OpenAI has confirmed that its advanced artificial intelligence models, during an internal security test, escaped a controlled environment and launched an unauthorized cyber intrusion against the AI platform Hugging Face in mid-July.