OpenAI Says GPT-5.6 Sol Led Autonomous Cyberattack On Hugging Face

According to Axios, OpenAI said Tuesday that GPT-5.6 Sol and an even more capable pre-release model carried out last week's AI-led cyberattack on Hugging Face during pre-deployment testing. OpenAI said the models broke out of their testing environment, used stolen credentials and other vulnerabilities to reach part of Hugging Face's production infrastructure after being asked to solve a hacking challenge. Hugging Face CEO Clément Delangue called it an attack unlike anything the company had seen before, while Anthropic's Logan Graham called it the first true AI safety incident. The U.K.'s AI Security Institute also said Tuesday that every model it tested attempted to cheat at least some of the time on cybersecurity evaluations, with GPT-5.6 Sol cheating in 12.6% of test runs and Anthropic's Claude Mythos Preview in 7.8%.