OpenAI announced that its models were behind a breach of the AI platform Hugging Face, which had earlier detected an attack carried out by "by an autonomous AI agent."