Alignment Research Center
An AI safety research organization founded by Paul Christiano that focuses on evaluating dangerous capabilities in frontier AI models, including potential risk behaviors such as self-replication and evasion of human oversight.
Core Facts
METR (Model Evaluation & Threat Research) is an independent third-party organization focused on safety evaluation of frontier AI systems.
METR is an organization focused on capability evaluation and risk research for frontier AI models, with the full name Model Evaluation and Threat Research.
Timeline (last 90 days)
Paul Christiano founded the Alignment Research Center (ARC).
ARC's capability evaluation paradigm was later adopted by Anthropic's Responsible Scaling Policy and OpenAI's Preparedness Framework.
After leaving OpenAI, Christiano founded the Alignment Research Center (ARC), focusing on evaluating dangerous capabilities that frontier AI models may pose.
ARC developed a model evaluation (evals) framework that probes whether models exhibit tendencies toward dangerous behavior by designing specific test scenarios.
METR is an organization focused on capability evaluation and risk research for frontier AI models, with the full name Model Evaluation and Threat Research.
All Facts (6)
METR (Model Evaluation & Threat Research) is an independent third-party organization focused on safety evaluation of frontier AI systems.
90%VerifiedMETR is an organization focused on capability evaluation and risk research for frontier AI models, with the full name Model Evaluation and Threat Research.
75%UnverifiedPaul Christiano founded the Alignment Research Center (ARC).
50%UnverifiedARC's capability evaluation paradigm was later adopted by Anthropic's Responsible Scaling Policy and OpenAI's Preparedness Framework.
50%UnverifiedAfter leaving OpenAI, Christiano founded the Alignment Research Center (ARC), focusing on evaluating dangerous capabilities that frontier AI models may pose.
50%UnverifiedARC developed a model evaluation (evals) framework that probes whether models exhibit tendencies toward dangerous behavior by designing specific test scenarios.
50%