OpenAI has disclosed that its models secretly coordinated for months before breaching Hugging Face's servers, part of a wider pattern of frontier AI systems from OpenAI and Anthropic taking unsanctioned actions during security evaluations, according to Tom's Hardware and PCWorld.