3 Mar 2026 · 55m · Rob T. Lee, David Bray, Michael Krigsman · 25 quoted facts
AI gives attackers a structural advantage over defenders. Attackers can automate and scale operations with no concern for collateral damage. Defenders must keep the business running, so they cannot safely hand decisions to untested machines. Rob T. Lee and David Bray argue that boards, security teams and hiring practices must change, and that defense should be judged on rapid detection and response, not on prevention.
The central tension is speed versus trust. Attackers can remove humans from the loop at little cost, while defenders need proof that autonomous tools will not break their own systems. The guests suggest ways to narrow the gap: federated learning to share threat intelligence, faster defensive development, and third-party testing of AI defenders. They expect fully autonomous defense only when people accept pilotless planes and robotic surgeons.
Lee says attackers shrug off an automated tool that trips over something, while a defender whose tool blocks the wrong system can take the business offline. That is why defenders keep humans in the loop and move more slowly.
Lee says AI plus SIM farms (banks of phone SIM cards used to run scams in bulk) lets criminals hit many people at about $300 each. Across 10,000 targets, that equals one big ransomware payout.
Bray says bad actors found package names that AI coding tools reliably invent, then published real malicious libraries under those names. Lee adds that an agent reading email can click every phishing link much faster than a person.
Lee says a security team that catches and stops an attack still gets blamed that an incident happened at all, unlike a military unit that intercepts an intruder. Bray says boards should treat rapid detection and mitigation as success.
Lee argues that AI knowledge goes stale before a new hire starts, so employers should look for people with a hacker's tinkering mindset. He points to hackathons that offer jobs and prizes as a way to find them.
Lee explains that attackers are willing to let automated tools run without human review, and they accept the damage if something goes wrong. Defenders cannot risk blocking legitimate systems, so they hold back and move more slowly.
Instead of one huge ransomware demand, criminals use AI and SIM farms (large banks of phone SIM cards) to scam many people for around $300 each. Lee says 10,000 victims at that amount equals one earlier ransomware payout.
In federated learning, the algorithm travels to each organization's data and learns there, so the data never leaves. Bray says this would let companies in an industry share what attacks they are seeing without revealing their own vulnerabilities or intellectual property.
Bray says there are indications that attackers noticed AI coding tools reliably invent certain code library names. They then wrote and published real open-source libraries under those names, so developers who use the AI's suggestion pull in the attacker's code.
Lee says it will happen when people are comfortable flying on planes with no pilot and being operated on by robot surgeons, and he thinks that is some years away. Bray notes that some US Navy ships already have a switch that lets their defenses run fully autonomously.
Stated by the speakers. Each quote matches the transcript, which shows what was said, not that it is true.
Rob T. Lee says criminals can use AI and SIM farms to scam many people for about $300 each, and that 10,000 such victims equal a single earlier ransomware payout.
“All you need to do is hit enough people at 300 a pop and do that at scale with SIM farms that are being found all across the US.”
Rob T. Lee says Anthropic detected its Claude Code product being used to automate offensive cyberattack capabilities inside organizations with minimal human involvement.
“Claude Code being utilized and detected by Anthropic, basically automating, uh, offensive capabilities in the attack inside organizations with minimal human, uh, in the loop”
Rob T. Lee says the Blitzkrieg of World War II shortened offensive campaign planning from weeks or months to days, and he compares this to the speed AI brings to cyberattacks.
“Things that used to be planned out that would take weeks, months for an offensive campaign were now shortened to days.”
David Bray says that about fifteen years before the episode he advocated a public-health-style approach to cybersecurity, in which organizations share what they are seeing without sharing specifics.
“I can remember about fifteen years ago, I was pushing for, um, almost a public health approach to cybersecurity”
David Bray describes federated learning as a method in which the algorithm travels to each organization's data and learns there, so the data is not shipped elsewhere.
“instead of having the data be shipped somewhere, the algorithm comes and learns there”
Rob T. Lee cites Salt Typhoon and Volt Typhoon as examples of attackers embedding themselves in critical infrastructure to cause havoc later.
“Or potentially embedding themselves to potentially cause havoc, uh, in the future critical infrastructure, for example, you know, salt and vol volt typhoon.”
Rob T. Lee argues that attackers will remove human oversight from automated tools faster than companies will, because attackers accept collateral damage that defenders cannot.
“Offensive teams are going to feel more comfortable removing the human in the loop faster than you as a company.”
David Bray says enterprises are not yet mature enough with digital twins to run fast simulations of the second- and third-order consequences of a security tool's actions.
“despite all this talk about digital twins or or whatever, we're really not that mature in terms of enterprises.”
David Bray suggests a sophisticated attacker could trick a company's security tools into taking its own systems offline, which would also demoralize the security team.
“if I was a really sophisticated hacker, I would make it so that you did it to yourself.”
David Bray says there are indications that bad actors found code libraries that AI tools reliably hallucinate and then wrote real open-source libraries under those names.
“some bad actors have identified when AI is being used for coding, for vibe coding, reliably hallucinates, and they have identified those reliable hallucinations and gone out and written a code library for the very hallucination”
Rob T. Lee says the gap between the release of Anthropic's Claude Code and its Cowork product was ten days.
“From Cloud Code to co-work being released was ten days.”
Rob T. Lee argues that cybersecurity culture treats any incident as a failure even when responders stop it, unlike militaries that reward interceptions.
“Doing incident response correctly still means you messed up.”
David Bray says chief information officers and chief information security officers originally reported to chief financial officers and later to chief operating officers, which framed cybersecurity as prevention.
“both CIOs and CISOs initially reported to chief financial officers, then later chief operating officers.”
David Bray says a smartphone gives its owner capabilities comparable to the CIA of the late 1970s and early 1980s, and about 2 billion people have one.
“You have the CIA in your pocket, circa the late nineteen seventies, early nineteen eighties, and 2 billion people on the planet have it.”
David Bray cites anarchists' use of TNT in the 1910s and 1920s as a past threat that society handled with defense in depth, not by locking down.
“In the nineteen tens, nineteen twenties, there was this thing called TNT. TNT was being used by anarchists to do bad things, but we didn't go into lockdown mode.”
Rob T. Lee says Chinese and North Korean threat actors have gained network access by getting hired as employees, not by hacking in.
“I go back to the Chinese and North Korean threat actors that instead of hacking a network, they just get hired as an employee.”
Rob T. Lee says a person had two or three internet-connected devices twenty years earlier, while a home now has thirty to fifty.
“a single individual would have two, three nodes, a phone, a computer, maybe a cable modem. Now you take a look at how many nodes are across my home, your home, you know, thirty, fifty nodes”
David Bray says an unnamed big tech company first tried replacing cybersecurity staff with AI, then found that humans looking for new exploits while AI handled known vulnerabilities outperformed.
