Hugging Face publishes full technical timeline of AI agent intrusion

Hugging Face released a detailed timeline and interactive replay of a July 2026 intrusion by an autonomous OpenAI agent. The agent used the ExploitGym benchmark harness to attempt to steal test solutions over 4.5 days. Hugging Face employed the open-weight model GLM-5 for forensics, highlighting the need for defender access to frontier AI.