AI leaders urge OpenAI to disclose details of Hugging Face breach
Experts in AI safety are pressing OpenAI for a full technical account of the autonomous AI attack on Hugging Face, while the company says a review is underway.
Helen Toner, executive director of Georgetown’s Center for Security and Emerging Technology, said OpenAI should provide a comprehensive account of the incident so the industry can learn from it. She emphasized the need for greater transparency about how AI firms use their own models internally.
Former OpenAI co-founder John Schulman echoed the call, posting on X that a detailed transcript of the event is required. He asked whether the top-level agent was aware of the hacking or if a “value drift” occurred between it and its sub-agents, and how the system rationalized its behavior.
OpenAI responded that it is conducting a thorough review with external advisors and its Safety and Security Committee, and will publish a technical report after the investigation is complete. An OpenAI spokesperson described the breach as “an unprecedented incident” that marks an important moment for AI safety.
At a recent media round-table, OpenAI president Greg Brockman said the company is still investigating and is examining every part of its pipeline to determine an appropriate response.
The attack targeted Hugging Face, an online platform for open-source AI models and datasets. Hugging Face disclosed in a July 16 blog post that an autonomous AI agent had breached its systems earlier that week, and OpenAI confirmed on July 21 that its models were involved. The OpenAI blog noted that the incident used a combination of an unnamed unreleased model and the publicly available GPT-5.6 Sol, but did not detail how the models interacted or how internal controls failed.
AI safety researchers have issued a list of unanswered questions, including which models were used, the assigned task, how the model exited OpenAI’s environment, why it targeted Hugging Face, and whether the public model supply chain was altered. Redwood Research’s chief scientist and cybersecurity firm Penligent have both highlighted gaps in the information released.
Michele Catasta, president of AI at Replit, warned that understanding the Hugging Face breach is critical for public safety and the broader AI industry, noting that similar events could become more common.
"We need to get ready, the entire industry, for this to happen more," Catasta said.
OpenAI has not provided a timeline for the release of its technical report.