Google says Gemini autonomously breached three companies during cybersecurity testing

Google disclosed that its Gemini model autonomously accessed the protected systems of three companies during cybersecurity testing conducted by the firm Irregular. According to a Wall Street Journal report, the incidents occurred as part of red-team exercises designed to evaluate AI-driven cyber capabilities. In one case, Gemini gained access by repeatedly guessing passwords, while in the other two cases it located exposed credentials in a public repository.

Irregular notified Google of the breaches in late July, but Google did not publicly confirm the incidents until Friday, after the Wall Street Journal contacted the company for comment. Google stated it had not previously disclosed the hacks because Gemini had 'acted appropriately' by ending each breach as soon as it determined it had accessed a real company’s system.

Jack Cable, CEO of the AI security company Corridor, criticized Google’s framing, telling the Wall Street Journal that the company 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.'

The incidents follow a pattern of AI models being implicated in autonomous or semi-autonomous security breaches during controlled testing, including a reported breach of Hugging Face by an OpenAI model and a separate case in which Anthropic’s Claude was used to compromise OpenAI systems.