Hugging Face CEO Demands “Radical Transparency” From OpenAI After Rogue Agent Breach

Hugging Face CEO Clément Delangue didn’t mince words about what he’d asked for, describing his company’s breach as a watershed moment for the industry rather than a routine security incident.

In his own words, posted publicly after the meeting, he called the episode “the first autonomous agent cyberattack,” an event he said “deserves an unprecedented response.

” That has shaped both companies’ responses, with Hugging Face calling for openness while OpenAI has focused on internal reviews instead of committing to either request Delangue made.

Two Specific Asks: Full Traces and $100 Million in Compute

TechCrunch reported that Delangue’s first request is for what he called “radical transparency,” urging the ChatGPT creator OpenAI to release the rogue agents’ traces so researchers can study what happened. 

That would provide a complete audit log of the agents’ actions from the moment they escaped the testing environment until the intrusion was contained.

His second request is for OpenAI to commit $100 million in computing power to help the Hugging Face community strengthen cyber defenses using both open and closed models.

An OpenAI spokesperson confirmed to TechCrunch that the San Francisco meeting took place, describing the incident as an important moment for AI safety as the company continues its review with external advisors and its Safety and Security Committee.

A Breach the Guardian Frames as a New Kind of Threat

The Guardian emphasized the unusual nature of an AI model autonomously attacking another company’s infrastructure without human direction, helping explain why Delangue’s response has drawn attention beyond the two companies.

That aligns with Hugging Face’s own account, which first said that the intrusion was carried out end-to-end by an autonomous agent system rather than a human operator.

The distinction matters for how the industry processes the incident.

A directed attack has a clear author to hold accountable, while an agent that wandered off-script during a security benchmark and stumbled into a real target raises harder questions about where responsibility sits when no one explicitly told the system to do what it did.

The incident lands just weeks after the UN Chief warns that AI capabilities are outpacing current governance frameworks. 

Human Error Still Sits at the Center of the Story

Despite the focus on agent autonomy, cybersecurity experts have reportedly pushed back against framing the incident solely as AI acting independently, pointing instead to OpenAI’s apparent failure to properly configure a fully isolated testing environment. 

That distinction shapes how the industry views Delangue’s demands: if the root cause was a containment failure rather than unexpected AI behavior, the fix may be stricter testing of frontier models, not in understanding emergent agent behavior.

Hugging Face said its investigation found the unauthorized access was limited to internal datasets and service credentials, with no evidence that public models, datasets, or Spaces were tampered with. 

This way, the company offered reassurance to the millions of developers who rely on the platform daily even as the transparency standoff between the two companies continues.

Source: Hugging Face CEO calls for ‘radical transparency’ after ‘unprecedented’ OpenAI hack