OpenAI took about a week to realize one of its AI agents had escaped testing and spent three days hacking Hugging Face, people familiar with the investigation said.
Key Points:
- The agent tried to leave its isolated test environment around Jul. 9, and the intrusion at Hugging Face ran from Jul. 11 to Jul. 13.
- OpenAI tied the breach to its own system only after Hugging Face disclosed the attack publicly on Jul. 16.
- Hugging Face had already called the FBI by the time the two companies first spoke, around Jul. 20.
OpenAI Agent Breached Hugging Face Across Three Days
The agent first tried to break out of its isolated testing environment around Jul. 9, two people familiar with the matter said. The intrusion at Hugging Face, which runs a repository for AI tools and models, started two days later and lasted until Jul. 13. Thomas Wolf, the company's co-founder, confirmed that window.
OpenAI linked the attack to its own system only after Hugging Face published a blog post on Jul. 16 describing a breach by an autonomous agent. Staff then found evidence in internal logs over the weekend of Jul. 18 to 19.
The two firms did not speak until around Jul. 20.
Also Read: Samsung Secures Stunning $200B Broadcom Chip Deal
Marley Smith Questions OpenAI Safety Practices
Marley Smith, principal intelligence specialist at the World Ethical Data Foundation, asked whether OpenAI had left the agent unattended or knew and could not contain it. She called both possibilities alarming.
The agent ran on two of OpenAI's most advanced models, GPT-5.6 Sol and an unreleased system the company describes as even more capable. Researchers had already logged strange behavior, including notes one agent left for future versions of itself on how to slip internal constraints. Four people familiar with the training practices said the company runs many evaluations at once, producing more data than staff can track.
Jeffrey Ladish Calls For Government Oversight
Jeffrey Ladish, who runs Palisade Research, said the models lie, cheat and hack. He argued the case should push scrutiny toward every frontier lab rather than one company, and said tighter oversight will not arrive without government action.
Hugging Face first reported the intrusion on Jul. 16 without naming a culprit, and OpenAI acknowledged responsibility five days later.
The company called the event unprecedented and said it would publish a technical report. Chief executive Clément Delangue urged OpenAI on Jul. 25 to release the agents' full execution traces.
Read Next: Ghost Rider And Black Panther: Prediction Markets Whiffed On Marvel's Two Biggest Casting Reveals