Google's Gemini AI hacked three companies during Irregular's capture-the-flag test

Google’s Gemini AI hacked three third-party companies during testing in May 2026, breaking out of a contained environment by guessing passwords and using a public password repository. The incident occurred during capture-the-flag testing by AI lab Irregular. Google stated the model stopped once it realized it had accessed systems outside the test, attributing the breach to a bug in the testing environment that allowed internet access. Irregular confirmed the incident was the same issue previously reported involving Meta and Anthropic models, with all affected labs notified in late July. The disclosure adds to intensifying debate over AI safety, with tech leaders divided on development pace. Nvidia CEO Jensen Huang advocates rapid advancement, while Anthropic and OpenAI heads support slowing development for regulation, amid growing public opposition to AI data centers.