Machine Alignment Transparency and Security (MATS)

MATS Research Inc. is the US 501(c)(3) nonprofit behind the ML Alignment & Theory Scholars (MATS) program, the largest AI safety research fellowship and talent pipeline. The program was originally launched in late 2021 as SERI MATS, a joint initiative of the Stanford Existential Risks Initiative and the Berkeley Existential Risk Initiative. It incorporated as an independent nonprofit in 2024, having previously been fiscally sponsored by BERI.

MATS runs twice-yearly 12-week intensive research fellowships in Berkeley, California and London, UK. Each cohort pairs approximately 100-120 fellows with leading mentors from organizations including Anthropic, Google DeepMind, OpenAI, MIRI, ARC, and others. Fellows receive a $15,000 stipend, a $12,000 compute budget, housing, catered meals, office space, travel support, and weekly one-on-one research management. At the conclusion of the research phase, fellows present their work at a Research Symposium, and approximately 75% continue into a fully funded 6-12 month extension program.

The program is led by Co-Executive Directors Ryan Kidd and Christian Smith, supported by a team of approximately 44 staff across research management, operations, community, compute infrastructure, and London extension teams. Research managers meet weekly with scholars and mentors to help with research strategy and project coordination.

Since its founding, MATS has supported over 446 researchers and 75 mentors. Alumni have produced more than 170 research publications with over 9,500 collective citations and an organizational h-index of 44. Approximately 80% of alumni now work directly in AI safety and security, and about 10% have co-founded AI safety organizations. Notable alumni outcomes include placements at Anthropic, OpenAI, MIRI, ARC Evals, and CHAI, as well as founding organizations such as Apollo Research, Leap Labs, Timaeus, ARENA, Cadenza Labs, Center for AI Policy, and Stake Out AI.

MATS fellows work across a wide range of research tracks including mechanistic interpretability, technical governance, empirical safety research, policy and strategy, theory, and compute governance. Research contributions from MATS fellows include sparse auto-encoders for AI interpretability, activation and representation engineering, emergent misalignment, developmental interpretability, and gradient routing, among many others.

Major funders include Coefficient Giving, Open Philanthropy (via Good Ventures), the Survival and Flourishing Fund, BERI, and Founders Pledge. The organization received approximately $6.1 million in donations in 2024 and approximately $20.7 million in 2025, reflecting its rapid growth. MATS is actively scaling its programs, with plans to expand to 120 fellows per cohort and launch additional fellowship cycles and residency programs for senior researchers.