your stuff. In my past career, I was in sales engineering, right?

And one of the things we used for, for, high-end network security equipment. Well, one of the things we had to prove was that, look, our product is, compatible with or follows this whole list of RFCs, right? Because that is the quality, that is the quality hurdle that the buyers expected.

Why did they expect the quality hurdle? Well, because they needed to put this piece of equipment into their large networks, and they didn't want the hassle of this not working. This is fundamental, economics.

It's information asymmetry, right? It's how do you validate the quality of something. Again, Nobel Prize work that, that, from George Akerlof, "Market for Lemons," right?

Paper is 1970. And this is about how do you evaluate the quality of something. " And that's how a buyer makes a decision.

That is a purely economic thing that in, in my view, is separate than regulation, right? Regulation is about an external entity defining, "Look, you need to follow these criteria or there will be this level of punishment for it," or, or... And the role that regulation plays is that none of us individually here have so much time on our hands to-- a-and so much economic power to be arguing with our telco providers that, they should be giving a service that follows a standard.

So we rely on the, on, on community to do that, right? It's a, uh... So again, I go back to I'm a, I'm-- I understand the point about regulation to some extent, limiting innovation, but you know what?

We absolutely need it. I don't want innovation when I am 50,000 feet up in the air and some-- and, and I don't want the, the, those engines on that airplane to be innovating at 50,000 feet. I want a very rigorous, regulatory process that certifies those engines time and time again until they can fly.

And honestly, as much as I love free markets, and, and trust me, I do, we are at the point where we need to do the same thing for technology and cybersecurity. It's that simple. It's amazing fa- how fast move fast and break things, changes when it's a motorcycle helmet or, you know, a car's safety system.

You know, your seatbelt should not move fast and break things. But- And, and, and it's amazing that we've gotten to this point with technology. It, it's, it's a testament to the creativity and innovation that everybody did.

But you know what? At some, at, at, at certain points, it becomes enough of a societal risk that, yes, you know what, guys, and ladies, and others, let's calm down. Let's make sure that whatever we're doing here works where it's supposed to work.

Generally comes from one of two things. Either you have a massive disaster, or you have a community that's willing to push so hard that it forces the people who are participating to reevaluate things. And that's where I wanted to pivot to is, as you're listening to this episode on March 17th, thereabouts, we are just a week out from RSAC.

And I think RSAC has had a huge impact on the way that we do things because the community of cybersecurity professionals is a very loud voice. And when they come back and say, "Don't do this thing," or, "There needs to be some kind of a system in here to, to prevent this from happening," people listen because me... A-and this is the joke that I have on social media, right?

" Uh, as someone once told me, this is not an airport, you do not have to announce your departure. Um, but if you can convince 50 people to do that, suddenly that company has a problem on their hands because it's become a bigger issue. And that, to me, is the, the third leg of the stool, is the ability of a community to organize and say, "You know what?

" And that's one of the things we saw last week, is when Anthropic announced they were gonna pull out of this deal with, the Department of Defense, they actually crashed Anthropic's servers because everybody went to download the, the app from the App Store. Likewise, when OpenAI announced that they were gonna step in to fulfill the contract, I think someone said that it was noticeable the number of uninstalls of OpenAI off of systems to the point where Sam Altman actually got an email about it. And, and, and- So- Yes Memes and instructions about how to move from OpenAI to Anthropic.

Uh, pivoting a little bit to, to RSAC, right? Uh, it is a forum that, I mean, many of us have been going for many, many years. Uh, it, it undergoes its own changes, right?

But it is so important that, that we, that we be able to have these conversations there and, and, and, and elsewhere. So yes, I'm, I'm really looking forward to it. Sorry.

You're kinda thinking about RSAC. You know, you, Fernando, you and I have, you know, a, a calendar full of appointments, right? Of companies that we're meeting with.

Tom, you've got, the Tech Field Day, the Tech Field Day event that you're, you're doing there and your own list of appointmentsAs I'm kind of looking, I have two lists of RSAC whenever I go. This is what I think I'm gonna hear about, and this is what I wish I would hear about, right? Mm-hmm.

