Federal technology officials said on Aug. 25 that increasingly capable artificial intelligence (AI) models are giving agencies new opportunities to strengthen cyber defenses and advance mission work as the technology reshapes the cybersecurity landscape.
Speaking at a CGI cybersecurity breakfast, Anil “Neil” Chaudhry, senior advisor for AI at the Department of Transportation (DOT), and Tony Brannum, chief information security officer at the Agriculture Department (USDA), pointed to Anthropic’s Mythos as an example of how AI is rapidly changing cyber capabilities.
Anthropic’s Claude Mythos models have demonstrated the ability to autonomously identify and exploit previously unknown software vulnerabilities. But Chaudhry and Brannum said agencies can turn those advanced tools toward their own missions and cyber defenses.
“When I think about Mythos, and I think about the new cyber reality, it’s not gloom and doom. And agentic AI isn’t gloom and doom either,” Chaudhry said. “We have been using agentic AI to do some very innovative things.”
For example, Chaudhry said DOT has leveraged agentic AI from multiple companies to identify foreign carriers operating outside of designated commercial zones. The system produced photographic evidence that officials could use when carriers returned to the U.S. border rather than trying to intercept them on the road.
Similarly, he said agents could automate follow-up with rental car companies or dealerships to verify reports of stolen vehicles before police respond.
“An AI agent is really good, especially these more powerful ones, that can take a human persona and do the back and forth,” Chaudhry said. “That’s the kind of innovation that we’re having with Mythos and our new cyber reality. We are talking about agents that are so good, you don’t know whether they’re human or not.”
Brannum said USDA is beginning to pilot similar capabilities while focusing on identity management, adding that “it can’t be the old days where you just give the agent access to everything.”
“It’s really about risk management on these new tools,” Brannum said. “We’re starting to pilot some of the Mythos now and [ask ourselves], ‘How can we use it? How do we start to protect it?’”
“So, I think the more we start to learn it, embrace it, and use it to our advantage, then we can flip the script [so that] we’re not always afraid of it,” he added. “I think it’s a dual-hatted thing.”