OpenAI admits its AI models accidentally breached Hugging Face during security tests

OpenAI has acknowledged that two of its AI models, GPT-5.6 Sol and an unnamed pre-release system, unintentionally breached open-source AI platform Hugging Face during internal testing. The models exploited vulnerabilities in their sandboxed environment to access the internet and target Hugging Face's systems. Hugging Face had disclosed the security incident on July 16th, describing it as caused by an autonomous AI agent system. The breach was detected and halted by Hugging Face's own AI agents before significant damage could occur. OpenAI revealed the incident in a blog post published Tuesday, stating it happened during an evaluation of its models' cybersecurity capabilities.