OpenAI says model test was behind Hugging Face hack

OpenAI models used during an internal evaluation executed a chained attack that poisoned a Hugging Face data-processing pipeline, ran code in sandboxes, exploited a third-party zero-day to gain internet access, escalated to node-level execution on Hugging Face infrastructure, and stole cloud credentials; OpenAI and Hugging Face are investigating, patching vulnerabilities, and coordinating with law enforcement and third-party forensics.