Uh, Rob, can you give us a little bit of the backstory as to how did this begin and how long has it been running?

Well, officially, it's been running for about two decades.

Our founder, Alan Paller, uh, started this in the mid two thousands, uh, with an effort to, you know, essentially it was a small luncheon at the time, to help identify and kind of reward and acknowledge those individuals working tirelessly throughout the industry.

He wanted a way to recognize those individuals as a thank you.

And it was a lot more intimate back then.

You know, cybersecurity was not as, you know, the, the industry that it is now and, you know, kind of was a small, you know, thank you from Alan to these individuals, and it's kind of grown since that point.

After COVID, uh, SANS decided to...

Can you touch on that element?

Is AI gonna kill us all, really?

Every massive technological change has come with its share of, "This is going to change everything, and it's not gonna be good." Um, my parents lived during a time in which the Doomsday Clock was invented, and they were trained to duck and cover under, you know, their chairs and, you know, tables, thinking that they'll somehow survive a nuclear blast.

A, a nuclear war was going to happen, um, if you listen to anyone, you know, who grew up during that timeframe in the fifties and even into the sixties.

And to a certain extent, you know, we obviously got through it, but that doesn't mean that, you know, hey, we navigated one doesn't necessarily beget the other.

I think with any technology, and this is one of the reasons why it's really important to lean on those who know how to do certain, you know, go back to the, um, Liam Neeson quote, you know, like you have a certain set of skills.

Cybersecurity knows how to contain things that are really bad, and we've been doing it for twenty-five years.

And cybersecurity can help, if not guarantee, to contain AI.

How have things evolved in your multifaceted career, and what's the overlap like, if any, as you mentioned, between these new AI and cyber applications?

Could you break that down in terms of, um, what you think, uh, are we maturing in terms of AI safety, or, uh, are we still learning some lessons here?

So there's a little bit of, you know, historical context, you know, historical, you know, six weeks old history, and then future that you hit upon.

And I'll say the his- historical first, and then I'll pause, and then, you know, I'll, you know, then take a moment to then cover, like, where are we headed.

So when we end up taking a look at the incident, um, you know, a lot of people are still calling it the hugging face attack, the OpenAI attack, and, you know, the intent...

You know, I always go back to, you know, this law enforcement background, act and intent.

You know, there was no intent to cause malicious harm, but there was an act.

It's almost, you know, Ed, we talked about in the previous webcast, it's equivalent to, you know, like a manslaughter incident in which there's accidental death, you know.

Um, how do you keep up with all this stuff, Rob? This stuff is just moving so fast

Could you break that down in terms of, um, what you think, uh, are we maturing in terms of AI safety, or, uh, are we still learning some lessons here?

So there's a little bit of, you know, historical context, you know, historical, you know, six weeks old history and then future that you hit upon.

And I'll say the his- historical first, and then I'll pause, and then, you know, I'll, you know, then take a moment to then cover, like, where are we headed.

So when we end up taking a look at the incident, um, you know, a lot of people are still calling it the hugging face attack, the OpenAI attack, and, you know, the intent...

You know, I always go back to, you know, this law enforcement background, act and intent.

You know, there was no intent to cause malicious harm, but there was an act.

It's almost, you know, Ed, we talked about in the previous webcast, it's equivalent to, you know, like a manslaughter incident in which there's accidental death, you know.

Um, how do you keep up with all this stuff, Rob? The stuff is just moving so fast

what is,

you know, ICMP echo request? You know, you're playing.

And I think a lot of security teams, because of the technological umbrella that's like, we're supposed to know everything, you should immediately be able to deploy agents.

I think you need to go back to Okay, can you get an agent to read your email and, you know, automatically put in calendar entries? Let's start with some fundamentals.

And I think people feel that it may be very boring, you know, for some who are just like, I pick things up naturally.

I think majority teams out there, you need to first look at how do you apply agents in your own personal lives, in your initial work structure.

And then you need to start taking a look at how do you employ agents across the team interacting with one another for doing very basic, menial tasks.

kind of as a new program, what does that function actually look like for a mid-sized company? I don't

you know,

inebriated or something like that, totally different thing.

So I think this is the digital equivalent of that because unlike Google, unlike, you know, I'm trying to say at 30,000, you know, really explain why this is a big deal.

Unlike Google, unlike other technology sets, you're not really able to cause widespread harm.

