We're launching the Anthropic STEM Fellows Program. AI will accelerate progres

Anthropic has announced the launch of its STEM Fellows Program, a structured initiative designed to embed domain experts from scientific and engineering fields directly into the company's research teams for focused, short-term collaborative projects. The program runs from June 15 to September 15, requiring fellows to work full-time and in-person at Anthropic's San Francisco headquarters. Participants receive a weekly stipend of approximately $3,800 alongside office access and benefits, and must hold a PhD in a STEM field or possess equivalent research experience, along with US work authorization. The fellowship is explicitly oriented around improving Claude's scientific capabilities, pairing fellows' deep domain knowledge with frontier model access and internal tooling to address specific gaps in how Claude reasons about and operates within scientific contexts. The program's technical scope is notably concrete and methodologically rigorous. Fellows are tasked with three intersecting workstreams: designing evaluations that test whether Claude can plan experiments, interpret data, and reason about mechanistic processes within their field; identifying or generating data and techniques that target identified capability gaps; and applying Claude to genuine open problems using test-time compute strategies and scientific software tools. Example projects cited include materials scientists constructing reasoning pipelines around phase stability and climate scientists integrating atmospheric modeling tools — suggesting the program is oriented toward applied, domain-specific AI augmentation rather than generalist capability research. Fellows receive direct mentorship from Anthropic researchers and work alongside a cohort of peers, creating both structured guidance and cross-disciplinary exchange. The program reflects a broader strategic posture within Anthropic toward accelerating AI's impact on scientific discovery. By recruiting credentialed STEM experts rather than AI generalists, Anthropic is effectively acknowledging that meaningful progress in scientific AI requires grounding in domain-specific epistemology — understanding what constitutes valid experimental design, plausible mechanisms, or interpretable results within a given field. This approach contrasts with earlier AI development paradigms that treated scientific reasoning as largely transferable from general language capabilities, and instead treats it as something that must be deliberately evaluated and engineered with expert input. The fellowship structure also functions as a talent pipeline and network-building exercise, connecting Anthropic with researchers who may later become long-term collaborators, advisors, or hires. Within the wider AI industry landscape, the STEM Fellows Program situates Anthropic alongside a growing cohort of frontier AI labs investing in domain-specific scientific AI programs. Google DeepMind's AlphaFold work, Microsoft's partnerships with research institutions, and various academic-industry collaborations all signal that the race to demonstrate AI's value in science has become a key competitive and reputational front. Anthropic's approach — structured fellowships rather than large-scale research partnerships — emphasizes interpretability and targeted evaluation, consistent with the company's general emphasis on safety and measurable capability assessment. The three-month timeline and scoped project structure also suggest a preference for rapid iteration and verifiable outcomes over open-ended research, which aligns with the evaluation-centric framing of the fellowship's stated goals. The program's launch in April 2026 comes at a moment when AI systems are increasingly being positioned not merely as productivity tools but as active participants in the scientific process. Anthropic's framing — that AI will "accelerate progress in science and engineering" — signals a long-horizon ambition beyond incremental assistance. By investing in fellows who can stress-test and improve Claude's capacity for experimental judgment and long-horizon scientific reasoning, Anthropic is building the empirical infrastructure necessary to make credible claims about Claude's scientific utility. Whether such programs yield durable capability improvements or primarily serve as structured external evaluation exercises, they represent a meaningful evolution in how AI companies are engaging the scientific community — not merely as end users, but as co-developers of the models themselves.