Field builder · Funders · Berkeley, CA
MATS (Machine Alignment, Transparency, and Security) is an independent research and educational seminar program focused on growing the AI safety and security research field by funding and mentoring early-career researchers and field-builders. MATS runs multi-track research fellowships in Berkeley (California) and London (UK), pairing fellows with top mentors across AI alignment/transparency/governance/security and related areas, providing research management, and offering compute, stipend, housing, and travel support.
In addition to short, intensive fellowship cohorts (e.g., 10–12 week programs plus an optional funded extension), MATS also operates a longer-horizon MATS Residency that provides 6–24 months of independence and resources for researchers working on longer-term AI safety agendas. MATS reports operating as a US 501(c)(3) public charity with a separate UK entity (a company limited by guarantee), and publishes financial transparency disclosures listing named donors and donation dates for post-incorporation periods.
No curated programs yet. Check the careers page ↗.
A Pre-mortem on The First Major AI-Enabled Terrorism Event and The Choices That Could Enable it: Plus 9 Predictions research by Max Kamachee, Jonas Raedler, Oct 20, 2026. Explore AI alignment and safety research from the MATS Program.
Gradient Routing Localizes and Removes Unintended Behaviors in RLGradient Routing Localizes and Removes Unintended Behaviors in RL research by Jake Ward, Shawn Hu, Aria Wong, Oct 19, 2026. Explore AI alignment and safety research from the MATS Program.
Can Power Draw Constrain Covert Compute? Limits of Analogue Verification for AI GovernanceCan Power Draw Constrain Covert Compute? Limits of Analogue Verification for AI Governance research by Tom Kimpson, Emlyn Graham, Oct 18, 2026. Explore AI alignment and safety research from the MATS Program.
Adaptive Attacks on Trusted Monitors Subvert AI Control ProtocolsAdaptive Attacks on Trusted Monitors Subvert AI Control Protocols research by Alexander Panfilov, Oct 06, 2026. Explore AI alignment and safety research from the MATS Program.
针对可信监控器的自适应攻击会破坏 AI 控制协议针对可信监控器的自适应攻击会破坏 AI 控制协议 research by Alexander Panfilov, Oct 10, 2026. Explore AI alignment and safety research from the MATS Program.
AI 研究人员如何看待 AI 研发自动化与智能爆发AI 研究人员如何看待 AI 研发自动化与智能爆发 research by Severin Field, Oct 10, 2026. Explore AI alignment and safety research from the MATS Program.
AI Researchers' Views on Automating AI R&D and Intelligence ExplosionsAI Researchers' Views on Automating AI R&D and Intelligence Explosions research by Severin Field, Aug 17, 2026. Explore AI alignment and safety research from the MATS Program.
Automatically Finding Reward Model BiasesAutomatically Finding Reward Model Biases research by Atticus Wang, Aug 17, 2026. Explore AI alignment and safety research from the MATS Program.
自动发现奖励模型中的偏差自动发现奖励模型中的偏差 research by Atticus Wang, Oct 10, 2026. Explore AI alignment and safety research from the MATS Program.
评估并改进 LLM 对分布外对齐失败的监控评估并改进 LLM 对分布外对齐失败的监控 research by Dylan Feng, Oct 08, 2026. Explore AI alignment and safety research from the MATS Program.
Benchmarking and Improving Monitors for Out-Of-Distribution Alignment Failure in LLMsBenchmarking and Improving Monitors for Out-Of-Distribution Alignment Failure in LLMs research by Dylan Feng, Aug 17, 2026. Explore AI alignment and safety research from the MATS Program.
Biases in the Blind Spot: Detecting What LLMs Fail to MentionBiases in the Blind Spot: Detecting What LLMs Fail to Mention research by Iván Arcuschin Moreno, Aug 17, 2026. Explore AI alignment and safety research from the MATS Program.
MATS lists multiple open roles with update dates in early September 2026, including General Counsel and several research/program/operations positions supporting the organization’s ongoing expansion.
MATS Autumn 2026 applications are open with new research tracks—apply or refer a friendAnnouncement of the first-ever Autumn cohort and two new tracks (Biosecurity and Founding & Field Building), including the fellowship dates and application closing date.
MATS Autumn 2026 program details (10 weeks; Berkeley + London; tracks include Founding & Field-Building and Biosecurity)Program page detailing the Autumn 2026 fellowship duration, locations, and the newly added tracks, plus the stated mentorship setup and extension option.
MATS Winter 2027 program runs January 19 to April 10, 2027 (Berkeley + London; optional funded extension)Program page with key dates and structure for the Winter 2027 program, including locations and the option to apply for a 6–12 month funded extension.