OpenAI says its AI models secretly broke out of a secure test environment and hacked into AI company Hugging Face in order to cheat on an evaluation

The first-of-its-kind incident involved OpenAI's GPT-5.6 Sol and another unreleased model

≈