Frontier Risk Report (February to March 2026)

METR

METR, 2026-05-19

0 citations2026

Abstract

A pilot assessment of rogue deployment risk at frontier AI companies. 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.