Google announced on Monday that it had disrupted a criminal group's attempt to use artificial intelligence to exploit a previously unknown digital vulnerability in another company's systems. This incident has intensified concerns across government and private industry about the risks AI poses to cybersecurity.
Google shared limited information about the attackers and the target. However, John Hultquist, chief analyst at Google's threat intelligence arm, stated that this represents a moment cybersecurity experts have warned about for years: malicious hackers using AI to supercharge their ability to break into computer systems worldwide. "It's here," Hultquist said. "The era of AI-driven vulnerability and exploitation is already here."
AI's Role in the Cyberattack
Google observed a group of prominent threat actors planning a major operation that relied on a bug they had discovered. The vulnerability allowed them to bypass two-factor authentication and access a popular online system administration tool, which Google declined to name. The company described this as a zero-day exploit, a cyberattack that exploits a previously unknown security vulnerability. The term "zero-day" indicates that security engineers have had zero days to develop a fix.
Google notified the affected company and disrupted the operation before any damage occurred. However, while tracing the hackers' digital footprints, Google found evidence that they had used an AI large language model—the same technology powering popular chatbots—to discover the vulnerability. Google did not reveal which AI model was used, only that it was likely not its own Gemini or Anthropic's Claude Mythos. The company also did not identify the suspected group but noted no evidence of ties to an adversarial government, though groups linked to China and North Korea have been exploring similar techniques.
The Implications of AI-Driven Attacks
Hultquist explained that criminal hackers have much to gain from AI's speed in finding and weaponizing security bugs, compared to government spies who typically operate slowly and quietly. "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."
Regulatory Responses and Industry Reactions
The incident comes amid significant advances in AI's ability to find vulnerabilities, including Anthropic's Mythos model announced a month ago. The White House under President Donald Trump has shifted its approach to vetting powerful AI models before public release. After repealing former President Joe Biden's AI guardrails, the current administration has sent mixed signals about government oversight. Dean Ball, a senior fellow at the Foundation for American Innovation and former White House tech policy adviser, said, "Some people don't want there to be a regulatory response to this and others do. I don't like regulation, but I think we need to in this case."
Last week, the Commerce Department announced new agreements with Google, Microsoft, and Elon Musk's xAI to evaluate their most powerful AI models before public release, building on previous agreements with Anthropic and OpenAI. However, the announcement later disappeared from the Commerce Department website, highlighting the administration's inconsistent approach.
Anthropic's Mythos and Industry Collaboration
Anthropic's Mythos model, described as "strikingly capable" at hacking and cybersecurity, has sparked panic and calls for regulation. Anthropic launched Project Glasswing, bringing together tech giants like Amazon, Apple, Google, and Microsoft, along with JPMorgan Chase, to secure critical software from potential fallout. However, its relationship with the U.S. government has been complicated by a public and legal fight with the Pentagon and Trump over military use of its AI technology.
OpenAI has since introduced a similar model, releasing a specialized cybersecurity version of ChatGPT only available to "defenders responsible for securing critical infrastructure" to help find and patch vulnerabilities.
Looking Ahead: A More Dangerous Transitional Period
Ball expressed optimism that AI tools will eventually improve coding and make systems safer from routine cyberattacks. However, he warned that the world's computing systems rely on "untold trillions of lines of software code" that are at risk if AI tools are unleashed to exploit all their bugs. Hardening this software could take years and would benefit from U.S. government coordination. Ball predicted a "transitional period" where cybersecurity risks rise significantly, making "the world might actually be more dangerous."
