“Hugging Face used a Chinese open-weight model to defend its systems during a security test where a rogue, unreleased OpenAI model had hacked its platform.”