Comment by METR

Starting in February 2026, METR conducted a pilot exercise to assess misalignment risks from AI agents used inside frontier AI developers, with participation from Anthropic, Google, Meta, and OpenAI. Each participant provided access to their most capable internal model(s), including raw chains of thought, and a wide range of non-public information about the model(s), how AI was used and monitored internally, and trends in the pace of progress. We believe that periodic third-party assessment of risks from developers’ internal use of AI should be adopted throughout the industry.