MATS Summer 2026 launches with 120 fellows and 100 mentors — the largest cohort to date lands as DeepMind questions the SAE research direction the program funds
The ML Alignment & Theory Scholars (MATS) Summer 2026 program runs June through August with 120 fellows and 100 mentors — the largest cohort in the program's history. MATS has been the dominant pipeline for new mechanistic-interpretability researchers since 2023; the program scales just as DeepMind publishes negative results on the SAE methodology MATS has heavily funded.
The scale-up signal is the substantive piece. 120 fellows × 100 mentors is roughly double the program's 2024 cohort and signals that interpretability and alignment remain the highest-leverage research-talent investment frontier labs and the broader safety community can make. The funding side — partial support from OpenPhil, Anthropic, and several frontier labs — has held even as the technical methodology under review.
The methodology tension is the under-discussed piece. DeepMind's deprioritization of SAE research arrives as MATS scales to a record cohort partly committed to SAE-adjacent projects. The cohort will be the first generation of safety researchers trained explicitly inside a methodological transition — from the dominant SAE paradigm of 2024-2025 to whatever replaces it in 2026-2027. That's an inflection point for the discipline regardless of which methodology wins.
Zylos Research — AI Safety, Alignment, and Interpretability in 2026 → · Leonard Bereska — Mechanistic Interpretability for AI Safety — A Review → · IntuitionLabs — Understanding Mechanistic Interpretability in AI Models →