MATS Summer 2026 launches the largest alignment-research cohort to date — 120 fellows, 100 mentors, June through August across multiple methodological tracks
The MATS (ML Alignment & Theory Scholars) Summer 2026 program runs June through August with 120 fellows and 100 mentors — roughly double the 2024 cohort. The program is the dominant pipeline for new mechanistic-interpretability and alignment researchers, and the 2026 expansion comes alongside funding commitments from Anthropic, OpenAI, DeepMind, and Open Philanthropy.
The pipeline expansion is the substantive update. Frontier labs have been signaling that alignment-research headcount is the binding constraint on the safety side of the capability/safety balance; MATS doubling its annual throughput addresses that constraint at the entry tier. 120 fellows produce roughly 60-80 publications across the August demo period, which then feed the labs' own hiring pipelines for the September-October cycle.
The methodological diversity matters more than the count. The cohort includes tracks on mechanistic interpretability, behavioral analysis, training-data influence, capability-evaluation frameworks, and theoretical alignment work. Given DeepMind's just-announced SAE deprioritization, the methodological diversity is timed well: the field needs research breadth at exactly the moment its dominant methodology is being questioned. MATS Summer 2026 will produce the first generation of alignment researchers trained inside the post-SAE-dominance era.
Zylos Research — AI Safety, Alignment, and Interpretability in 2026 → · LessWrong — AI Safety at the Frontier: Paper Highlights → · IntuitionLabs — Understanding Mechanistic Interpretability in AI Models →