So once it makes one error, if you have too many steps in a row, it's creating compound errors.

Yes.

So you're very generous to say it's the world's smartest intern.

I wouldn't even go that far.

I think it's just like any capable intern in that it's smart enough to be dangerous.

Like you took some classes and now you think you know the subject.

Right.

What's your hope for this AI revolution? Like, where can this take us if it goes the right way?

But, like, why is it so important, Camille, to have this conversation around what cyber looks like today in the age of AI? 'Cause I think the book really brings that out, too, in terms of governance is really required as well.

Yes.

So first I decided to focus on organizations because we are in a policy moment where policy is not moving fast enough.

Um, we're not wrapping our heads around it enough.

And in certain places, like here, we're paused on action at a national level in terms of a concrete AI policy.

And that means that industry is on the front lines of writing the rules and the societal norms and the way this technology is going to impact our lives.

And unwittingly, organizations have stepped into a role that they weren't [laughs] designed to fill.

Mm-hmm.

So in some ways, it seems to me, I don't know, I'm interested in your opinion, but in some way, it seems to me that one of the keys to managing these AI systems is learning the lessons we didn't learn from managing people.

Exactly that.

And honestly, even just learning the lessons we did learn through managing people.

If we even started there, we'd be in a much better position.

But yes, to your point, there are so many lessons we need to continue to learn about not overtrusting a technology or a person and not losing the managerial discipline that we have come to know, like things like separation of duties or checking someone's work.

to your point, become easy to let go as the trust builds, but are implemented for a reason.

And they're definitely necessary in this environment.

And I think one of the impediments is people don't understand the availability and utility of the kind of systems you're talking about.

So in some ways it seems to me, I don't know, I'd be interested in your opinion, but in some way it seems to me that, uh, one of the keys to managing, um, these AI systems is, is, is learning the lessons we didn't learn from managing people

peopleExactly that.

And honestly, even just learning the lessons we did learn through managing people.

If we even started there, we'd be in a much better position.

But yes, to your point, there are so many lessons we need to continue to learn about not, um, over-trusting a technology or a person, and not losing the managerial discipline that we have come to know.

Like, things like separation of duties or checking someone's work, to your point, become easy to let go as the trust builds, but are implemented for a reason, and they're definitely necessary in this environment.

I think what also complicates this is that not only is the flattery enticing and trust-building or, you know, at least making you connect to the technology, it's right just enough, right? Or sounds right enough.

And I think one of the impediments is people don't understand the, uh, availability and utility of the kind of systems you're talking about.

What were you trying to solve when you developed that framework?

That framework is an opportunity for an organization to align how it looks at some of the common, um, practices that we use to constrain technology, like detection and enforcement and incident response and, you know, thinking about the learning that happens after, whether that's through a postmortem.

But in this case, I advocate for continuous learning and a more agile way to kinda be iterative about how you intake information from an incident or a potential incident.

That framework is really designed to help an organization think through how those things evolved, how the organization realigned to support doing those in a way that actually will help constrain agent behavior and operationalize it to create some operational reliability rather than just to mitigate risk.

And it's not about attribution in the sense of...

Attribution is important in terms of articulating who is accountable, which agent did what, but not attribution in the sense of, you know, what we commonly think about in cybersecurity.

It's not for consequence, but for-- to facilitate learning.

What, what kind of constraints or mechanisms should organizations have to execute this idea of governed execution?

Yeah,

I mean, that's the whole job.

I mean, a lot of the work is about getting groups that are motivated by different things to align in a way that advances a larger set of values or goals.

Sitting in government, my job was to pull together the private sector, civil society, academia, and all of the agencies within the executive branch to align around where we should go.

So when you think about AI, there's a real opportunity.

This is like the burgeoning of the internet.

So much about society is going to change.

I'd love to hear a little bit more about that, both from the standpoint of what was accomplished, but also how influence was part of that.

[upbeat music] Camille Stewart Gloster is former US Deputy Cyber Director and author of the new book, The Insider You Built: How Organizations Stay in Control of Autonomous AI Agents.

I began working with companies to deploy AI within their organizations after leaving the White House and remarked at just how many organizations were confused, worried, um, unsure about the effort that they were undertaking and wanted to move forward because the technology was something they felt they needed to keep pace with how change was happening across industry.

But were not making the necessary investments across security and even just operational resilience to understand how the technology was changing the nature of their organizations.

And I was learning so many lessons through my client work that I wanted to capture that for a framework that is in the book to help organizations really understand that as they move into agentic AI and deploying more in their organizations, here are the organizational realignment, the security, the risk mitigation, things you need to do and how your organization needs to come together such that you can realize the value of AI.

It's really about maximizing the ROI and less about actually being a security book itself.

What were you trying to solve when you developed that framework?

That framework is an opportunity for an organization to align how it looks at some of the common, um, practices that we use to constrain technology, like detection and enforcement and incident response and, you know, thinking about the learning that happens after, whether that's through a postmortem.

But in this case, I a- advocate for continuous learning and a more agile way to kinda be iterative about how you intake information from an incident or a potential incident.

[laughs]

Because what that means is when they get access to your whatever, they have access to so many things, and they're gonna test every site that you use to see if that password can be used there as well.

The other thing you shouldn't do is authenticate into something with something else.

So you'll often see, "Use your Google, use your Facebook to get into this other site." I don't recommend that.

It's so easy, but when you do that, you make this kind of connection that we're talking about, particularly if, not, if it's done without connecting it to a multi-factor authentication.

So if they get into your Google account, they've gotten into all of these other sites, and potentially can even lock you out of your Gmail account.

Keep your s- work and personal accounts separate, and consider using one of the privacy-focused browsers, like DuckDuckGo, and block the trackers.

Can you talk about just the impact that has on our mental health and how it impacts us psychologically?

So you don't have to believe in the end of the world to say, "Gee, we're entering a new kind of, uh, cybersecurity environment." Right?

Oh, for sure.

I mean, the fundamentals of the discipline have to change.

It's no longer a, you know, deterministic environment.

You have to think about how behavior shifts happen, when there's emergent behavior, when, um, you know, you've given access to an entity that will do things thinking it is operating in your intent, but there is an, a misalignment of incentives that act, as- that, that demand of it or cause it to act in a way that you did not intend.

Also, let's think about it this way.

There was a, um, a research c- that came out recently through CrowdStrike about the DeepSeek model.

Well, first of all, I mean, do you see those as positive things? And secondly, what's working right now? What are we doing better now than we, we haven't done before 'cause we're understanding the issues better?