Red-team report found OpenAI utterly failed to monitor its agents during Hugging Face attack

Monitoring was only partial and was easily bypassed