And sometimes they're the same thing. Sometimes they're sort of adjacent, and oftentimes they're a little far apart. But yeah, certainly I think we're not only gonna hear about AI this year, I think we're gonna hear a lot more about agentic AI, this moving into more autonomous type of work, 'cause the capabilities are there and people are starting, starting to use it.

It's not norm, it's not there, but I think security professionals want to think ahead. They don't like to be surprised and be... You know, they're out of the mode of let's, let's build the moat and the wall and prevent it from happening and try to keep it out of our environment.

They know that doesn't work. There may be a few places in the world where that, that can be possible. But I think that that's gonna be one theme.

And I think to that point, that's sort of on my what I expect to hear, what I would hope to hear. What I'd love to hear is, my own, my own belief is one of the models we have to change about security, that we're at this pivot point, is we have to get out of the scan and fix later mode. The, the idea we're gonna scan and have human, for whatever purpose, governance, security, incidents, et cetera, you name it, and non-security, whether it's DevOps, other things, the idea that we're gonna have humans in the loop reviewing everything is impossible.

It'll, it won't scale. It already doesn't scale today, and it won't scale with AI as velocity increases. So to me, I think the industry has to move to security vendors become fixers of problems, resolution of problems, not just identifiers of problems.

Doesn't mean they're gonna be, you know, the solution to everybody's problem. But if all you're gonna do is report more problems that humans have to intervene on, that's a failure mode. Now, you could argue, and I think this is a very viable approach, is yeah, you, you need to know about those things.

I think you need to find them closer to the point of origin, but even so, you still have to have other places where you look for issues, that that can be funneled into other agents that then are fixers that do, that resolve those issues. That's fine too. But when I see announcements by vendors of, "Yeah, we're now in, in the agent development process and reporting more issues," I kind of s- shake my head and like, "Yeah, that's not gonna go very well for you," at least I don't think in the long term.

So that, that's sort of my two top of the list things that I'm thinking are gonna be topics and would I'd like to be the topic. Yeah. I, I, I make the joke, right, I've, I've, I've said it sometimes I think Brazilian Portuguese, the letters A and I spell ai, and ai is the sound you make when something hurts.

Like if I'm gonna punch you in the shoulder- ... " Right? Perfectly tracks.

Like as a matter of fact, I think I used that joke in, in my report on the Futuron, my Futuron report for RSA Conference, RSAC Conference 2025. But, I... Yes, we'll absolutely see, agentic all over the place, right?

" Yeah, okay, fine. We, we, we can have that conversation, right? And I, I like your, your model of having things you want to hear versus things you think you'll hear, right?

I would like to... I would, I very much want to hear where agentic is working and not working and, and, and, I would like to hear, a lot more around, guardrails for reasoning, right? Everybody here is talking, we're deploying agentic everywhere.

We are, but the core of that, the, the agentic reasoning engine still seems to be LLMs, right? Still seems to be, "Look, here is a very capable large language model. Here is significant context around it.

We moved from prompt engineering to context engineering, and now we're moving, further along the stack. We think that this model reasons over something and, and duck," right? And, and I don't know, I'm, I'm, I'm, skeptical that that works outside of very controlled cases.

The question be- be, can we define the, the cases well enough where this works? Perhaps it is code fixing. Perhaps it is, it, it is SOC triage, right?

But beyond that, we'll see. But the other thing is that, I very much want to see, that, that, that, that I love your point about fixes. I think that there needs to be a broader conversation around things that are honestly outside the vendor's control.

If I take a firewall and I configure and, allow any, any rule, right? And that allow any, any rule is the source of a breach, with all due respect, that's not the vendor's fault, right? So, as much as we need secure by default, and oh, by God we do, right, we also need, this level of, of assurances and, and oversight over configurations as well, right?

If I choose to make somebody domain admin, is that really Microsoft's problem that somebody was a domain admin, or is it me who gave that permission to somebody because I didn't want to do something? We go back to regulations, right? We go back to insurance, right?

Who-- How are we going to hold anything people or how are we going to hold people accountable foruh, failures that result in cybersecurity related losses, right? I would love to have more of that conversation at RSAC. I'll leave you with two thoughts.

Um, if you're gonna be at RSAC, you should think now about what you feel like is the biggest problem, the biggest concern for you. And that's the thing you should ask everybody you talk to. How are you going to address this?

