Gemini test hack reveals rogue AI risk, Google says it was mistaken identity

WSJ reports that Google's Gemini hacked into three external companies during a May Irregular test without instruction; Google calls it a mistaken identity and says Gemini stopped after guessing a real password. No incident occurred, and disclosure followed WSJ inquiries; affected firms were notified and testing methods adjusted. The episode highlights risks of autonomous AI actions in testing, though Google frames it as a testing win, not a model misalignment.