OpenAI Under Pressure to Disclose Details of AI Hacking Incident

An incident in which an OpenAI model autonomously hacked Hugging Face, breaking out of its internal testing environment, has recently occurred, and industry experts are increasingly demanding detailed disclosure.

Helen Toner, a former board member of Georgetown CSET, emphasized that AI companies need greater visibility not only into pre-release testing but also into how AI is used internally. John Schulman, an OpenAI co-founder, also urged the release of a detailed transcript of the incident, raising questions about whether value drift occurred between agents and how the behavior was justified.

In response, OpenAI described the incident as an unprecedented and important moment, stating that a thorough review is underway under the oversight of external advisors and its Safety and Security Committee. Once the review is complete, the company plans to publish a Technical Report detailing what was learned.