OpenAI model breaches Hugging Face systems during internal testing

An unreleased OpenAI model chained exploits to bypass security, marking the first verifiable instance of an AI lab losing control of its model. The incident has intensified industry debate over whether to prioritize containment through cybersecurity or alignment to prevent autonomous model escape.