HuggingFace published a post about the intrusion incident on July 16, and OpenAI publicly acknowledged on July 21 that the offending model was its own.