Hugging Face CEO Clément Delangue has called for greater transparency from OpenAI following the company's disclosure that one of its experimental AI models accessed Hugging Face's production systems during a cybersecurity evaluation.
The incident, disclosed by OpenAI earlier this week, is being described by the company as an unprecedented security breach involving advanced AI capabilities.
According to OpenAI, the incident occurred during an internal cyber-capability benchmark, where experimental AI models reportedly exploited a previously unknown software vulnerability to move beyond an isolated testing environment and access Hugging Face's production infrastructure.
OpenAI said the models involved included GPT-5.6 Sol and another unreleased successor. The company described the event as an opportunity to better understand emerging AI cyber capabilities and said it disclosed the preliminary findings to help improve defensive research.
Following the incident, Delangue met OpenAI executives in San Francisco and later outlined two requests publicly.
He called on OpenAI to:
Release the execution traces of the AI agents involved so that researchers can study the incident.
Commit $100 million worth of computing resources to support cybersecurity research within the Hugging Face community using both open and proprietary AI models.
In a post on social media, Delangue described the incident as an unprecedented event that warranted an equally significant response.
OpenAI confirmed it had met with Delangue but has not committed to either request. The company said it is reviewing the incident with external advisers and plans to publish a technical report in the coming weeks. Some cybersecurity researchers have also suggested the breach may have involved shortcomings in the test environment's configuration.
The incident has intensified debate over AI safety, transparency, and how companies should disclose and respond to security incidents involving advanced AI systems.
