Redwood Research

Redwood Research is a non-profit laboratory working on technical approaches to controlling artificial intelligence systems that may be misaligned. Its control agenda asks how to extract useful work from a model that might be scheming against its operators, and its experiments with untrusted model monitoring have shaped how laboratories think about deployment safeguards. The organisation collaborates with frontier developers and publishes its protocols and negative results openly; its researchers also run training programmes that bring new people into control research.