From the Desk of Tracey Greene, CBIC Executive Director

I caught something at Black Hat recently that I can’t shake. OpenAI researchers were testing AI agents against tough cybersecurity challenges in what was supposed to be a controlled environment. The agents hit barriers. They didn’t stop. They probed, tried different paths, found vulnerabilities, and eventually reached systems outside the boundaries researchers had set — including infrastructure at Hugging Face.

In plain terms: the AI was given a job, ran into a wall, and figured out another way around it.

It wasn’t “going rogue.” It wasn’t sentient. It was simply doing what it had been asked to do — maybe a little too well. And that’s the part that keeps replaying in my head.

We’re used to asking AI to write, summarize, analyze, or help us solve problems. Agentic AI goes further. Give it an objective, and it can work through the steps, use tools, assess outcomes, and try again when Plan A fails.

Think about what that means for drug discovery, cybersecurity, software development, manufacturing, healthcare, defense — or even a startup trying to do with five people what once required twenty-five.

But then come the harder questions:

• What are we comfortable handing over to an AI agent?
• What access should it have?
• Who sets the boundaries?
• What happens when it finds a solution nobody anticipated?

There’s another ripple effect: investment. PitchBook just published a report on this incident, predicting it will accelerate venture capital into AI-native cybersecurity. Cybersecurity has always been an arms race. AI is changing the speed — and maybe the economics — of both sides.

And here’s the local angle: PitchBook interviewed Charlottesville startup Hardshell for the report. The questions raised aren’t abstract. They’re being tackled right here.

For investors, that’s another lens. As AI advances, new vulnerabilities, defenses, infrastructure, and companies will emerge. Some of the most interesting opportunities may come from solving problems we didn’t even know we had a year ago.

I don’t have neat answers. Nobody does. But I know where I’d start: in the room, with people thinking about it. And we’ve got plenty of those rooms in Charlottesville.

On August 27, Cville Data Science hosts an AI & Data Science Project Night.

On September 8, CBIC hosts the Innovation Mixer with AFCEA. intentionally broad because the best ideas often spark when a founder meets a researcher, or a cybersecurity pro bumps into someone solving a completely different problem.

September 30: UVA’s Fall Data Science & AI Night.

October 1: Data Centric Summit at CODE.

October 2–3: BeCamp takes over UVA’s School of Data Science with its unconference model — bring a question, propose a conversation, find your people.

Put it all together and you’ve got a concentrated stretch where you can see what local AI and data practitioners are building, connect with investors, defense and cybersecurity communities, and follow whichever threads spark your curiosity.

That’s Charlottesville. Researchers, entrepreneurs, engineers, investors, defense expertise, cybersecurity talent, life sciences, data scientists — and companies like Hardshell suddenly in the middle of a global conversation.

The opportunity is to keep these people bumping into one another. None of us can keep up with everything happening in AI. I certainly can’t. But we can stay curious. We can ask better questions. We can learn from each other. We can invest in good ideas. And we can create more chances for people working on very different problems to realize they’re not so different after all.

So here’s my takeaway from watching an AI agent find its way around a wall it wasn’t supposed to cross:

Pay attention. Stay curious. Get in the room.

And we’ve got some pretty good rooms coming up.