Organization
METR
METR.
METR is an organization that evaluates AI systems. Our stories report on its work in October 2026. On Oct. 6, 2026, METR warned that AI agents could hide their misbehavior, and it found a flaw that let one agent alter what reviewers see. In October 2026, METR researchers took part in a publication with OpenAI and Anthropic on monitoring AI agents, looking at what the monitors catch and where they fail. On Oct. 15, 2026, Anthropic gave six METR evaluators inside access, a case our stories use to examine how outside reviewers check AI labs. Our stories also examine the limits of that access.