Field builder · Funders · Berkeley, California · Est. 2023
MATS (matsprogram.org) is an independent U.S. 501(c)(3) research and educational seminar organization focused on building the AI safety talent pipeline through mentored fellowships and longer-horizon research support. The program’s core model pairs fellows and residents with “top mentors” across AI alignment, interpretability, governance, and security, while providing structured research management, compute/stipend/housing support, and programming intended to help researchers produce publishable and durable research outputs.
MATS runs multiple fellowship-style cohorts (e.g., Summer 2026 and Autumn 2026) and a longer-duration “Residency” program that funds independent agendas for 6–24 months. Program tracks and streams cover both technical and policy/security themes, including an explicit Founding & Field-Building track (and—during Autumn 2026—the Biosecurity track), with the stated aim of strengthening field capacity by both training researchers and supporting high-agency ecosystem builders. From an institutional standpoint, MATS functions as a “field-builder” by concentrating mentoring talent, logistics, and financing into time-bounded research cohorts, then connecting alumni into labs, nonprofits, and academic groups in AI alignment, transparency, and security. MATS also publicly reports select donation categories and values via its financial transparency page, and it maintains an internal research/risk governance posture (e.g., fellowship agreements and confidentiality / pre-publication review for dual-use risks are referenced on program materials).
Towards Predictive Models of Strategic Behaviour in Large Language Model Agents research by Jennifer Nzambi, Ares Panos, Aug 31, 2026. Explore AI alignment and safety research from the MATS Program.
Non-Great-Power Conflict and AI RiskNon-Great-Power Conflict and AI Risk research by Kristina Kempkey, Aug 31, 2026. Explore AI alignment and safety research from the MATS Program.
Studying Coordination and Collusion in Multi-Agent LLM Code ReviewsStudying Coordination and Collusion in Multi-Agent LLM Code Reviews research by Jennifer Nzambi, Ares Panos, Roger Dearnaley, Aug 31, 2026. Explore AI alignment and safety research from the MATS Program.
Think Fast: Estimating No-CoT Task-Completion Time Horizons of Frontier AI ModelsThink Fast: Estimating No-CoT Task-Completion Time Horizons of Frontier AI Models research by Alex Serrano, Jo Jiao, Twm Stone, Ariana Azarbal, William L. Anderson, Aug 24, 2026. Explore AI alignment and safety research from the MATS Program.
MATS updated its careers page (updated Sep 4, 2026) listing open roles including General Counsel, Research Manager, (Senior) Program Manager (Mentor Selection), Program Coordinator (Main Program), Impact Analyst, Community Manager, Head of Finance, and executive support roles.
MATS Autumn 2026: 10-week program dates and track expansion (Founding & Field-Building and Biosecurity)MATS announced the Autumn 2026 program will run September 28 to December 4 in Berkeley and London, with option for a 6–12 month funded extension, and noted that (for the first time) Founding & Field-Building and Biosecurity tracks are being run.
MATS Summer 2026: stated cohort scale (120 fellows, 100 mentors) and application timingMATS described Summer 2026 as its largest program to date, with 120 fellows and 100 mentors, and stated the application deadline as Sep 6.
MATS Residency: application outcome timeline and intake window (Aug 25–Oct 31, 2026)MATS stated that Residency applications for its next intake window are open Aug 25 to Oct 31, 2026 (AoE), with applicants hearing outcomes by end of November 2026 and shortlisted candidates invited to interviews or paid work tests early December 2026.
MATS Fellowship outcomes/scale claims (Winter 2027 application page)MATS’ Winter 2027 application page includes program scale/outcome claims such as 631 researchers participating “since late 2021,” 220+ research publications, and other citation/impact-style metrics.
Same Question, Different Lies: Cross-Context Consistency (C³) for Black-Box Sandbagging Detection research by Yulong Lin, Benjamin Arnav, Aug 20, 2026. Explore AI alignment and safety research from the MATS Program.