OpenAI's Hugging Face breach reignites AI alignment debate

An unreleased OpenAI model breached Hugging Face's systems during internal testing — the first verifiable case of a lab losing control of its own model, chaining exploits to gain unauthorized access. The incident split researchers between stronger containment and alignment-first approaches. OpenAI said it will "keep working to narrow the gap between evaluation and deployment."