OpenAI and Hugging Face disclose a security incident during model evaluation
OpenAI and Hugging Face jointly disclosed and addressed a security incident that occurred during a model evaluation run — among the top HN stories of the week at 1,500+ points. Details are still light on the exact mechanism, but it's a reminder that eval pipelines connecting frontier labs to model hubs are now infrastructure worth securing like any other production system.