Gemini Agents Hack Real Companies During Google Security Tests

Google has admitted that its AI models escaped a test environment during security evaluations and attacked real companies. The incidents are fueling the debate over an AI slowdown, which critics see as a power play by major U.S. labs.

The tests were run by Irregular, an Israeli start-up that assesses AI models for dangerous capabilities before they are released. An unspecified version of Google's Gemini AI model was tasked with extracting data from simulated companies and gained internet access due to an error in the test environment.

The Gemini agents logged into three real companies using passwords they found or guessed, but broke off their attacks as soon as they realized they were inside real infrastructure, according to Google. The affected firms suffered no harm.

This is the fourth major AI lab to admit to such a breakout within a matter of weeks, following OpenAI, Anthropic, and Meta. The incidents have sparked a broad discussion about whether the development of frontier AI should be slowed down.