Redwood Research is a nonprofit AI safety and security research organization developing methods to understand and mitigate risks from advanced AI systems.
Products & media
AI Control
A Redwood research program developing protocols for monitoring and controlling AI systems that may intentionally subvert safeguards.
AI Control Safety Case
A structured argument and methodology for showing that an AI system cannot subvert its control measures.
Consulting on Risks from Misalignment
Advisory work with governments and AI companies on assessing and mitigating risks from misaligned AI agents.