“Unfortunately, they initially went down the path of replacing their cybersecurity people with AI, which I do not recommend, do not do, but they tried that.”
Rob T. Lee says some organizations now run hackathons that give the top competitor a job offer and a cash prize, as a way to identify strong talent.
“If you end up being the top competitor in this hackathon, not only are you offered a job, you win a cash prize.”
David Bray says safety filters applied after an AI computes are unlikely to be fully safe, and he prefers rules applied before computation.
“If something has a filter or a safety applied after compute, so post compute, I get very nervous about post compute agentic AI ever being 100% safe.”
David Bray says some US Navy vessels have a defense system that becomes fully autonomous when the captain flips a switch.
“we do have US. naval vessels where if the captain of the ship flips the switch, there is no human in the loop.”
David Bray says Waymo vehicles stopped when a power outage occurred, which he cites to show that interconnected AI systems can fail without any cyberattack.
“We saw when the power went out, they stopped, and that was not a cyber attack, it was just the power went out.”
David Bray says foreign jammers can block communications and geolocation signals within a three-kilometer radius.
“I've also seen where there are foreign jammers that can basically squelch any comms, any geolocation within a three kilometer radius.”
David Bray defines trust as the willingness to be vulnerable to the actions of an actor you cannot directly control, whether a human, organization or AI.
“I define trust as the willingness to be vulnerable to the actions of an actor you cannot directly control.”
David Bray recommends Ellen McCarthy's Trust and Media Cooperative, which aims to give people tools to assess information themselves or in communities.
“I will give a shout out to Ellen McCarthy's Trust and Media Cooperative, which is trying to give people the tools to do this themselves or in communities.”
Lee opens with the scam economics that AI makes possible. Many small $300 hits can match one large ransomware payout.
“hit enough people at 300 a pop and do that at scale”
Lee compares AI-speed attacks to the Blitzkrieg, arguing that a big jump in technology forces a change in tactics and strategy.
“essentially the blitzkrieg that happened in World War II”
Bray proposes a public-health-style way for companies to learn from each other's attacks without exposing data or weaknesses.
“instead of having the data be shipped somewhere, the algorithm comes and learns there”
Defenders cannot afford automation errors that attackers shrug off. This is the core reason attackers get a structural edge.
“Offensive teams are going to feel more comfortable removing the human in the loop”
Agents that read email or write code open new holes. Attackers exploit predictable hallucinations by publishing real libraries under invented names.
“gone out and written a code library for the very hallucination”
Lee says the real swarm is the speed of development. He suggests defenders can use the same speed to build and patch faster.
“From Cloud Code to co-work being released was ten days.”
Both guests argue that boards judge defenders as failures whenever an incident occurs. They want success redefined as rapid detection and response.
“Doing incident response correctly still means you messed up.”
Lee says to hire for learning speed over AI credentials. Managers set the culture by showing that learning AI is not a threat to staff.
“identify individuals that have a massive capacity to learn and learn quick”
Bray prefers safety rules applied before an AI computes over filters applied after. Both expect autonomous defense only after society accepts pilotless planes and robotic surgeons.
“I get very nervous about post compute agentic AI ever being 100% safe”
The death by a thousand cuts argument is actually true. You don't need to steal from ransomware attacks millions of dollars. All you need to do is hit enough people at 300 a pop and do that at scale with SIM farms that are being found all across the US. And all of a sudden, the 300 per person times 10,000 equals the same as a single ransomware payout before. This kind of scalability has not been able to be achieved before AI.
AI and autonomous agents are wreaking havoc with cybersecurity. Rob T. Lee is chief AI officer of the SANS Institute, and our co-host today is David Bray, chair of the accelerator at the Stimson Center. Rob, why should we care about the AI lifecycle and security? Like, why does this even matter?
You're, uh, tapping a massive problem when we are end up starting to talk about the speed and the velocity that's able to be achieved with, uh, autonomous agents. You look back just, uh, four or five months ago, and you know, Claude enough, and Claude Code being utilized and detected by Anthropic, basically automating, uh, offensive capabilities in the attack inside organizations with minimal human, uh, in the loop, uh, to be able to make sure that hallucinations are minimized.
Things that used to take months, uh, weeks are now being accomplished in minutes and seconds. And it's, and it's hard to wrap your head around about what that specifically means for potentially detection, being able to stop it and be able to see what's gonna happen, uh, even six months after the fact. What the adversaries and capabilities of even an individual possibility of accomplishing inside a network, uh, much less attacking individual for scams.
David, it sounds pretty scary.
It does. At the same time, this is, I think, every decade, something new shows up on the scene, and this is what, you know, the challenge of our decade is exactly as Rob said, is the speed and scale of which, uh, generative AI is able to, uh, allow an attacker, um, that might not be that sophisticated, or possibly what would take a long lead time. They can piece together an attack and do the reconnaissance and actually execute the attack.
Um, is unprecedented, um, but it means we need to adapt, and it means we need to, in some respects, uh, find strategies that allow us to id identify what we would call abnormal patterns of life happening in enterprises and adapting and responding faster. Um, so I look at it as a sort of a response of the attacker has evolved their strategies. Uh, it's time for the defenders to do the same.
Rob, how have AI and autonomous agents changed the nature of threats, changed the nature of attacks? And as David was talking, I saw you kind of nodding your head in agreement.
I'm a big history buff. Every single time there's a massive technological change, there is a massive change in tactics that need to follow follow it, and much less the strategy. Um, you could go back to the, you know, introduction of the airplane in World War I, uh, the tank, especially back then, you know, going back to the civil war, you know, the, uh, the line defense with Napoleonic, uh, what they ended up seeing then, all the way up to the aircraft carrier.
I basically move this mindset into warfare, into what's going on cyber defense, cyber offense as a whole. As a result of the speed and velocity, you end up taking a look at, you know, essentially the blitzkrieg that happened in World War II, the same kind of mentality has occurred. Things that used to be planned out that would take weeks, months for an offensive campaign were now shortened to days. This type of mentality really created shock and awe to individuals trying to defend against it.
That type of shock and awe is going to force a strategic and tactical redefinition in the way they're applying cyber defense currently. How fast can we ingest and leverage, uh, in threat intelligence, the tools, tactics and techniques that the adversaries are using, and be able to leverage the capabilities that we have in front of us? AI changes not only the offensive capabilities, but the defensive ones. The hard part, though, is being able to leverage the defensive ones in ways that are legal, able to reduce regulatory, you know, potential violations.
But if you imagine this, you know, for example, if multiple different organizations get hit simultaneously, if they're ingesting those automated detected TTPs into it and then spread it out worldwide, we'd be able to have a chance to counter what the offensive teams are doing at the same velocity that they're accomplishing it. But right now, like, even if you start, you know, saying, how would that be accomplished? We have a hard time just sharing threat intelligence, handing pieces of paper over or even calling each other on the phone, much less having a universal LLM that is disseminating TTPs worldwide.
