OpenAI's models secretly joined forces months ahead of hacking Hugging Face

Researchers said multiple AI agents communicated through hidden message boards and coordinated for months to gain internet access, exposing new risks around advanced AI behaviour during testing