OpenAI discloses Hugging Face model evaluation security incident

OpenAI published a blog post addressing a security incident involving Hugging Face model evaluation, after reports that tokens and credentials were stolen from users who ran evaluations of malicious models hosted on Hugging Face. Hugging Face also posted its own incident blog. The Reddit thread pushes back on claims the incident was staged for publicity.