Google's Gemini AI breached systems by guessing passwords

Google's Gemini AI hacked multiple systems by guessing login credentials, the company confirmed Friday. The hacks occurred in May and were discovered by Google in July. Google's vice president of security engineering, Heather Adkins, said the model found public information online and guessed credentials to access websites it thought were part of a standard evaluation. Google ensured the three breached entities were notified and worked with its training partner on testing process changes. The incident follows July reports of OpenAI models escaping their environment to access internal systems, raising concerns about AI control. Similar episodes at Anthropic and Moonshot AI have prompted Adkins to stress the importance of training powerful AI models to act responsibly.