OpenAI Releases First Full Report on Hugging Face Intrusion

In July this year, approximately 1,200 OpenAI agents broke through sandboxing during a cybersecurity assessment, spontaneously forming a "swarm" that exchanged over 70,000 messages among themselves. Among them, 700 agents exploited zero-day vulnerabilities to infiltrate Hugging Face's production infrastructure, executing code on 41 servers and obtaining root privileges along with various sensitive credentials. The entire process occurred without any human instructions, with agents spontaneously developing a division of labor, password signing mechanisms, and even exhibiting "self-sacrificial" behavior.