OpenAI and Hugging Face team up to fix security flaw found during model testing

OpenAI has finally accepted the responsibility that its AI models mistakenly hacked Hugging Face last week.

The ChatGPT maker termed the incident “unprecedented," involving state-of-the-art cyber capabilities.

OpenAI shared a blog post on Tuesday, July 21, and said that the models intruded into Hugging Face’s system, which hosts AI models and datasets, during testing of their models.

“We consider this incident to be an unprecedented cyber incident, involving state-of-the-art cyber capabilities, and are responding accordingly,” OpenAI stated in the blog post.

According to OpenAI, “We are sharing preliminary findings at this stage to help defenders understand what happened and to help calibrate on what models are now capable of.”

Calling all forces to join hands, OpenAI said that it had asked the models to pursue “advanced exploitation” and develop “complex attack paths in a bid to test their cyber strengths.

OpenAI said the attempt was driven by a combination of OpenAI models—including GPT-5.6 Sol and an even more powerful beta version.

These models had reduced cyber refusals for testing while also undergoing internal changes, the ChatGPT maker explained.

Hugging Face CEO Clem Delangue said, “We’re grateful for the collaboration with OpenAI on this and other topics.

Calling for joint efforts to curb such attacks in future, Delangue added, “This incident, possibly the first of its kind, proves a point we’ve long believed: AI safety won’t be solved by any single company working in secret.

“It will be solved in the open, collaboratively, with broad access to AI for every defender, everywhere.”

Hugging Face revealed in a blog post shared on July 16 that a new kind of security flaw was detected in an AI agent that hacked the company's infrastructure early last week.