With the latest frontier models, and I actually had a conversation with several Hill staffers about this specifically saying, are you tracking this? And they wasn't on their radar that much because it sounds cybersecurity and like, oh, it's technical.

Someone breaks into something, who cares? And I said, well, let's take an extension of that, biosecurity.

And they said, well, what do you mean? I said, the same models.

Are we seeing that as part of this assessment strategy and how the NSA and some other agencies are going to be building benchmarking process to assess these most advanced models that are being provided into this assessment?

you know,

inebriated or something like that, totally different thing.

So I think this is the digital equivalent of that because unlike Google, unlike, you know, I'm trying to say at 30,000, you know, really explain why this is a big deal.

unlike Google, unlike other technology sets, you're not really able to cause widespread harm.

With the latest frontier models, and I actually had a conversation with several Hill staffers about this specifically saying, are you tracking this? And they wasn't on their radar that much because it sounds cybersecurity and like, oh, it's technical.

Someone breaks into something, who cares? And I said, well, let's take an extension of that, biosecurity.

And they said, well, what do you mean? I said, the same models.

Are we seeing that as part of this assessment strategy and how the NSA and some other agencies are going to be building benchmarking process to assess these most advanced models that are being provided into this assessment?

In addition to being fantastical and non-realistic, Rob was struck by the black boxiness of it.

It requires an individual to understand the underlying data, add up all those equations, and say, "Here's how we made that determination." Those equations don't exist here.

If you're expecting a courtroom to use, "Hey, you got to trust me," you know, like DNA evidence.

It's like, "Hey, we have DNA evidence." "What's DNA?" "Oh, I can't tell you." [laughs] "I have, I have fingerprints." "Well, what's a fingerprint?" "No, I can't tell you.

Yeah, I could tell you that the person was there because of the fingerprint on a glass." "Well, what the heck does that mean?" "Well, I can't tell you." Everyone needs to tip their hand of how they're coming to these conclusions.

You have to articulate the science.

But he was able to definitively say that it shouldn't be used in the justice system.

It's astrology.

Does that really matter?

It matters, but also is a opportunity.

Folks that are looking at it through the lens of obviously Anthropic did press releases around it.

They're trying to look at this from the lens of, "Look at the good we're doing.

We're pumping the brakes." It did get noticed by a lot of teams and organizations that have not had their security teams get a lot of, you know, questions toward it over the past few years.

Hiring's been flat.

New skill development has been flat.

What do you think CISOs should be doing faced with this reality? I mean, what, what are the immediate things that need their attention?

Rob, how have AI and autonomous agents changed the nature of threats, changed the nature of attacks? As David was talking, I saw you nodding your head in agreement.

I'm a big history buff.

Every single time there's a massive technological change, there's a massive change in tactics that need to follow it, much less the strategy.

You could go back to the introduction of the airplane in World War I.

The tank, especially back then, going back to the Civil War, the The line defense with Napoleonic and what we ended up seeing then all the way to the aircraft carrier.

I basically moved this mindset into warfare, into what's going on, cyber defense, cyber offense as a whole.

As a result of the speed velocity, you end up taking a look at, you know, essentially the blitzkrieg that happened in World War II.

Rob, thoughts on how incident response and forensics must evolve, especially since AI can navigate breached networks, erase its own tracks, corrupt evidence? The rules have changed, or the technology has changed, so how must incident response and forensics evolve as a result?

What must these evaluations include to actually catch some of the things like you talked about poisoning, jailbreaking and distribution shift before these systems go live, Rob?

Well, that's also, you're highlighting a very important part of the action plan, which again, I very much applaud, is that in order to deploy AI, and again, we're using very generically here, depending on the capability you're deploying, there are key questions that leaders should probably ask before these deployments are put onto their network.

If you have a massive amount of data that you're trying to utilize in AI to be able to analyze, is there ways for adversaries to potentially exploit that data or potentially poison that data against the organization itself say if you do a lookup of saying critical information that's needed for a supply chain or to be able to run your operation it'll potentially give you the wrong data back Kind of like similar, let's make it real world a little bit, is you put a prescription in to the pharmacist and the pharmacist ends up prescribing you something else that is lethal.

That is essentially what might happen under a prompt injection style attack is that you not even sure what the pills are and what you think they are.

because there's no way to validate it.

These regulations in this act is aimed at trying to encourage companies to follow some sort of standards that are out there.

SANS has their own, the AI security controls as well, and we're trying to work with other organizations to advise them to be able to do this.

Any thoughts on that?