Anthropic STEM fellowship

Anthropic has posted a STEM fellowship that brings physicists, biologists, and engineers into its research teams for a few months in San Francisco. (job-boards.greenhouse.io) The posting says fellows will work on scoped projects with Anthropic researchers, using Claude models and internal evaluation tools to test where the systems can plan experiments, interpret data, and reason about mechanisms in a scientific field. Prior machine learning experience is “helpful but not required,” and the role targets people with a PhD in a STEM field or equivalent research experience. (job-boards.greenhouse.io) Anthropic frames the program around building “an AI scientist,” meaning a system that can handle long-horizon reasoning and experimental judgment rather than just answer questions. The job post gives examples including a materials scientist building property-prediction tasks around phase stability and a climate scientist integrating atmospheric modeling tools. (job-boards.greenhouse.io) The fellowship extends a broader Anthropic strategy of using paid, mentored programs to bring in researchers who are not already inside major AI labs. Anthropic’s main Fellows Program says it is designed to foster AI research and engineering talent “regardless of previous experience,” with cohorts starting July 20, 2026 and applications for that cohort due April 26, 2026. (job-boards.greenhouse.io) That main fellows track pays $3,850 a week in the United States for four months and provides about $15,000 a month for compute and research expenses. Anthropic says more than 80% of fellows in an earlier cohort produced papers, and more than 40% later joined the company full-time. (job-boards.greenhouse.io, alignment.anthropic.com) Anthropic’s public research agenda helps explain where a STEM fellow could land. Its research page lists teams focused on alignment, interpretability, economic research, and societal impacts, and its recent science posts include “Long-running Claude for scientific computing” and “Vibe physics: The AI grad student,” both published on March 23, 2026. (anthropic.com) The company also ties the work to older safety lines that predate Claude’s current product push. Anthropic’s fellows postings point back to research areas including scaling laws, artificial intelligence and compute, concrete problems in artificial intelligence safety, and learning from human preferences. (job-boards.greenhouse.io) Learning from human preferences is Anthropic’s term for training models with human judgments about better and worse answers, while Constitutional AI uses written rules to steer outputs instead of relying only on human ratings. Anthropic has published both methods as part of its alignment research and updated Claude’s constitution again on January 22, 2026. (anthropic.com, anthropic.com) Anthropic says its mission is to build “reliable, interpretable, and steerable” AI systems and to push a “race to the top” on safety. The STEM fellowship turns that language into a hiring pipeline for scientists whose expertise is in labs, simulations, and fieldwork rather than machine learning résumés. (anthropic.com, job-boards.greenhouse.io)