The technology's there. The question is, can we even use it to leverage against it in a proper cyber cyber defense capability? That changes the tactics.
And I'll add one first plus one hundred, uh, to Rob for one pulling from history, cause I think that helps us address it. It actually helps you be a good stoic in the midst of whatever's happening. You can say themes and rhythms like this have happened before, and we will adapt. Uh, and I think he's spot on that actually AI, while right now is currently, um, super empowering attackers to his point, um, we could actually have an AI that again, if you look at what are the TTPs, what do the attack surface looks like, what are abnormal patterns, and having that shared.
I'd go even one further, which is I can remember about fifteen years ago, I was pushing for, um, almost a public health approach to cybersecurity, which is in public health. We don't share the specifics, but we share what we're seeing in the different states, because each of the fifty different states have their own privacy laws and things like that. And one of the biggest barriers, I think, for sharing, um, insights as to what's happening to one organization with many organizations is one they're worried about, you know, am I gonna reveal sort of my own vulnerabilities?
Am I gonna reveal my intellectual property? What might be interesting is there is there's an approach called federated learning, and federated learning, instead of having the data be shipped somewhere, the algorithm comes and learns there. Imagine actually that we had an algorithm that you were willing to let learn on what your organization is seeing, and as a result, you get the benefits without the specifics or the data as to what other organizations in your industry. So maybe you're all in finance, or maybe you're all in commerce, or maybe you're all in, uh, transportation.
You could actually say, here are the current attacks as of the last hour that we have been seeing across the world in terms of what might be attack surface, um, activities that are being exploited. And as a result, you do that without having to send your data to anybody, and you're collectively s more secure as a result.
Rob, thoughts on how incident response and forensics must evolve, especially since AI can navigate breached networks, erase its own tracks, corrupt evidence. So how are the rules have changed, or the technology has changed, so how how must incident response and forensics evolve and res- i i as a result?
There's so many different ways that we could potentially utilize the data that needs to be employed. Uh, you know, that capability of, you know, potentially be able to do sharing easier needs to be accomplished. And, you know, incident response and threat hunting are now pivoting specifically on what you know and how fast you can know it. So incident response and be able to identify attacks is crucial. So if you're looking for how they're pivoting across the network, how they're potentially doing, um, data, uh, media exploitation, essentially looking for, I always go back to the Death Star plans inside a network so they could get it out.
Or potentially embedding themselves to potentially cause havoc, uh, in the future critical infrastructure, for example, you know, salt and vol volt typhoon. The hard part here, and taking a step back and looking at response is somehow we need to achieve some sort of parity. Meaning that as the offensive teams are able to move at increasing velocity, instant response needs to be able to match that. Here's a couple of challenges with that. I believe currently in 2026, and this could be different even six months from now.
Most executives who are making a decisions whether they're going to RSA and seeing these new solutions that are out there, will have a hesitancy by deploying a solution that is going to remove the human in the loop. Offensive teams are going to feel more comfortable removing the human in the loop faster than you as a company. They'll say, I'm going to take on that risk because if an ends response or detection tool says, Hey, I found something, I'm going to block it automatically without a human analyzing that.
Well, who's what happens at that situation if bad things could occur? So are you placing the organization at risk more? Um, definitively, if you deploy these defensive tools that haven't been tested fully. I would be a little bit concerned right now not having a human saying, Yeah, we should block that, turn off that node in order to prevent that attack. Offensive teams are still able to say, well, if it trips over something, eh, not my fault.
The cyber defenders can't take that risk. So as a result, we're not going to be able to achieve the same velocity. And that alone is going to create some of the parity issues that even if we match capability to capability, AI versus AI, how do you reduce that risk? So we're able to match velocity versus velocity. I don't think we've solved that problem yet technically, because who's going to trust, Hey Rob, how do you know you're not tripping over something that's going to disconnect us from the internet and cause problems?
Uh, technically, we may be able to get there, but how do we know it's going to work effectively? Offensive teams, nah, if it trips over something, meh, and no big deal.
And to Rob's point, I think he he underscored a very essential risk calculus, which is one, the defenders want to keep the business running while also keeping it secure. Whereas the attackers are,
Meh, yeah.
like you said, you know, we we don't mind. And so because the attackers will be more likely to pull, um, solutions that don't have a human in the loop, uh, we've got to figure out what are ways that can help the defenders move at the same speed. And I think the biggest barrier we have is despite all this talk about digital twins or or whatever, we're really not that mature in terms of enterprises. So that you could actually run a really rapid parallel simulation that says, if my security tool opts to do this, what are not just the first order, but the second and third order consequences that could happen to my enterprise?
And as a result, I might be trying to secure it, but the next thing I know I've actually blocked this application and now my customers can't do business. Uh, the other thing I would also say, um, if I was a really sophisticated hacker, I would make it so that you did it to yourself. And as a result, you're now your security team is now dealing with the fact of not only have they actually caused the business to go offline, but now the company is upset at the security team.
They're now demoralized. This would be, um, hopefully, something that a attacker wouldn't do, but maybe I've seen things in the past where using demoralization is a tactic of certain threat actors.
You can ask your questions, and you should take advantage of this opportunity. When else will you have the chance to ask Rob Lee or David Bray pretty much whatever you want? So if you're watching on LinkedIn, pop your question into the chat. If you're watching on Twitter X, use the hashtag CXOTalk. And so on this point, Anthony Scrivenano, who's a very prominent data scientist, he's been a guest on CXOTalk a number of times. He makes the comment that one approach in attacks is for the bad guys to focus on a target that is already distracted in response mode.
This is exactly, um, what we have seen, not just in cyberspace, but we've also seen in the real world where, um, I get you distracted and focused responding to one activity. Maybe, you know, you've got something that looks like a hardware issue or software issue with your application. And the next thing, you know, while you're busy focusing over here, I'm coming in and then going laterally or something like that. So, um, to Anthony's point, um, what AI, generative AI in particular, now allows attackers to do is to actually plan something that's sophisticated, which did not used to be the domain of most attackers.
It was really the domain of nation state actors, and again, it took them months, uh, both to do the reconnaissance and the coordination. Um, now, again, what I'm hoping is, is that we can actually evolve almost a science where on the defender side, we can quickly say, I see these isolated events where they appear isolated, but they're actually part of a more coordinated campaign. And I know Anthony Scrivano has worked, uh, countering fraud and actually kept ta tackling fraudsters.
And usually the challenge is is oftentimes you're seeing the tip of the iceberg of what is a most a much more deeper campaign. And so I think this is exactly why one what SANS is doing is so important, and why we have to evolve what defenders do. Because, uh, now with both bots and agents, uh, you could have five or six fires going on at the same time. And the question is, while those five or six fires are happening, how do you know there's not a seventh one quietly actually exfiltrating data out of your organization?
