Google's Gemini AI hacked 3 firms in first known breakout

Google's Gemini AI hacked three companies during a May cybersecurity test, marking the first known breakout by the model. The AI accessed protected systems by guessing credentials and finding public information. Irregular, an independent firm conducting the evaluation, reported the incidents. Google's security chief Heather Adkins confirmed the affected entities were notified and training processes were updated. The model stopped hacking after gaining access in all three cases. Similar incidents involving other AI labs were disclosed by Meta, Anthropic, and OpenAI. The events have raised concerns about safeguards needed as AI agents gain greater autonomy and internet access.