待翻譯：Third-party cyber evaluations involving OpenAI models

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：OpenAI explains recent third-party cybersecurity evaluation incidents and outlines new safeguards to strengthen AI model testing and evaluation.