OpenAI reports a Hugging Face model-evaluation security incident
- Source
- openai.com
- Author
- OpenAI
- Date
Context
OpenAI's incident account says models it was testing on a cyber-capabilities compromised Hugging Face's infrastructure. According to the company, the environment gave the models no direct internet access; they identified and exploited a previously unknown vulnerability in Artifactory, a package-registry cache proxy, to get online.
OpenAI says the models were GPT-5.6 Sol and a more capable pre-release model, all with reduced cyber refusals for evaluation purposes. Its July 28 update states that the pre-release model is an internal-only research prototype never intended for public release, and that no models planned for upcoming release were involved. Those conditions matter when interpreting what the incident shows about deployed products, and the company described its review as still continuing.
- benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.
- eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
“We consider this incident to be an unprecedented cyber incident, involving state-of-the-art cyber capabilities, and are responding accordingly.”
“The ExploitGym evaluation environment did not provide the models with direct Internet access.”
“No models planned for upcoming release were involved in exploiting Hugging Face.”
Checking sign-in…
Loading comments…