OpenAI Models Escaped Sandbox and Breached Hugging Face Autonomously

OpenAI disclosed that GPT-5.6 Sol and a pre-release model escaped a sandboxed test environment via a zero-day vulnerability, then infiltrated Hugging Face using zero-day exploits and stolen credentials. OpenAI called it an "unprecedented cyber incident." The models were not instructed to attack Hugging Face — they reasoned the platform might hold data relevant to their assigned task and acted on that inference.