A lot of times people can go into your system through the AI, right? Is there like prompt injection where they can ask the AI to pull data from your company that shouldn't be public? Is that one of the things that people can do?

Prompt injection is its own thing.

Whether you're doing social engineering prompt injection or you're actually doing drop malware, those are a little different.

I think of it more like this.

If I'm using ChatGPT or Copilot or Anthropic Squad, The first thing it asks me to do is sign in and likely link one of my online profiles, Gemini for Google.

All of them are doing through Google, your Gmail accounts, and you give it access to a whole bunch of information.

And then it asks, do you want access to your calendar? Do you want access to your email? If somebody were to get into my email, how hard would it be for them to get into my AI or vice versa? And so.

And since those are interviewing for a company, what are red flags that they'd see where it's a company that's trying to fill a role, but doesn't want to make any changes?

First off, what are your thoughts on using AI to solve some of these problems in the security space? as it relates to maybe alert fatigue or billing your team out and things like that.

I think security's been doing it for 30 years, right? I mean, in 2019, 2020, I was a part of a team that built two machine learning models for network detection response tool, fully patent the way it actually did, the way it did its actual detection response component, fully unsupervised, which was a precursor to kind of what we're seeing with Gen AI.

And it was reaching 98.9% accuracy on unknowns, right? Knowns are different, but unknowns.

And so we've been doing it for a long time, right? Machine learning is just an early variation of artificial intelligence.

So I think what we're seeing now is an advancement of that, right? How quickly can I turn an artificial intelligence tool, a generative AI tool, an agentic AI tool into something into a threat hunter, into an immediate responder, instead of just quickly finding the alerts.

Now, I think I saw something recently that those that are using like a Cloud Anthropic or a ChatGPT or even a Microsoft Copilot to review and correlate logs, are in that 99.8% accuracy of removing false positives.

That's insane, right? You're never going to be perfect.

So how do you stay on top of that? What's your