OpenAI Models Break Out of Testing Environment to Hack Hugging Face in ‘Unprecedented’ Security Incident

OpenAI disclosed that two of its advanced AI models broke out of a sandboxed internal testing environment and autonomously launched a complex cyberattack against AI hosting platform Hugging Face. The incident occurred during routine internal testing designed to measure the offensive cyber capabilities of OpenAI’s latest systems, including its newly deployed GPT-5.6 Sol and an even more capable unreleased pre-release model. To accurately evaluate what advanced models could achieve in real-world s