Google said Monday that it had upended a criminal group’s effort to harness artificial intelligence to crack open a previously unknown software vulnerability, a breakthrough one of the company’s senior threat analysts described as the arrival of the “AI‑driven vulnerability and exploitation” era that cybersecurity professionals have long predicted.
Google offered few details about the perpetrators or the target, but the company’s threat intelligence arm said its investigators caught a group of prominent “threat actors” planning a major operation built around a bug they had discovered. The flaw let them sidestep two‑factor authentication and gain access to an online system administration tool that is used broadly; Google declined to name the tool or the company that was in the crosshairs.
The vulnerability was a so‑called zero‑day exploit — a term that means the software’s maintainers had zero days of warning to craft a fix. Google notified the affected firm and law enforcement agencies in time to stop any damage, but as its team traced the intruders’ digital footprints they found proof that the same kind of large language model that powers popular AI chatbots had been used to find the weakness. The company would not say which model the attackers used, only that it was probably not Google’s Gemini or Anthropic’s Claude Mythos. Google also withheld the name of the suspected group but said there was no indication it was tied to a hostile government, although crews linked to China and North Korea have experimented with similar techniques.
“It’s here,” John Hultquist, Google’s chief analyst at its threat intelligence unit, said in an interview. “The era of AI‑driven vulnerability and exploitation is already here.”
Hultquist stressed that criminal hacking gangs — as opposed to government spies, who often work slowly and methodically — stand to gain the most from AI’s “tremendous capability for speed” when hunting for and weaponizing security holes. “There’s a race between you and them to stop them before they can essentially get whatever data they need to extort you with, or launch ransomware,” he said. “AI is going to be a huge advantage because they can move a lot faster.”
The episode lands in the middle of a regulatory scramble touched off by the rapid advance of AI’s ability to find and exploit vulnerabilities. Anthropic sent shockwaves through the tech world a month ago when it unveiled a new model, Mythos, that it said was so “strikingly capable” at hacking and cybersecurity work that it could only be released to a small circle of trusted groups. Anthropic created a consortium, Project Glasswing, that brings together Amazon, Apple, Google, Microsoft, JPMorgan Chase, and others in an effort to harden the world’s most critical software against what the company called “severe” public‑safety and national‑security fallout.
Amid that anxiety, the White House’s posture has been anything but steady. President Donald Trump’s Commerce Department announced last week that it had signed fresh agreements with Google, Microsoft, and Elon Musk’s xAI to evaluate their most powerful models before public release — building on similar Biden‑era deals with Anthropic and OpenAI. Days later the announcement vanished from the Commerce Department’s website, the latest in a string of conflicting signals from an administration that had already fulfilled Trump’s campaign promise to scrap the Biden administration’s guardrails around the technology.
“Some people don’t want there to be a regulatory response to this and others do,” said Dean Ball, a senior fellow at the Foundation for American Innovation who previously served as a White House tech‑policy adviser and was a lead author of Trump’s AI policy roadmap last year. “I don’t like regulation. I would prefer for things not to be regulated. But I think we need to in this case.”
OpenAI, Anthropic’s top rival, moved in a different direction on Friday, announcing a specialized cybersecurity version of ChatGPT that will be available only to “defenders responsible for securing critical infrastructure” so they can locate and patch flaws in their code more quickly.
Ball, who favors a government‑coordinated effort to harden the “untold trillions of lines of software code” that run the world’s computing infrastructure, said that effort could take years. In the meantime, he predicted a “transitional period” in which cybersecurity risks rise dramatically and “the world might actually be more dangerous.”