If you can't think of a good one though, here's another trick that you can use. This is a journalist, trick. Ask them what they think the biggest threat, biggest problem, biggest issue to solve is gonna be.

Yeah, their answer's probably gonna be interesting, but I'm more curious as to the reasoning behind it. If they think that the biggest problem to solve is, is this gonna make me a trillion dollars? That's one way to look at it.

If they think that the problem is, is that eventually it will become sentient and wipe out all of, life in the universe, that's a slightly different problem. A-and you should, you should kind of be interested in how both of those people approach that. And, and that should at least give you some fun topics of conversation.

Speaking of which, the two gentlemen who join me every week have some great stuff coming up, and I hope they can, tell you a little bit about it. Starting with you, Fernando, what have you got working on that people should, be checking out? Uh, so Tom, you and I are, are-- we're, we are moving our way through a, a report on SASE, Secure Access Service Edge, because as I like to say, there is things that-- in cybersecurity outside of AI, right?

Mm-hmm. So that, that should be coming up in the, in the near future, right? And be-- and, and alongside that, we are, helping-- We're, we're looking at what is the, the, the, the basis that organizations need for this level of trust, right?

And sorry to bring it back to AI, but it, but it comes down to is do you trust your infrastructure? Do you trust your identities? Do you trust the data that you have?

So, so those kinds of, of topics are, are top of mind for me right now as I, as I'm putting some research together. How about you, Mitch? Well, several things.

One, I'm gonna be issuing a new analyst insight report, about agent control plane. And, in this case, I'm, like I did with Observability Native, I'm issuing a framework for that just so we can kind of put the parts and pieces in the places where they are part of this discussion, because the thing that's elevated in our conversation around AI is accountability. It's not normally what we talk about in security.

Yes, yes, we do, but not in terms of-- It's almost equated to governance, right? Accountability through governance, through security, through guardrails, et cetera. Um, in, in, in the world, you know, to your example, Fernando, of, you know, is it the fault of Microsoft for allowing you to make someone a, a, root admin, or is it the fault of the person who did it?

We're now in a world where that accountability has to apply to agents also, like who did it and why. So what was the intent? What was the reasoning that led AI to do that?

Were there guardrails or policies in this code or whatever mechanisms we have to keep those actions within some, within some, allowed lanes of operation? And then of course, what happened, and do we know who that is? And so there's a lot of discussion around agent control plane, and it's interestingly enough, it's not just one thing.

Yes, Microsoft has and, and GitHub has, Agent HQ. Um, we have Agent Core from AWS. There are elements of this in a bunch of different products.

And just in the last couple of weeks, GitHub and Microsoft issued the new release of VS Code, Visual Studio Code, which is very-- the most popular IDE environment, where now some of those things, you could do them within the, within the, IDE, but now they're being automated, some of these control plane capabilities, the application of governance, the application of control as you're doing more and more agent work. So this is a big rising part of not only the operational capability of agents and agentic AI, but also the security, the governance, and operational capabilities. So I think it's gonna be a big part of our conversation.

I don't think it's g- elevated to an RSAC topic quite yet. Maybe in a year or two, that will be the hot buzzword. We'll see.

I can't wait for that. Uh, I also can't wait for RSAC. As, as Fernando and Mitch have mentioned, we're gonna be doing Tech Field Day there.

com, that are gonna be presenting. Uh, and I'm also gonna be running around talking to some interesting companies while I'm there. Uh, I'm probably also going to be running, so if you see me in the street, try to keep up.

Uh, but we wanna thank you very much for listening to this episode of Security Boulevard Podcast. If you enjoyed this conversation, do us a favor, subscribe on YouTube or in your favorite podcast application 'cause we don't want you to miss any of our episodes. And then do the thing where you click the thumbs up button, you give us a bunch of stars, and you leave a comment, positive comment, and that helps the show grow.

com and the Futurum Group. com, the, Techstrong TV website, or the Techstrong TV app. Uh, as previously mentioned, I am now trying to get it to run on my thermostat, which should be really interesting to see Mitch welcoming me to turn the temperature up.

Um, we will be back with more great stuff, and you make sure to follow us on Security Boulevard, on social media and on LinkedIn by looking for SecurityBLVD. Uh, lots more content there. We'll be back next week with another great episode.

Until then, we'll see you soon.