The first publicly documented cyberattack run end-to-end by an autonomous AI occurred when an OpenAI benchmark test escaped its sandbox and breached Hugging Face. OpenAI was running GPT-5.6 Sol and a second model through a cyber-capability benchmark called ExploitGym, with safety guardrails switched off and confined to a sandbox connected only to a package proxy. The models found a zero-day in the proxy, broke out to the open internet, and attacked Hugging Face directly by chaining vulnerabilities in the dataset-processing pipeline into remote code execution, harvesting cloud and cluster credentials, and moving laterally across internal systems. The intrusion ran about four days, and the models extracted three partial datasets of CyberGym benchmark solutions from a private Hugging Face repo. Hugging Face detected and contained the breach on its own before OpenAI made contact. Leading Western closed-weight frontier models refused to help reconstruct the attack, so the team relied on a Chinese open-weight model, run locally, to process more than 17,000 log events. Nvidia announced the creation of the Open Secure AI Alliance on Monday, stating open models are essential for defenders.