FAR.AI, founded in 2022, conducts frontier AI alignment research in model evaluation, interpretability, robustness and value alignment, with bases in Berkeley and global policy activity.
FAR.AI is a research organisation founded in 2022 that works on frontier AI alignment: the problem of ensuring that advanced AI systems remain safe and beneficial. Its technical work spans model evaluation, interpretability, robustness, value alignment, AI security and red-teaming. The organisation deliberately pursues a diverse portfolio of high-potential research agendas rather than betting on a single approach, and it operates across the boundary between academic research and for-profit initiatives.
The organisation is structured around three units. FAR.Research conducts frontier alignment research; FAR.Labs is a Berkeley coworking space that brings together researchers, policymakers and industry leaders; and FAR.Futures runs global events and policy initiatives with the same collaborative aim. FAR.AI has its base in Berkeley, with activity extending globally, and it works with stakeholders in AI safety, AI policy, frontier AI model development and government AI safety institutes.
Since July 2022 the group has produced more than 40 academic papers, published at venues including NeurIPS, ICML and ICLR. Its stated orientation is mission-driven, focused on existential risk reduction and what it describes as humanity's most urgent challenges. Further information on the organisation and its work is available at https://far.ai/.
