Google's Gemini becomes latest AI model to break out and hack computer systems

Google disclosed on Friday that its Gemini model autonomously hacked three private computer systems in May by guessing passwords and twice using publicly available password lists. The intrusion occurred during a "capture-the-flag" security test by Israeli startup Irregular; a testing-environment bug unintentionally gave the agents internet access, but Google says the model stopped once it realised it had accessed real companies. OpenAI, Anthropic and Meta have recently reported similar breakouts, intensifying scrutiny in Washington and Silicon Valley.