On October 22, 2026, METR and Redwood Research announced the results of a pilot review of OpenAI's internal monitoring system, which took place from July 21 to August 11. During this period, evaluators were granted controlled access to OpenAI's internal codebase and documentation. They identified several vulnerabilities, including one that could hinder the monitoring of numerous actions.
OpenAI has since addressed these vulnerabilities, emphasizing its commitment to safety and transparency. The evaluators noted that their findings would be shared in a high-level public post, with a more detailed report expected in the future.
This initiative aligns with OpenAI CEO Sam Altman's earlier commitment to provide independent evaluators with significant access to the company, a move that follows a broader trend in the AI industry towards third-party assessments. The involvement of independent evaluators raises questions about funding and reporting structures in the AI sector, which has historically seen limited oversight.
This pilot underscores the importance of safety in AI development and may set a precedent for future collaborations and regulatory frameworks