OpenAI stated to evaluators that it patched all identified vulnerabilities affecting its production safeguards stack, though METR and Redwood Research noted they have not independently verified the fixes