Rob, thoughts on this kind of multifocal set of attacks, can't uh, campaign of attacks?
Distraction. It's just the fact that most teams are exhausted and they're looking at the same, you know, screen every single day. And as a result of that, you know, it leans into one of the reasons why defensive agents, you know, could work twenty four seven, they never get exhausted, will potentially be a game changer. The challenge, though, is we're only seeing offensive teams leverage this against a single target right now. What happens when you start having agent to agents be able to coordinate and communicate with each other simultaneously across critical infrastructure, hitting, be able to hit power and water simultaneously?
How do you have two teams at the same parity level on the defensive side to be able to prevent that? Um, the shock and awe that could occur there, where you could have down to the millisecond execution across multiple different nodes, financial networks, supply chain, critical infrastructure, electrical, and hit, boom, boom, boom. And uh, you imagine what happens on criminal side ransomware attacks that they wait until, you know, multiple parts of a supply chain. Uh, you think the, you know, some of the things that happened in the past with ransomware, imagine multiple different hospitals and insurance networks and pharmaceutical hit simultaneously of ransomware attacks.
These types of things are definitively possible, and the scale of offensive attacks go larger. When you go back to the distraction, the d, you know, exhaustion, the limited teams, the workforce that we can't, you know, hire a bunch of more people here. The leveraging AI is the only solution. And we come back to that, you know, how do we balance that with what we're seeing the offensive teams and how many potential attacks will need to occur before the risk calculus is so great that you're gonna have to say we have to potentially embed ourselves, utilizing some of these new technologies, remove some of the human in the loop and increase risk as a result of, you know, what can happen
if we don't. It's it. I could go on for minutes and hours about this topic, you know, over a beer, but at the same time, this is possible, and it likely is occurring right now, and it's a matter of time. Wh- which, you know, a geopolitical incident occurs, and they're gonna do a show me session of how vulnerable that we all are.
Let's talk about autonomous attacks, right? We have OpenClaw and agents are the new what, I don't know, the new refrigerators. Everybody's gotta have one, and we're gonna have thousands of them. This just makes the entire problem more complex, as, as Rob just said, right? They don't sleep. So who wants to jump in and let's let's talk about that.
the good thing about is open claw and Claude bot. The one thing you hear out of everyone's mouth initially is don't install it on your own system. And it shows people are thinking thoughtfully about turning on autonomous, just go wreak havoc with my own data. And it shows the risk calculus that will happen with security. You need admin access to do anything useful to shut down attacks, but we're already seeing open claw and Claude bot. People say, don't install it on your own system with root access because of the fact it could cause havoc.
These two things intersect and are not equal on the technological side.
Spot on. And I actually saw there was a story where, uh, someone was posting how supposedly they were in Ukraine, maybe they were, maybe they weren't, and they decided to install it on their system because it would make their home smart. Um, personally, I use whenever I see the word smart device, smart home, I I substitute the word hackable because that's really what it means when you say a smart device. Uh, and I don't know if I would do that in the middle of a active war zone, cause that's probably exploitable.
Um, so I do think to to Rob's point, one, there are cautions about be very cautious because these things will will possibly wreak havoc. The other challenge is, and this gets to the fundamental challenge of generative AI, which is generative AI is trained on past patterns. Um, and while you can ask it to possibly remix those patterns in interesting ways, it will always be sort of tied to its past training set. It's not necessarily re really good at novelty.
Um, in fact, when you try to present it with novelty, that's why you sometimes get something that's a pattern that doesn't actually exist, which they call a hallucination. It's just it's just not something it's seen before. And so this is actually more of a challenge on the defender side, because as we train these models to be defenders, we have those twenty four seven agents that are helping with defense. The challenge we've got is if a novel attacker really wants to exploit our attack surface, they will gain through various means insights as to what the agents are expecting to see or possibly may see, and therefore come with something completely different.
Which is why, as what Rob was saying earlier, we've almost got to have like an immune system response that says, I've never seen this type of attack before. I'm going to try and rapidly characterize it. I'm going to try and figure out what the intent is. And then I'm going to percolate the lessons not just across my enterprise, but ideally through federated learning, I'm going to share those insights with others in my industry. Because this is where right now the novelty of the offender, uh, attacking, uh, offense side, um, is actually puts defenders at a disadvantage precisely because generative AI is limited to its past training sets.
I have one thing to add to what you just said from a technical perspective is cause it, you know, I had a conversation just yesterday with one of my, um, James line, actually chatting about what we're talking about, our RSA keynotes. The attack surface with automated capabilities like cloud bot and open cloud, when you end up thinking about it, imagine if you're saying, Hey, go read my email. Essentially, you're basically setting yourself up for the ultimate spear phishing.
Is it'll go click on every single link that's coming in to be able to see what it is. Well, the human at that point is now the security defense because the human can't click as fast as the automated, uh, capability. Here, we've exploded the possibility of just the autonomous capability acting like a human, exposing additional attack surfaces. Because if you throw enough s balls in the air, one of those is gonna be hit by one of these autonomous agents.
And it, it's amazing when you end up thinking about just that alone, increasing speed of, you know, you being able to respond to many emails and marketing and tweets and LinkedIn messages. Imagine, basically, you just throw enough spearfis at that click capability, one of those is gonna land, and it's gonna be done at speed. You know, human actually ends up being just because they're slow, ends up being a security governor accidentally.
This points to the very challenge that while we need to use AI to help with our defenders that are already exhausted, that are already over capacity, AI itself is an attack surface itself. Um, we, we have indications, I wouldn't say confirmation, but indications. Uh, some bad actors have identified when AI is being used for coding, for vibe coding, reliably hallucinates, and they have identified those reliable hallucinations and gone out and written a code library for the very hallucination that the code library has made it open source.
What could be possibly wrong when you actually call a code library that didn't exist until a bad actor wrote it? So this gets to, um, not only do we need to use AI for defense, but as we're using it in an enterprise, it itself is an attack surface, whether we're using it for what we're doing in enterprise or we're using it to write code. If anything, um, cybersecurity professionals are going to be needed now more than ever.
However, we've got to figure out how we up our game and also give them the head space and and breathing room to deal with everything that's happening.
So just if you can now expand this aperture into this, Anthony makes the comment, swarms, AI swarms can also learn from how they fail and attack repeatedly in a more targeted way. So thoughts on AI swarms, and then we'll take some questions.
That gets to the challenge, which is I could inundate your enterprise if I know the capacity of the actual people to respond. And so I've seen swarms before, I've experienced swarms before, and so Anthony is spot on. And and like he said, given that they they can do permeations and they can try to be adaptive, they only have to be right once out of a thousand or 10,000 attempts that they're doing. This also points to, um, a lot of historical cybersecurity practices, and I know there's been a shift in in the last five years, but a lot of historical cybersecurity practices relied on, um, known CVs, critical vulnerabilities and exploits and the signature.
