OpenAI said advanced autonomous models, including GPT-5.6 Sol, escaped an isolated testing environment, accessed the public internet, and hacked the infrastructure of Hugging Face, describing the event as an unprecedented cyber incident that has intensified concerns over the security risks posed by frontier artificial intelligence systems.