Analysis of the Hugging Face incident revealed that OpenAI had turned off known safeguards during the evaluation