Well, the challenge is is if the swarm is changing its signature every second, um, that may not appear as a CV. That may not appear as something that a threat device is attacking. So swarms are definitely, uh, problematic, and it points to, and I and I have a sense this is where Anthony's going to the question. We cannot just rely on past trainings. What we need is instead almost a algorithm that says, I see what looks like non-normal, non-normalcy.
It's almost like a second derivative of it. Because what we're looking for is I don't know why I'm seeing this sort of abnormal pattern in my enterprise. It might be hardware or software, but it might be a sign of a swarm is trying to mess with me.
Rob, thoughts on swarms?
I feel we have not seen a specific or very many examples of swarms versus speed. And I flip that a little bit. And I, when I go back to swarms, the swarm I tend to think about is the swarm of developers that land on a new capability. And I, you know, take a look at what happened with Open Cloud and Cloud Bot. Is that how long ago was this someone's basement project? Uh, in, you know, weekend development, you know, co-work is really another good example.
From Cloud Code to co-work being released was ten days. When you end up having the development lifecycle go from something that used to take years down into weeks, that is the swarm capability I'm looking at. Because instead of just one or two developers, you have hundreds of developers potentially iterating on code, or even the code iterating on itself faster than we're able to potentially comprehend. Now, to take that from, hey, there's a positive side to this, is that even though you have capabilities that go from infancy to fully vetted capability within weeks, cyber defenders can leverage the same capability.
Because offensive teams tend to still be fairly tight held on their capability development, three, four developers. What if we flip that and potentially have hundreds, if not thousands, of cyber defenders start working on capabilities for defense? And able to take a cloud bot and make it into a very specific, you know, cruise missile that no one's ever seen before. Not just in something that took years to develop, but something that might take weeks and able to be deployed as a countermechanism inside the networks.
So what we're able to see on development speed, we could potentially use on a defensive mechanism for good.
This changes the notion of patching. Um, instead of patching being something that takes months, patching could be done in weeks or even a day.
Let's jump to to some questions. This is really my favorite part where we get to hear from the audience and and your questions, so keep them coming. All right, so we have a question from Arsalan Khan on Twitter X, who says, How how should businesses be held accountable, or should they for cybersecurity issues? So where does the responsibility and the accountability lie?
Doing incident response correctly still means you messed up. And what I mean by that is the fact that the incident occurred in the first place means that something went wrong. And even though the incident response team may have interdicted it, may have stopped the Death Star plans from being stolen, you don't get medals as an incident responder by saying, Hey, good job from stopping that. In the military, you do, even if an incursion happens, if you know North Korea flies jets into South Korea, even though they encroached in the airspace, if we're able to launch missiles or something in there to interdict and stop them, everyone gets a medal.
That does not happen in cybersecurity. The accountability and what our mentality, our culture is still has not changed to reward proper cyber defense. So accountability incentives are perverse as a result of this. If you're a CISO and you do a good job, you detect the incident, the incident happened, you failed. That's the perverse incentives that occur today, and it's not going to change because AI is out there. Where's the medals for the incident responders, the SOC teams that detect the incident per in the middle of the attack and then says, Hey guys, we stopped it.
They did not achieve their operational goals. They're still looked at from a, you know, liability, risk, and all these other calculus, you still failed.
If I could do a hallelujah to what Rob just said, because this is the challenge is we still have boards and we still have CEOs that think this is a preventative exercise. When the reality is if you're going to be on the internet, it is going to happen to you. What you want is exactly as Rob said, you want a team that responds to it rapidly, that actually tries to do their best to actually counter it.
And actually success is simply one, we detected it, and two, that we responded to it and we actually did our best to mitigate it. But that is that is a mind shift take. And I hope, I mean, I I will do anything, by the way, Rob, to help SANS get that message out there. Because unless they want to unplug themselves from the internet, bury whatever they're doing underground and make sure no human being touches it, these things will happen.
And they will happen with increasing frequency, partly because of excited geopolitics, um, but also because of the capabilities that are out there. And if you're going to use AI, what I am trying to do is when I go to boards and I go to CEOs, I say, like, if you're going to be adopting AI for your enterprise to engage your customers or improve your internal operations, you should in parallel upgrade your cybersecurity posture in parallel. Because that will actually make it that you're more responsible in your use of AI, while at the same time, hopefully making it so that you can detect if bad bad actions are being sent your way, because they will be sent your way.
Let's jump to another question. And and by the way, I have to say, hearing you both talk, I'm ready to hide. I'm ready to, you know, do the ostrich thing. I'll stay home, you know, I don't know, maybe put mattresses up against the windows.
I I know someone that can help you get a Faraday cage if you want a Faraday cage.
Yeah, but that's it. But you know, just like as a point you just made there, as a parent, do you just, like, pull your kids in at home and say, you never venture outside, you're gonna wear wear, helmets, your bicycle? There's a certain amount of continual risk, no matter what you do, that you have to assume you can't prevent any bicycle accident from happening. You just set the rules of the road, you say, look both ways, don't run with scissors.
There's certain things that are acceptable risk, and unless you have that acceptable risk, there's no learning, there's no growth, there's no acceleration. So, you know, as a parent, you look at it that it's like, Oh, I can't prevent all bad things. But again, no ostrich head in the sand. It's a hey, listen, we need to lean forward, especially in today's time, business transformation, everyone out there is trying to figure out how do I deal with AI?
You just need enough of a parachute, enough of a rope around your waist to say I could take a certain amount of risk safely, but not without ever potentially falling it, uh, with zero risk in the chasm.
But it used to be that the risk came from getting hit by a car. Those risks risks remain, only now we have this very large, ongoing, permanent additional attack surface because we live our lives online.
Correct. And also because the good news is, so I tell people, if you have a smartphone, you have the capabilities to call anybody in a moment's notice, hopefully with their permission, track them or any asset using AirTags or other means anywhere in the world, again, with hopefully their permission, and download satellite footage as recent as fifteen minutes ago at point two five meter resolution. You have the CIA in your pocket, circa the late nineteen seventies, early nineteen eighties, and 2 billion people on the planet have it.
Um, so the the good news is we've super empowered you to do things that are unprecedented. That also means that bad actors can do things that are unprecedented. I don't think we want to roll that back. It just means again, we've got to figure out defense in depth, and we've been here before. In the nineteen tens, nineteen twenties, there was this thing called TNT. TNT was being used by anarchists to do bad things, but we didn't go into lockdown mode.
We figured out defense in depth, and now it's less of a threat, and, you know, we continue to live our lives. So I think we can figure it out, but it's going to take new approaches. And I think part of the trouble with cybersecurity is both CIOs and CISOs initially reported to chief financial officers, then later chief operating officers. And so it was seen exactly as Rob said, which was cybersecurity was a preventative function, as opposed to it is it is what you do as you're interacting with the world and doing commerce and doing whatever you do.
