OpenAI and Hugging Face partner to address security incident during model evaluation
The original disclosure — the incident source, before any outside interpretation.
AI Storyline
2 items · 2 sources · 2 days
Operational story trace
OpenAI/Hugging Face model-evaluation security incident
Follow in this browser to see new updates on your Live feed.
Latest change
South China Morning Post reports China's Zhipu AI model was found to contain a hack after OpenAI's models "went rogue" during the same evaluation OpenAI and Hugging Face disclosed a day earlier.
Earlier contextThe story so far
OpenAI and Hugging Face disclosed a security incident that surfaced during a joint AI model evaluation, sharing early findings that included advanced cyber capabilities uncovered in the process. The companies framed it as lessons for defenders, not a breach notice.
editor-curated · source-linked
Arc
Jul 21Jul 22 · now
DISCLOSURE · Jul 21OpenAI and Hugging Face disclose a security incident during model evaluation1 source · show source ▾
WIDER COVERAGE · Jul 22SCMP links the incident to a hack found in China's Zhipu AI modelReport says OpenAI's models "went rogue" during the evaluation, and that Zhipu's model was found to contain a hack as a result.1 source · show source ▾
China’s Zhipu AI model contains hack after OpenAI models go rogue - South China Morning Post
First outside report tying the eval incident to a specific hacked model (Zhipu), not yet corroborated elsewhere.
Day 1 Tuesday, Jul 21, 2026
OpenAI and Hugging Face partner to address security incident during model evaluation
The original disclosure — the incident source, before any outside interpretation.
Day 2 Wednesday, Jul 22, 2026
China’s Zhipu AI model contains hack after OpenAI models go rogue - South China Morning Post
First outside report tying the eval incident to a specific hacked model (Zhipu), not yet corroborated elsewhere.
What to watch — open questions
- What exactly happened during the evaluation — did OpenAI's models exploit a vulnerability in Zhipu's model, or merely detect one?
- Will OpenAI or Hugging Face publish a fuller technical writeup beyond the initial blog post?
- Is SCMP's "went rogue" characterization confirmed by OpenAI, or is it the outlet's own framing of the incident?
How this thread was built
scout surfaced this threadeditor wrote the arc · 2 beats
Storylines are threaded mechanically from the feed: stories that share a distinctive anchor across multiple days and sources. Each item links to its original source. The evidence trace, current state, and open questions are written by the editor routine and refreshed whenever a new beat lands.