People
Updated 04/02/26Co-Founder
Funding Details
Updated 04/02/26- Annual Budget
- -
- Current Runway
- -
- Funding Goal
- $500,000
- Funding Raised to Date
- $190,000
Org Details
Updated 04/02/26Luthien is a Seattle, WA-based nonprofit public benefit corporation founded in September 2024 to translate AI control research into production-ready tools. The organization builds on the foundational AI control research published by Redwood Research in "AI Control: Improving Safety Despite Intentional Subversion" (Greenblatt et al., 2023), which demonstrates that safe outcomes can be achieved even under the pessimistic assumption that AI models may behave adversarially. Luthien was founded by Jai Dhyani, a former Meta ML software engineer and MATS alumnus with over a decade of experience training and deploying AI models at scales of billions of monthly active users. Dhyani is also a co-author on METR's "RE-Bench: Evaluating frontier AI R&D capabilities of language model agents against human experts" (2024). Scott Wofford serves as co-founder. Luthien is the only AI safety organization headquartered in the Pacific Northwest and operates out of CoMotion Labs at the University of Washington. The organization's flagship project is the Luthien Proxy, an open-source OpenAI-compatible LLM gateway that brings Redwood-style AI control to production agentic deployments. Luthien is iterating on this proof-of-concept system and coordinating with AI control experts at research and frontier labs to develop a production-ready solution. The organization has published research on prosaic AI control and presented at ControlConf London in March 2025. Luthien received early support from Impact Academy, the Survival and Flourishing Fund, Entrepreneur First's def/acc, and Catalyze Impact. Within roughly two months of founding, the organization raised approximately $190,000 through Catalyze's Seed Funding Circle. It subsequently received a $150,000 seed grant from the AI Safety Tactical Opportunities Fund (AISTOF). The organization is targeting $500,000 to grow the team with additional software engineers and a product manager to manage partnerships. Donations are accepted through Manifund.
Theory of Change
Updated 04/02/26Luthien's theory of change holds that current approaches to AI safety are insufficient because they assume models will behave as intended. By instead assuming adversarial behavior and building oversight infrastructure that demonstrably prevents misaligned systems from achieving their goals, safety can be guaranteed even if alignment fails. Luthien's role is to take this theoretical insight from Redwood Research and make it practically deployable: by building open-source tooling like the Luthien Proxy, they lower the cost for organizations to adopt control measures. Widespread adoption of AI control infrastructure in real-world production environments reduces the probability that a misaligned frontier AI system could cause catastrophic harm, creating a direct causal path from their engineering work to reduced existential risk from AI.
Grants Received– no grants recorded
Updated 04/02/26Projects– no linked projects
Updated 04/02/26Discussion
No comments yet. Be the first to share your thoughts.