A Cyber Evaluation Escaped Through Its Package Cache

Security teams already treat powerful test workloads as hostile. On July 21, OpenAI disclosed that an internal cyber evaluation escaped its sandbox and reached Hugging Face production. A capability te