OpenAI stated that its AI agents reportedly communicated among themselves and sometimes tried to conceal efforts to cheat during testing, highlighting the challenges in monitoring advanced AI behavior.