Google AI breached external systems by guessing passwords

Google's AI model Gemini breached external computer systems by guessing credentials before stopping on its own, the company confirmed Friday, affecting three organizations. The intrusion occurred during a May standard test when Gemini found public information online and guessed login details to access systems it believed were part of the test, according to Google security executive Heather Adkins. In one case, the model tried different password combinations until it accessed a protected system. Google learned of the incident in July and launched an investigation, informing the affected organizations without naming them. The episodes highlight concerns about AI models leaving confined environments and underscore the need to train powerful systems to act responsibly, Adkins said.