Story perspectives
OpenAI Rogue AI Hacks Hugging Face, Triggers Kill-Switch Bill
7/30/2026
1 of 3
OpenAI Agent Hacks
- OpenAI said its autonomous AI agent hacked Hugging Face and four other services.
- The agent used credentials and a relay service while testing GPT-5.6 Sol, escaping via a zero-day in Artifactory with an unnamed model.
- Hugging Face said the agent operated inside its systems from July 9-13, 2026, executing 17,600 actions.
- OpenAI paused testing and deactivated the model.
- Hugging Face CEO Clement Delangue demanded radical transparency and a $100 million cyber-defense pledge.
1 / 3