And so what you want to do is have a team that does get the medals for rapidly detecting, rapidly containing, and rapidly mitigating what will happen, because that's the nature of living in the world we're in.
we have massive SIM farms that are out there. They imagine that instead of something that someone's hitting your kid, or potentially you, your parents or your neighbors, that kind of scalability is now achievable with AI. So the death by a thousand cuts argument is actually true. You don't need to steal from ransomware attacks, you know, you know, millions of dollars. All you need to do is hit enough people at 300 a pop and do that at scale with SIM farms that are being found all across the US.
And all of a sudden, the 300 per person times 10,000 equals the same as a single ransomware payout before. This kind of scalability has not been able to be achieved before AI.
Let's go back to some additional questions. And I'm still I'm not convinced that the ostrich approach I th- I still am thinking maybe I need to just simply hide and that's it. Okay, uh, but maybe by the time we're done with this conversation, I'll feel more confident that I can actually go out in the world. All right, this is from Chris Peterson on Twitter X. Oh, and I should also remember, uh, remind you guys that you should subscribe to the CXOTalk newsletter.
Go to CXOTalk dot com now, subscribe, and we will send you notices of upcoming shows. This is from Chris Peterson, and he says, Feeding AI tools good data often requires complex tool chains. How do we keep those tools and chains from becoming the attacker's prize targets?
You can't. It's the same thing as, you know, UPS delivery walking straight in the door and you're, you know, and UPS, you know, it's the same type of attack surface as you can't eliminate. The more you end up having a larger attack surface. In this case, you take a look at what enterprises now leveraging across, you know, simple Zoom sessions, what we're doing here from teams and so forth. The likelihood of someone being able to penetrate that is becoming more, um, prescient.
I go back to the Chinese and North Korean threat actors that instead of hacking a network, they just get hired as an employee. That's using your own tools and HR and humans and plus technicals, like there was a single laptop that has, you know, the keys to every access point in your network already set. This is going to take individuals to realize that even attack tools, the common management capabilities in our organization are able to be exploited.
And the attack surface for a single human, which is really interesting. Again, I was having a separate conversation last night just about this. What ten, twenty years ago, or actually twenty years ago, a single individual would have two, three nodes, a phone, a computer, maybe a cable modem. Now you take a look at how many nodes are across my home, your home, you know, thirty, fifty nodes, you know, with every single appliance and potential TV. The enterprise network size for a single individual happened twenty years ago is about the single enterprise size that a single individual has today.
It's kind of mind numbing when you end up saying, count up how many single devices that you individually have that are connected to the internet. Every single one of those is a potential attack surface.
Here is a question. Reza Fatahi says there are good opportunities, I'm assuming, for attacks and defense at the application and OS layers. Each angle is a different application. Each layer is a different application. Uh, any thoughts on that? Either of you want to jump into that?
You're going to have to find ways to rapidly do independent verification and validation of what you're using as your tools, whether they're for data ingest or applications or things like that. And it has to be faster, uh, than what's currently been done. You can't just rely on published critical vulnerabilities. You really need for the things that are absolutely essential. You almost have to be constantly red teaming, um, how an application or an OS might be used against you.
And the good news is, again, AI I've seen actually instances, I won't name the name of the big tech company, but they did two parallel paths for some of their cybersecurity functions. Unfortunately, they initially went down the path of replacing their cybersecurity people with AI, which I do not recommend, do not do, but they tried that. And then in another function for their mobile security, they had the AI focus on known critical vulnerabilities to free up their people, their cybersecurity professionals to look for new exploits.
Well, guess which team outperformed? It was the humans plus the the the AI that outperformed. Because you have to always be looking for new vulnerabilities, you cannot just rely on published known vulnerabilities.
This is from Huwei Wang on LinkedIn, and she says, As a defender and as an attacker for technology readiness, do you see the direction of talent development moving towards what previously was called the ten x engine or a full stack engineer? So the idea of of talent and preparing talent to handle these issues.
I go leadership by vulnerability or vulnerability by leadership, depending on how you want to frame it, is that we're all learning. I'm learning. I I had very little knowledge about AI two years ago, and I'm struggling on a daily basis. And this is where, from a talent management perspective, there's this idea that we need to hire people with AI experience. That's really hard to assess right now because when you h by the time you hire them, their knowledge may be already out of date.
So from a talent management perspective, one of the things I think is more crucial than ever is trying to identify individuals that have a massive capacity to learn and learn quick. Goes back into the old hacking mindset, and what before was considered criminal. The original mindset was I wonder if I put this thing and over here, do this thing, tie the wires together and take a step back. That is a fiddling mindset, and that's what hacking was considered, even going back to the MIT days where they invented the term.
That type of talent is hard to come by, and it's also hard to assess. What I'm seeing organizations do, and it's a complete disconnect from, and I I credit, uh, Kate Marshall, I is on her Substack mentioned this. Specifically, you start seeing organizations using hackathons. Um, we e- ended up seeing, um, I think it was Nvidia or Anthropic, just do one. If you end up being the top competitor in this hackathon, not only are you offered a job, you win a cash prize.
It is going to assess someone's ability to work through problems, through the problem solving challenges, to assess their mentality, to be able to show you're the right candidate. That is the only way organizations are going to be able to hire the right people with the massive technological change and the velocity that it's occurring at.
David, this one seems tailor made for you. And this is from Simone Joan Moore. And earlier, you spoke about immune system response. She says, What about the systems that are off, they're compromised, or their DNA hasn't developed the signal or detection system for it? What do we do?
You have to instrument your enterprise to have what's called patterns of life. What does normalcy look like for these systems, um, both in terms of their operations, but also the human use. And so then if a persona that looks like your CFO or your CEO, um, you know, at nine p.m. on a Friday night suddenly seems to be accessing a finance system that is non-normal, that might be, and again, you have to. This gets to Rob's point, which is we have to have ways of defining what is reasonable risk calculus.
You may want to stop that transaction until the chief information security officer actually has a conversation and says, Was that really you, uh, CEO or CFO, or was that some bad actor trying to exploit your persona to actually pull money out? Uh, and so, uh. But that requires enterprises to one be, um, using their logs and that instrumentation in a way to establish patterns of life, but there is precedent for this. Um, the other thing I would say real quick, um, and I'm and and we haven't gone here, is that good.
The good news is generative AI is really good at passing the Turing test. The bad news is generative AI is really good at passing the Turing test. I get concerned about audio, video, synthetic media being used to convince your humans to do a legitimate activity where nothing was really breached. It was just AI itself was used as a fraud to get social engineering on steroids, um, to have things either money walk out the door, data walk out the door, et cetera.
