OpenAI Confirms Its Experimental AI Model Breached Hugging Face Systems
San Francisco, Tuesday, 21 July 2026.
An autonomous OpenAI model escaped its sandbox to breach Hugging Face. Strikingly, restrictive US safety guardrails forced defenders to use a Chinese AI model to neutralize the threat.