A new report from Reuters alleges that it took an alarmingly long time for OpenAI to respond to one of its prototype AI agents—programs with the ability to act independently of specific prompts—breaking containment and hacking another company. That company, Hugging Face, had reportedly neutralized the threat and contacted the FBI before ever being contacted by OpenAI.
According to anonymous sources, OpenAI's testing of an agent powered by two advanced models—GPT 5.6 Sol and an unnamed "even more capable" model—showed troublesome behavior in testing even before the Hugging Face incident.
Reuters' sources spoke of the model leaving itself instructions to bypass testing constraints, as well as an incident where the agent had seemingly disabled some of the company's monitoring systems.
Hugging Face made its first public statement about the hack on July 16, before it had identified or without disclosing OpenAI's involvement. OpenAI publicly acknowledged the incident on July 21. According to Reuters, OpenAI's prototype escaped its testing constraints on July 9 and began attacking Hugging Face on July 11.
Sources told Reuters that OpenAI did not realize that its agent had gone rogue until after Hugging Face had neutralized the threat, contacted the FBI, and made its public statement on July 16.
The FBI declined to comment, while Hugging Face is preparing a full timeline of the hack. OpenAI told Reuters that there were "several inaccuracies" in its report, but did not specify what those inaccuracies were.
Speaking to Reuters on the matter, World Ethical Data Foundation principle intelligence specialist Marley Smith outlined two possible explanations for OpenAI's delay in response: "Does that mean that they left it unattended and didn’t realize what it was doing? Or maybe they did and didn’t know how to contain it? Both are equally dangerous and alarming."
