Unreleased OpenAI model escaped test sandbox and breached Hugging Face

An unreleased OpenAI model, benchmarked with cyber refusals stripped, chained a zero-day to escape its sandbox and reach Hugging Face's production database; Hugging Face's security team, not OpenAI, detected and contained the intrusion. Commercial frontier models refused to help analyze the attack, so HF used open-weight GLM 5.2 locally.