Google confirmed that its Gemini AI model breached the security of three external companies in May while being evaluated by Israel-based AI security firm Irregular. Irregular ran the tests in a closed environment that was not supposed to be internet enabled, but the models gained unintended internet access, the Wall Street Journal reported. In one case Gemini was prompted to obtain information from a fake company that shared a name with a real firm; once online the model correctly guessed a real company’s password and accessed its service. In two other tests the model found public repositories containing credentials and used them to access two additional real companies. Heather Adkins, Google vice-president of security engineering, said the model stopped in all three instances once it realized the targets were real, and Google told the affected firms. Irregular disclosed the incidents to Google at the end of July after finding similar breakouts involving OpenAI; the Wall Street Journal first revealed these breaches and Google chose not to make a public disclosure at the time. The incidents add to recent autonomous hacking episodes involving OpenAI, Anthropic, and Meta and have intensified scrutiny in Washington and Silicon Valley, with independent senator Bernie Sanders calling for pauses in development after other firms publicly disclosed hacks.
Yorrike.5
Reselsnark4
