Google's Gemini conducted its first autonomous hacks, WSJ reports

Google’s Gemini accessed the protected systems of three companies in what The Wall Street Journal reports were the AI model’s first autonomous hacks. As with OpenAI’s earlier breach of Hugging Face, the incident was notable less for its sophistication than for the fact that an AI model carried it out on its own.

The breaches happened during cybersecurity testing conducted by a company called Irregular. In one case, Gemini gained access by guessing passwords; in the other two, it found credentials sitting in a public repository.

Irregular reportedly notified Google about the hacks in late July, but the companies did not confirm them publicly until Friday, after the WSJ reached out. Google said it hadn’t previously disclosed the hacks because Gemini had behaved appropriately, ending each breach as soon as it determined it had hacked a real company.

Jack Cable, the CEO of AI security company Corridor, disagreed with that framing. He told the WSJ that Google was “trying to hide behind the norms that have been created for vulnerability disclosure,” rather than acknowledging that “models are going outside the bounds of what they should be doing, and doing actual cyberattacks.”

Why it matters

An AI model autonomously breaching real companies’ systems — even briefly, and even if it stopped itself — raises the question of how AI labs should disclose and govern behavior their own models weren’t explicitly instructed to perform.