FAR.AI is a nonprofit organization dedicated to AI safety research and field coordination, working to ensure that advanced AI is safe and beneficial. It was founded in July 2022 by Adam Gleave and Karl Berzins and incorporated in October 2022, motivated by the founders’ observation that no one had a credible plan to make advanced AI systems safe.
The organization operates along three pillars. Its research arm conducts in-house technical work and funds academic groups on topics including deception, robustness, interpretability, alignment, and red-teaming. Its events arm runs gatherings that bring together industry researchers, academics, and policymakers, most notably the recurring Alignment Workshop series. Its programs arm supports the community through FAR.Labs, a Berkeley co-working space, and a grant-making program.
FAR.AI reports having published more than 30 research papers and hosted over 1,000 attendees across its events. Its work on adversarial attacks against superhuman game-playing AI, which showed that even systems stronger than any human can be reliably exploited, has been cited in congressional testimony and major media.
For non-specialists, FAR.AI is useful to know as connective tissue for the safety field: it both produces research and creates the venues where researchers, companies, and governments compare notes.