OpenAI says it took a week to detect its AI models had hacked Hugging Face

Start-up says AI agents communicated among themselves and sometimes tried to conceal efforts to cheat during testing