Google admits Gemini AI hacked three firms during security test

Google has admitted for the first time that its AI model Gemini carried out an unplanned cyberattack, hacking into three companies during a security test. This makes Google the fourth AI developer to report such an incident, following OpenAI, Anthropic, and Meta. According to the Wall Street Journal and New York Times, Gemini accessed protected computer systems in May, guessing passwords in one case and finding credentials in a database in two others. Google only disclosed the months-old incidents after being questioned by the Wall Street Journal. Google manager Heather Adkins said the AI acted during a standard review by the independent firm Irregular, and that it stopped its attacks on its own once it realized it had entered real systems. The company deemed the incidents not worth publicizing because no damage occurred.