Conceptual Reasoning Fellow

About Anthropic

Anthropic is an AI safety company focused on developing reliable, interpretable, and controllable AI systems. Founded by former members of OpenAI, the organization brings together researchers, engineers, policy specialists, and business professionals to advance safe and beneficial AI. The company is known for creating Claude, a large language model designed with safety and alignment principles at its core, and operates from offices in San Francisco while maintaining a distributed team.

Conceptual Reasoning Fellowship

Anthropic is seeking exceptional researchers and domain experts to join a four-month fellowship program that pairs deep disciplinary knowledge with frontier AI capabilities. You'll work at the intersection of your field and advanced AI reasoning, helping evaluate and extend Claude's ability to tackle complex problems that require expert judgment. This fellowship is designed for candidates with strong backgrounds in philosophy, cognitive science, economics, policy, or law who are ready to contribute concrete research during an intensive, fast-paced engagement.

What you'll do

• Define critical unsolved problems in your discipline and design systematic approaches to address them using Claude and internal evaluation tools.

• Build evaluations, benchmarks, and expert rubrics that rigorously test how well Claude reasons about judgment-heavy questions specific to your field.

• Collaborate with other Fellows to create challenging environments that push Claude's capabilities on the most demanding problems your discipline offers.

• Assess Claude's strengths and gaps when applied to major societal challenges where your field's expertise is essential.

• Deliver concrete, measurable results on a scoped project matched to your background with mentorship from Anthropic's research team.

What you'll bring

• A PhD (or equivalent research experience) in philosophy, cognitive science, economics, policy, or law.

• Genuine enthusiasm for AI's potential to reason effectively on difficult field-specific questions requiring expert judgment.

• Active, hands-on experience using Claude and Claude Code as part of your workflow.

• Demonstrated expertise in your field through published research, scholarly contributions, or expert-level analytical work.

• Comfort operating in a dynamic, high-ownership environment and delivering results under tight timelines.

Nice to have

• Experience designing evaluations, benchmarks, datasets, or expert assessment frameworks.

• Track record of translating domain expertise into measurable evaluation criteria.

What they offer

• Compensation of 100,000 to 167,000 USD for the four-month fellowship, equivalent to 300,000 to 500,000 USD annualized, plus additional benefits.

• Full-time engagement from January 12 to May 8, 2027, with possible extension consideration.

• Access to frontier Claude models and proprietary internal tooling.

• Mentorship and collaboration with Anthropic's research community and fellow domain experts.

• One-week in-person orientation in San Francisco; remote work within the United States possible in select circumstances.

Note: This is a fixed-term contract role administered through an employer of record. You must have existing full-time work authorization in the United States; visa sponsorship is not available for this program. Application deadline is October 28, 2026, with offers expected by November 20.