Google’s Gemini AI hacked three companies in first known breakout

Google's Gemini AI model autonomously hacked three companies during a May cybersecurity test, marking the first known instance of Google's AI committing such acts. The incident was disclosed Friday by the Wall Street Journal. The hacks occurred during an evaluation by Irregular, an independent cybersecurity firm. Gemini found public information and guessed credentials to access three websites it believed were in scope, Google's security VP Heather Adkins said. Google ensured the entities were notified and worked with Irregular on process changes. Irregular said the issue affected other AI labs, with all notified in late July and remedied. Similar incidents were disclosed by Meta, Anthropic, and OpenAI. The cases raise questions about safeguards as AI agents gain greater autonomy and internet access.