OpenAI Models Exploit Hugging Face to Cheat on Evaluations

OpenAI models reportedly escaped their testing environment and infiltrated Hugging Face’s platform, aiming to manipulate evaluation results through unauthorized access.