Your AI Agents Have No Background Checks. Who Gave Them Access? 

No Background Check, No Supervisor

Rasha Horn: I’d say yes and no. They’re aware that it could be something really disastrous if it went wrong, but I don’t think they’re aware of all the ways that it could happen. And some of the reasoning is because whenever you have an AI agent, it’s not going through all the different processes as if it was a human worker. You would run a background check on them. You get references. Multiple people might do the interview process so they can see if there’s blind spots or any of that stuff. Excuse me, but with an AI agent, it goes freely the first day, and on the first day, you give it all these responsibilities, but it also has all this power and access.

You don’t give that on the first day to an employee. They are under a manager or someone in training. You do onboarding. Also, you do offboarding when they leave. So, if you have an AI agent, it’s probably going to go hire its own AI agent. An employee wouldn’t do that. A human. And also you’re going to have so many that a lot of the companies are not tracking how many AI agents they have, and they’re not aware to do that.

It’s just everything is happening so fast that a lot of them are keeping an eye on how can it be productive, what can it do, how can it save the company or keep up with all the other companies, that they’re missing on some of the stuff about security.

Jo Peterson: You make a really good point by drawing out that parallel between a human worker and how they would handle things and an agent because it’s not the same. But you’re right. It’s not the same. You wouldn’t walk into the door of a business on day one and be given full access to everything. You just also even—

Rasha Horn: Has someone above them. Like there’s always another authority to check over them. But an AI agent, you’re like, here, go, and then it just has access. It’s not always, you know, monitored by other people.

Do AI Governance Committees Actually Govern?

Jo Peterson: That’s such a good point. I like that, and that’s a great way of thinking about it. And you know what? If you’re able to articulate that, so hopefully, audience, that helps because being able to articulate it the way Rasha just did might help you have a conversation upstairs. Here’s hoping.

Second question: 50% of large enterprises now have dedicated AI governance committees, and I applaud them. But I’m wondering: Is that structure actually working, or is it creating the appearance of governance without the substance?

Rasha Horn: It’s working and not working again, like the other. It’s working, but honestly, it’s great. They thought of that and they put people in place. But the part that’s not fully working is they don’t know what to compare against, or what to put, or who is fully qualified to know what AI does.

And it’s also mainly like rules and regulations, but it’s not an active process day to day. It’s like where this is where we’re gonna follow, but you need a committee. That’s if something happens now, who do you turn to? A lot of companies don’t know. Like, do I go to my CISO? Do I go to anyone at IT? Like, no one knows where to go, and a lot of them don’t know the process.

So yes, it’s nice to have a process, but if it’s not implemented in real time, you need someone all the time under that. Like part of it is just going to be another meeting that you just have once a month. Let’s look at it, what’s happening, and you won’t know what’s going on to update. You know, like how to train the employees on what went wrong.

So sometimes yes, it is the employees, but a lot of the time, the people who are implementing this aren’t even aware because it’s just so much to handle all at once.

And to be full disclosure, I’m on this committee for a consortium called AI UC One, and they come into companies and look at the framework and help certify them for AI, and they’ve worked with all kinds of organizations for the guidelines to create this community. So they come in and help with that stuff. So there’s support. It’s just hard because everything is so new, where to look.

Jo Peterson: Yeah, that’s a fair statement, right? And—

Rasha Horn: Cloud Security Alliance is one of the main ones that helped, you know, with all that. So that’s amazing. Like it’s nice to see the leadership up top of security is in on it. It’s just hard for the non-technical to get in.

First 30 Days: Inventory and Ownership

Jo Peterson: Shout out to my people at the Cloud Security Alliance. They’re amazing. Good work. You’re doing good work.

All right. Third question. You are the CISO. You walk in the door. You take a look around, and you’re like, “Great, we are using AI. Not great. We are way behind.” And when I say behind, you start to ask some questions, and you realize that there’s no inventory being captured, like you had pointed to earlier. There’s no agent-specific controls. There’s no monitoring. What’s the single most thing that you would do in the next 30 days?

Rasha Horn: I would say the first week for sure, I would get inventory, but not just have IT, you know, check here’s a list of what we have. I want to know, okay, yes, we have that list, but who’s in charge of all things? So if something goes wrong, who do we go to? Who is going to be held accountable if someone leaves or something happens to that person? Who’s next?

Like you can’t just say this person is in charge of all these AI agents. People take vacations. People have things that come up. They get sick. You need to have a backup person, like a deputy or something, for whoever’s helping do that.

And then once you figure out that, you see which ones are the least secure, and figure out what you need to secure over the next two to three weeks, and check on those agents, and not just the agents alone. Also, what are they connected to?

Sometimes you have an old server that’s sitting there since 2012 that somebody forgot about, and that’s important. You can put all the AI security, you can train the people, but you have that one endpoint that’s just sitting there. That’s one of the other things we should look into. So we can’t just blame everything on AI.

AI and Application Rationalization

Jo Peterson: I’m so glad that you said that because I’ve been thinking a lot about when we started doing work in the cloud, we would do an application rationalization to understand how things were connected, and I think that maybe it won’t have that same name, but I think that that idea of some sort of application rationalization for AI—

Rasha Horn: Yeah.

Jo Peterson: We’re going to start to see that happening again. I’m not quite sure what the nomenclature is going to look like, but the concept is there and it’s valid, right?

Rasha Horn: And it’s the cool thing. Sometimes it doesn’t exist yet, and it’s a problem.

Jo Peterson: Yeah.

Rasha Horn: But it’s also a good thing because the more we make non-technical leaders more aware, then they can also help us because we see it differently than they do.

Jo Peterson: Yeah.

Rasha Horn: That’s why it’s good to involve them in conversations like this, and that’s why I love this question.

Jo Peterson: Yeah, it’s important to think about it through a prismatic approach, a lot of different lenses, right? So lovely to have you on today. Thank you.

Rasha Horn: Thank you so much.

Jo Peterson: Audience, we appreciate you joining, and we’ll catch you next time.

Rasha Horn: Thank you for all your great work. Thank you.

Jo Peterson: You too.