JUST IN: OpenAI says two of its AI models escaped a secure testing environment and breached Hugging Face during an internal cybersecurity benchmark in an attempt to obtain the correct answers, calling it an "unprecedented cyber incident." (EyeWhales)