You might think that the information-security professionals who show up in August at the annual Black Hat conference in Las Vegas would know to bring their A game to that gathering. But every year, the administrators of the show’s Network Operations Center (NOC) see a different, messier reality play out over the conference Wi-Fi.
And at last week’s Black Hat, AI played an even larger part in that drama—as a tool and a threat.
Agentic triage
AI has delivered the biggest overall boost to NOC staffers in threat assessment, both by spotting suspicious behavior on the network and by helping determine whether it’s standard-issue conduct at a conference where attendees regularly demonstrate attacks or legitimately hostile behavior that requires attention.
“We have alerts; we combine those alerts into detections,” says NOC admin James Pope, whose day job is senior director of security product research at the security service Corelight. “We’ve then taken those detections and rolled them up into what we call agentic triage.”
Pope’s assessment: “When it works, it’s amazing.”
A presentation on the conference’s final day helped illustrate why NOC admins describe their task as “finding a needle in a needle stack”: Of 8,892 issues observed on the network, 2,593 became cases, 2,543 of which were handled automatically.
The remainder demanded human attention, judgment, and sometimes intervention. “Illegal stuff at Black Hat is still illegal,” Pope says.
One advantage of this triage tool—which combines Corelight’s existing network-security tools, components Pope built himself, and third-party components from Splunk and other developers—is that it lets users verify its output rather than simply accept it. “It’s tied all the way down to the raw log, so it’s not just ‘Trust me, bro, AI is great,’” he explains.
It also doesn’t make users go out of their way to check its data: NOC admins can interrogate the system via a bot in the Network Operation Center’s Slack workspace. “That was one of my big lessons learned from last year,” Pope says. “Meet them where they are, wherever they normally do operations.”
Yes, even technical types who deal with programs all the time don’t appreciate being told to reprogram themselves before using a new tool. Telling them “Go log into a portal,” Pope says, is not likely to get the desired uptake. “You lose so many people.”
Bad vibes from vibe coding
The news of multiple AI agents running amok in tests and escaping test environments to hack other companies’ systems—an OpenAI model’s intrusion into Hugging Face’s computers being the most prominent—was a frequent topic of Black Hat banter, but Pope says that kind of rogue behavior has not been a real issue from the NOC perspective.
Planned AI operations staged for educational or research purposes were another thing. “Are people using AI to run attacks? Yes, absolutely,” Pope says. “Is it an agent they forgot about that actually did it? I’m not seeing a lot of that.”
A different sort of AI downside, however, has been impossible to miss: software largely or entirely coded by AI models that somehow lacks basic security measures.
To be sure, so-called vibe coding can democratize software development and unlock individual creativity. But on the Black Hat network, too many of the resulting apps lack the essential feature of encrypting data in transmission. “Everybody’s vibe coding,” Pope says. “All these apps are getting slapped up without TLS.”
Pope is referring to Transport Layer Security, or TLS, the encryption protocol that protects data as it travels between a user and a website or app. It’s hardly new: TLS was already common enough a decade ago for Gmail to begin warning users when incoming messages weren’t encrypted.
Pope cites one case where a camera in somebody’s home was sending an unencrypted video feed that at one point had their kids in view. In another case, an attendee used an internet-calling application also absent of encryption—to call his bank. “It’s all in the clear,” Pope laments. “It’s kind of a mess right now.”
Then, he continues, the NOC admins found 10 or so attendees using the same broken calling app by the same developer. “Now I gotta go reach out to that company,” he says.
Pope’s to attendees: When you get home, run a network analysis tool like the free Wireshark to see whether your own apps are leaking data.
Many taps on shoulders
Pope notes that AI “is absolutely very good” for creating systems that can warn people at scale about security problems observed over a network. But being a Black Hat NOC admin provides an even bigger advantage: instant credibility when sending those virtual taps on shoulders.
“We’re sending these from a blackhat.com address,” he says. “We’re not saying, ‘Oh, I found a thing, and I want to be on your leaderboard,’ or ‘I need you to give us some money.’”
One thing he and his colleagues try to get across is that staying off the Black Hat Wi-Fi won’t necessarily keep people safe when they face the exact same risks bringing an insecure app to any other network.
“They perceive the Black Hat NOC is attacking them,” Pope says. “We’re just telling you, this is what we’re seeing; there is zero touching of your stuff.”
Using a virtual private network service to encrypt your entire connection can protect against eavesdropping by somebody on the same network, but not if the VPN, like too many, has bugs of its own.
“Unfortunately, some of those VPN providers are hot garbage, and they are leaking your stuff out,” Pope warns. That, combined with the near-universal adoption of TLS encryption by websites (Chrome will now refuse to connect to a site without it), is why so many experts have long since consigned “Use a VPN on public Wi-Fi” to the realm of tech myths.
Instead of suggesting that people use new and different software tools to shield vulnerabilities, Pope and his colleagues have a simpler request: “Stop having janky apps.”