The full-time programme runs from January 12 to May 8, 2027, offering selected scholars a rare opportunity to shape how frontier models reason through complex, open-ended problems.
The initiative builds on the success of an initial pilot cohort of more than 20 STEM Fellows earlier this year. According to Anthropic, the evaluation frameworks and datasets generated by those researchers now directly inform Claude's training regimen and internal science evaluations.
"Applications are now open for two new Anthropic Fellowships," Anthropic wrote in a post on X. "STEM Fellows (science and engineering) and Conceptual Reasoning Fellows (philosophy, cognitive science, economics, policy, and law) will spend four months at Anthropic from January 2027, working to help Claude reason well about some of the hardest problems in those fields. For more information, see the descriptions below. The application deadline is October 28."
Bridging the gap in science & engineering
The STEM Fellows track targets researchers with a PhD and professional experience in mathematics, physics, life sciences, engineering, and earth or climate science. The programme addresses a persistent limitation in current AI systems: a lack of the deep experimental judgment and long-horizon planning required to push boundaries in disciplines like mathematics or drug discovery.
Working in small domain-specific pods, STEM Fellows will collaborate weekly with Anthropic researchers to identify long-horizon open problems where models currently fall short. Their primary responsibility involves building comprehensive datasets, simulations, and rigorous, hard-to-game evaluations. These tests assess whether Claude can effectively plan experiments, interpret raw data, reason through physical mechanisms, and identify its own blind spots when using domain-specific software and test-time compute strategies.
Tackling uncheckable questions in conceptual reasoning
Running in parallel, the Conceptual Reasoning Fellows track invites scholars in philosophy, cognitive science, economics, policy, and law — holding or completing a PhD or possessing equivalent research experience — to focus on problems that lack a single checkable answer.
These fellows will design evaluation frameworks and reinforcement learning environments that test model performance on judgment-heavy tasks. Representative projects span philosophical cases requiring normative reasoning in formal epistemology, experimental design tasks for cognitive scientists, modeling and identification challenges for economists, and regulatory cost-benefit analyses or AI-policy datasets for legal and policy scholars.
Logistics, compensation & requirements
Both fellowships are structured as four-month, fixed-term contracts operating out of Anthropic's San Francisco headquarters, though remote arrangements within the United States may be considered under specific circumstances. All fellows are expected to attend a mandatory one-week in-person orientation in San Francisco.
Compensation is set between $100,000 and $167,000 for the duration of the programme, translating to an annualized rate of $300,000 to $500,000, alongside standard benefits. Candidates must possess independent, full-time work authorization in the United States, as Anthropic is not currently offering visa sponsorship for these specific positions.
Applications for both programs close on October 28, 2026, with final offers projected to be extended by November 20.