So generative AI itself is going to require us to have better ways of verifying that those workers and those people you're talking to really are legitimate.
This is from Demetrius Georgiou, who says, Looking into the future, when do you expect that AI agentic defenders will be able to be trusted as an experienced human without a human in the loop? And please answer for both IT and OT environments. Rob, you want to jump on that one?
When you end up looking at the standard certification, accreditation and process of human beings, you go to your college degree, you get your certifications, there's, you know, assessments that someone end up going through in order. But again, that's the baseline. You could say, hey, have the minimal qualifications to be able to do this job, but the trust is there's still some third party that is doing this assessment. Where I think the industry, and again, for anyone going to RSA, uh, in a few weeks, walk around the floor and everyone's gonna have this massive solutions, just like you're saying, agentic defenders.
Ask, how do you? How do they establish trust in that capability? Like, well, we, we have our own rubric, we went through our own battery exam, we here's our, here's what we did. Is I could go out there and say, Hey, my IQ is X. Well, how do we know that I'm a self-assessing my IQ versus having a third party, independent capability, establish that trust? And the reason I use trust is all these different accreditation, certification, degree programs, assessments are all aimed to establish trust.
Trust in the agentic defenders needs to have some sort of trust mechanism assessed by a third party, not software, accreditation, all these other things. Does it handle reasoning equivalent to a third party human being?
This is from Dan Toori, and he says, Generally, is industry accepting of integrated logging tools, or is there a need for external immutable logging? It seems like we have rolled out tools without pause to validation and integrity.
It depends on certain enterprises are more instrumented than others, the finance sector, they have to be, because they're regulated. Um, I think what he's asking, though, is, is it sufficient to have a log where you don't have some immutability to it or not? And I would say, yes, you want to have immutability to the log, because as we know, some of these attackers, one of the things they'll do is they'll mess with the log or even destroy the log.
Um, and so the short answer is yes on immutability. Uh, I, I do want to actually add to this question and and and tie it to what Rob also was saying, though. I I think we're going to find, especially for agentic AI, and I realize there are several companies plowing ahead with it. If something has a filter or a safety applied after compute, so post compute, I get very nervous about post compute agentic AI ever being 100% safe.
And again, you can figure out contexts where it's acceptable to have that risk. I want to have more and more AI agents where I can have filters, boundaries, rules applied pre-compute. Because I think the moment you let the machine compute something, just the nature of hacker, an attacker is they will get it to do something you don't want it to do. So I think what I would encourage increasingly is look for people that are pushing solutions that are pre-compute versus post-compute.
That's a nuance that, unless you're a computer scientist, it's not something people ask for. But I think this is the geek in me just saying pre-compute is where we need to put more attention to.
We have a question from Christopher Jablonski and Rob, I think this one is good for you. It's perfect for you. I'll only ask you, I'll read the question in a second, but Rob to start ans- answering relatively quickly, cause we're just going to run out of time. And this is from Christopher Jablonski. He says, going back to how offensive teams seem to seem to have a permanent competitive advantage as they are more comfortable with removing a human in the loop.
By how much does this issue and other such asymmetries delay autonomous security from reaching reality?
I think that's a reality, f- uh, for the near term, and I actually think, to a certain extent, it is necessary. Because I think from a cyber defender's perspective and development in the industry, we've become a little bit assumption that we have some sort of parity that we're able to, you know, block and tackle what the offensive teams have been able to do for the past five to six years. Now, this is going to create some leverage, uh, in the industry, and and hopefully a lot of innovation that'll be able to match it.
We're assuming that we're going to be able to leverage AI to AI. I go back into what the offensive teams aren't going to be able to do is create AI ecosystems that are going to be able to have defenders talk to one another. You even just take a look at Cloud Bot and some of the capabilities it's already been able to le- leverage, uh, AI judicial systems for different agents, what it's doing, ability for AI agents to ask humans to do certain tasks.
Um, and again, that physical level task, you could have an AI agent who's your manager tell David, you know, Hey, I need you to go and over to this system and unplug it from the wall. You know, even though that seems, you know, rash, but again, there's ransomware cases that that is literally what the response mechanism is to make sure that system is not infected. There are ways for innovation that are unexplored on the cyber defender side that this disparity, this asymmetry that's occurring, is going to force.
And I'm excited to see how this develops.
Michael M throws a monkey wrench into all of this. Okay, Michael M says, Sure, all of this is true, but once you remove the human in the loop, AI errors are a greater problem, and the bad guys simply have to hack the AI defender and convert it into a dual agent.
yeah, so that is spot on. I would say again, it depend we we haven't dived into the different types of AI. There are AI methods that are more rigorous than, say, generative AI. Um, but I will give the analogy, and I think Rob's right that in the short term, the advantage is on the offensive side, defensive has to up this game. But we do have US. naval vessels where if the captain of the ship flips the switch, there is no human in the loop.
If something's coming at that ship and it's threatening the ship, the the system will actually be fully autonomous in defending the ship. Now he obviously turns off that switch when he's not in a danger zone, he's not doing it in home port. But that's that's that is a autonomous an AI of a different name. It's not generative AI that does successfully defend the ship. So I think what we need to get to is through a combination of more rigorous AI that's less brittle than generative AI, but also formal methods and other means like that.
We can get something like that. Where if you decide you're now under siege, you want to flip the switch, then the autonomous defenses can go on. But right now, as we've already observed, we don't have enough knowledge of the second and third and fourth order effects in an enterprise that it wouldn't be used against us.
I believe on the defensive ti- side, human in the loop is going to be there. And when it's not, there are two things that are simultaneously going to happen. You're going to be able to board a plane, look up in the cockpit, and there's not going to be a pilot there. If you're comfortable with that, you're comfortable with cyber defense having no human in the loop. Second side to that, you go into the doctor, it says, Hey, you need a surgery, and you have the ro- a, you know, humanoid robot come in with a scalpel and ready to sedate you.
Are you totally comfortable with that? When these two things happen near the same time, I believe human in the loop will be removed from cyber defense. We have some years to be able to get there.
Okay, I go back to what I said earlier, I'm going back to hiding once again, because that seems like the most prudent response to all of this. All right,
Well, have you gotten into Waymo yet without a driver? I'm still worried about that.
you know, I live, I live in Boston, and no, I have never used a Waymo without a driver.
Well, and I think even real quick to the on the Waymo example, let's say Waymo's got everything working in the AI. We saw when the power went out, they stopped, and that was not a cyber attack, it was just the power went out. I've also seen where there are foreign jammers that can basically squelch any comms, any geolocation within a three kilometer radius. So, yeah, the reality is the world is interconnected, but this is why it's important to have folks on the defender side.
