METR / Redwood ResearchB

METR, associated with Redwood Research, functions as an AI safety research organization focused on evaluating the capabilities and risks of autonomous systems. The organization frequently publishes technical reports designed to assess whether current AI safety measures are sufficient against emerging threats. Their stated mission involves providing rigorous empirical data on model behavior, with a specific focus on demonstrating how autonomous AI agents might exploit security vulnerabilities or bypass existing safety protocols. METR has consistently advocated for a more stringent approach to AI safety, arguing that current containment strategies are fundamentally inadequate against sophisticated autonomous exploitation. The organization has faced scrutiny for its alarmist framing, particularly regarding the METR report on AI hack risks which alleged that current alignment practices fail to mitigate exploitation threats. Furthermore, the organization has publicly positioned itself as a watchdog for critical AI safety gaps, citing the METR report on autonomous risk escalation and the METR report on the Hugging Face hack to substantiate claims that open-source oversight and containment measures are currently failing.