METR Publishes First Frontier Risk Report — Tests Internal Models From All Four Major Labs, Finds No Dramatic AI R&D Speed-Ups

METR·high signal

METR released its first comprehensive Frontier Risk Report on May 19, testing the best internal models from Anthropic, Google, Meta, and OpenAI with Chain-of-Thought access and reviewing non-public information about capabilities and alignment. Key finding: companies did not report evidence of dramatic speed-ups in AI R&D from automation, and Anthropic explicitly stated they had not seen a 2x increase in pace of progress as of April 2026. The report also flagged that a large fraction of agent activity was not reviewed by any human, and automated monitoring systems were not universally applied.