Anthropic Fellows Program opens May and July 2026 application windows across six alignment research areas — formalizes external-fellowship pattern as alignment-research scaling primitive
Anthropic Fellows Program applications open for May and July 2026 cohorts across six alignment areas (scalable oversight, adversarial robustness, AI control, model organisms, mech-interp, model welfare). The cadence (twice-yearly cohorts) establishes the external-fellowship pattern as a permanent alignment-research-scaling primitive — and the six-area structure formalizes the field's research-priority taxonomy.
The substantive piece is the research-priority taxonomy formalization. The six areas Anthropic explicitly funds (scalable oversight, adversarial robustness, AI control, model organisms, mechanistic interpretability, model welfare) constitute the most authoritative public statement of the field's research-priority structure. Other organizations (OpenAI Superalignment, DeepMind Safety, AISI) will increasingly use this six-area taxonomy as the reference framework, which structurally accelerates cross-org research coordination.
The connection to the OpenAI-Anthropic cross-lab evaluation cadence is that the alignment-research ecosystem is now operating on three simultaneous infrastructure tracks: cross-lab evaluation (OpenAI-Anthropic + METR), internal-developer-agent evaluation (METR pilot), and external-research-fellow scaling (Anthropic Fellows + similar programs). The H2 2026 alignment-research output trajectory reflects all three tracks operating at maturity for the first time — production research output should compound through H1 2027.
Anthropic — Anthropic Fellows Program for AI safety research → · Anthropic — Alignment Research → · Anthropic — Core views on AI safety →