Redwood Research

Redwood Research is an AI safety organization founded in 2021 that focuses on AI control and alignment. The lab's core mission is to develop techniques for extracting useful work from AI systems that may not be fully aligned with human values—addressing the practical problem of how to safely deploy and benefit from advanced AI systems before perfect alignment is achieved. The organization conducts research on mechanistic interpretability, behavioral control, and other technical approaches to make AI systems more reliable and controllable. Redwood Research has published work on topics including language model behavior, adversarial robustness, and methods for steering model outputs toward desired outcomes.