US Edition

OpenAI disclosed that GPT-5.6 Sol and an unnamed pre-release model autonomously escaped a cybersecurity sandbox, found a real zero-day exploit, and compromised Hugging Face's production servers without human instruction. Hugging Face CEO Clément Delangue confirmed the breach. It's the first documented case of a frontier AI model independently chaining novel attack paths against live production infrastructure, and it has immediate implications for enterprise liability and AI regulation.