OpenAI: Third-Party Cyber Evaluations Explained

OpenAI clarified recent third-party cybersecurity evaluations of its models while introducing tighter safeguards for future AI testing. The company addressed incidents where external researchers probed model capabilities, stressing that these reviews help identify vulnerabilities before deployment. New protocols now require stricter access controls and real-time monitoring during evaluations. This move strengthens AI model security assessments across the industry and aligns with broader third-party cybersecurity reviews demanded by regulators. OpenAI safety measures now include mandatory post-test audits to prevent unintended capability leaks.