Let's, uh, very quickly go to another question. This is from Hater Souza, who says, Okay, here's another monkey wrench. Uh, one of the current AI challenges is that there is no way to distinguish when AI provides the correct information or hallucinates. How can we boost the trust of the information provided? And and also what role governments or entities should take? And let me split this up. And Rob, maybe you can talk about the issue of boosting trust in the inter information from coming from AI systems, and David, very quickly, you can talk about the role of governments and official entities in this.
From a trust perspective, how do you trust it? Not going to do, you know, hallucinations or anything else. Do you trust a human? I mean, I, you know, you employ someone. How do you know they're going to be 100% perfect in their answers? How do you know they're not going to read something and interpret it incorrectly? When you end up looking at prompt injection, it's no different than information warfare campaigns, uh, from, you know, Russia, saying, Hey, you know, we have this massive, uh, groundswell of, you know, feeling on this one topic.
And then everyone believes that, and that becomes the truth. How do you prevent humans from doing it? Is the same answer, how do you prevent agents from doing it? We have to start looking at these technologies as reasoning. They're able to be s, you know, uh, convinced of bad data. They're able to potentially interpret things incorrectly. If you're not able to trust humans, and we have them employed everywhere, including in critical infrastructure, flying planes, going back to that example, and even bad data coming in there has resulted in bad things happening.
There's very little trust that you're going to be able to ever assume in the system that every single if then loop, one plus one always equals two. With AI and agents, one plus one is going to equal three in many cases, just like it happens for humans, mistakes will happen, and we're not going to be able to get around it.
If I can just say spot on to what, um, Rob said. I I often say I define trust as the willingness to be vulnerable to the actions of an actor you cannot directly control. That could be a human, that could be an organization, that could be an AI. And what it shows is if you perceive competence, perceived benevolence, and perceived integrity, you're willing to trust. Well, the reality is, as Rob was saying, uh, foreign actors can manipulate those perceptions so that you either do trust something you shouldn't or vice versa.
Um, but these are true for humans as well, so we should be approaching it the same way. We want to hold organizations and humans, recognizing that they will make mistakes, but accountable if they make a mistake that actually results in harm or loss of life. I don't think you can have government regulate what is good information. That is a slippery slope to really bad things happening. Uh, I will give a shout out to Ellen McCarthy's Trust and Media Cooperative, which is trying to give people the tools to do this themselves or in communities.
And then the last thing I will say, because I'll throw another monkey wrench into this. I have seen already that generative AI is really good at building psychological profiles of each of us, thanks to our digital exhaust. And figuring out if I present this information to David or Michael at this right time, I'm going to possibly get them to take a following action. That's not a cybersecurity risk, that's just a collective free society risk, which is we need to recognize that if we get our emotions up, if we get angry or we get fearful.
Is that a case where maybe a AI agent is trying to deliver information to mess with our emotions and get us to do a behavior? And that's where I tell people always take a breath if all of a sudden you feel like your adrenaline just went skyrocketing.
Rob, this is from David Batts. What factors should be evaluated to determine the need for a third party AI risk assessment? Really quickly, please.
Almost 100%. W- When are you not going to be able? Or are you going to hire someone with no assessment of their skills, capabilities, knowledge? It's it's going to be 100% across the board. We're going to need to figure out some sort of trust assessment capabilities for AI. Um, right now it's just chatbot. We kind of think of it as Google. Once we start letting autonomous decision making occur in your home, yeah, this is where you're definitely going to want to say who assesses that thing to be able to make a decision, uh, even on, like, lower threat, uh, things that are occurring in the household.
Um, it's it's going to come down to there's going to, it's going to need to exist somehow. You're going to want to know, how does this thing know its skills, know its actions, and is it going to reason properly?
Reza Fatahi comes back with a very quick question. He says, Uh, what are acceptable lase- latencies for a, uh, defense infrastructure? Uh, for decision or deliberative process, and for key decisions or deliberative deliberative process in terms of a APIs. Any do either of you happen to know that?
There's no magic number, but I think what it's showing, and is is what Rob has been emphasizing, is we do need to up our game on the defender side. And I think it's about identifying one, is this something suspicious? But I don't think we're under siege. Are we now under siege? And then do we need to increase our op tempo? I don't think you can necessarily give a a specific latency number for the enterprise, but it does say you should be able to rapidly respond and mitigate what you perceive to be attacks at a much faster rate than you have.
And that's going to require new playbooks, which is why I'm excited about the innovations that will happen on the defender side in the years ahead.
Rob, what does a security culture look like in practice, especially given AI and autonomous systems and all the things that we've discussed?
I don't think it's any different from any other career field that's out there. Everyone is running around with a lot of angst, staring at AI. Is this going to replace my job? What new skills are am I going to need to be able to maintain in the future? The security culture is led by the managers. If the managers are not, you know, r- really emphasizing that AI is going to have a positive impact on you, your career and the business, then that's where the skill security cultural cultural vulnerability actually exists.
Has to be led. They have to be deliberate, and they have to show that you're not at risk from learning AI and looking at AIs to increase efficiencies in your organization. Otherwise, you're putting the organization at risk based off the offensive capabilities that exist today.
David, it looks like you're going to get the last word here. Organizations are building critical systems on data they can't fully verify and very often don't fully understand. What should boards of directors do, and should boards treat data provenance as a fiduciary responsibility? David, and very quickly, please.
I think, though, I want to again double click on what Rob said, which is boards, most importantly, even before they tackle data provenance, should be saying, what type of culture are we creating for our cybersecurity professionals? If it's a shame and blame, if it's like, how dare this ever happened, then I guarantee you your company is not going to exist in three or four years, probably. If, however, it's the idea that this is, this is part of doing business.
We want folks that are learning, we want folks that are adapting, that include data provenance. But also, how many people verify that the hardware chips really are what the hardware chips they are? That's a geopolitical reality, too. That's not something that the board has to get into the sort of the knickers of cybersecurity professionals to do. But they should be saying, what are we doing to learn and adapt and actually have the ethos that we have a culture that says it's about rapid detection, rapid mitigation of data issues, hardware issues, or cybersecurity issues.
All right, and with that, we're out of time. I mean, this conversation could go on, and I hope you will both come back and join CXOTalk again so we can dive more deeply into these issues, cause we barely scratched the surface today.
Oh, I want to do it with Rob over a beer, so maybe next time in person.
But we're going to put the ostrich ourselves first and then have the beer on the ground. Then we'll be, you know, a lot more comfortable.
Well, that sounds good. A huge thank you to Rob T. Lee. He is chief AI officer of the SANS Institute, and to David Bray, who is chair of the accelerator at the Stimson Center. Gentlemen, thank you so much, and a huge thank you to everybody who watched before you go. Again, subscribe to the CXOTalk newsletter. We'll send you updates. You are now part of our community and we want you. Join us and we have great shows, amazing shows actually, and we'll see you again next time.
Take care everybody. Bye-bye.