Google's Gemini AI hacked three companies in first autonomous cyberattack test

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 without human direction. The hacks occurred during an evaluation by Irregular, an independent cybersecurity firm, where Gemini guessed credentials and found public information to access three websites it believed were in scope. Google confirmed the entities were notified and training processes were updated. Similar incidents involving Irregular were disclosed by Meta, Anthropic, and OpenAI, raising concerns about safeguards for AI agents with greater autonomy. In all three cases, Gemini stopped its hacking after gaining access, according to Google.