Can an AI model reason through a hard legal question or an ethical problem? Anthropic wants experts to find out. It has opened applications for the Conceptual Reasoning Fellow programme. This is a four month research fellowship for people in philosophy, cognitive science, economics, public policy and law. Fellows work with Claude to test how well it reasons and to find ways to improve it.
The pay ranges from 100000 to 167000 US dollars for the four months. But there is one hard rule. You must already have permission to work in the United States. Applications close on 28 October 2026.
Last Information Checked: 17 October 2026
Quick Facts
| Item | Details |
| Programme | Conceptual Reasoning Fellow |
| Host | Anthropic |
| Length | Four months full time |
| Dates | 12 January to 8 May 2027 |
| Pay | 100000 to 167000 US dollars for the four months |
| Default location | San Francisco California |
| Orientation | One week in person in San Francisco |
| Deadline | 28 October 2026 |
| Offers expected | By 20 November 2026 |
What the Fellowship Is
Anthropic builds AI systems that are meant to be reliable, interpretable and steerable. The main goal of this is to make sure that AI benefits society.
This fellowship focuses on questions that cannot be solved by simply checking a fact. Many legal philosophical, economic and policy problems need context and careful judgment. Fellows will work with Anthropic researchers. They see how frontier AI systems can support this kind of reasoning.
Each fellow does a defined research project that fits their background. They get mentors and access to Claude models and internal evaluation tools. They also work with the other fellows.
What Fellows Do
- Find important open questions in their field and test ways to answer them using Claude
- Study big social challenges that need expert knowledge and see where AI does well or badly
- Design evaluation benchmarks and reinforcement learning environments that test how a model reasons
- Turn expert knowledge into practical tests. It measures the quality, reliability and limits of AI analysis
- Work with other fellows and Anthropic researchers to build hard tasks that need advanced judgment
Example Projects by Field
These are examples and not fixed assignments. Fellows work with mentors to shape their question and scope.
| Field | Possible project |
| Philosophy | Build hard ethical problems or formal epistemology tasks that test if a model can reason through competing principles |
| Cognitive science | Design experiment tasks that test model comparison, hypothesis testing and research design choices |
| Economics | Create economic modelling and identification problems with reliable correct answers to test accuracy |
| Public policy | Build regulatory analysis and cost-benefit tasks with criteria set by experts |
| Law | Build datasets about legal reasoning and AI policy to see how well models read and analyse legal questions |
Fellowship Benefits
- Pay: 100,000 to 167,000 US dollars for the four months depending on the arrangement. Anthropic says this equals an annual rate of about 300000 to 500000 US dollars
- Other benefits: The call says other benefits are also provided
- Mentorship: Support from mentors while you shape your research project
- Access: Use of Claude models and internal evaluation tools
- Community: Work with other fellows and with Anthropic researchers
- Matched project: A research project chosen to fit your background
- Orientation: A one-week in-person orientation in San Francisco
- Real impact: A chance to add your expertise to the development and testing of advanced AI systems
- Possible extension: Some placements may be considered for an extension
Who Can Apply
You must meet all of these.
- You hold a PhD or are completing a PhD in philosophy, cognitive science, economics, public policy or law. You can also show equal research experience
- You use Claude and Claude Code in your regular work
- You are excited about AI reasoning through hard questions in your field
- You are comfortable in a fast-moving place with a lot of independence
- You can deliver real research results within four months
- You already have permission to work in the United States for the whole fellowship
Preferred skills
- Experience designing evaluation benchmark datasets or expert grading rubrics
- Proven expertise in your field through research, scholarship or high-level analytical writing
Location and Contract
The fellowship is full-time from 12 January to 8 May 2027. The default location is San Francisco. Anthropic may allow remote work inside the United States in some cases. Every fellow must attend the one week orientation in San Francisco.
The role is a fixed term contract. Fellows are hired through an employer of record. They are not direct employees of Anthropic.
Important: Anthropic cannot sponsor visas for this fellowship. You must already have or get full time US work authorisation on your own. International applicants without it do not meet the condition.
How Selection Works
Later stages usually include a take home exercise. You also talk with researchers and possible mentors about your research interests and project ideas.
How To Apply
- Go to the official Anthropic careers website.
- Find the Conceptual Reasoning Fellow position.
- Prepare your current CV or résumé.
- Prepare a LinkedIn profile or résumé as the form asks.
- Choose two professional references.
- Write your proposed research ideas.
- Fill in the questions about your Claude Code experience and any earlier AI or machine learning work.
- Name the field in which you want to contribute.
- Submit before 28 October 2026.
Confirm that the vacancy is current. Apply only through Anthropic’s official recruitment platform. Real Anthropic recruiters use official email addresses that end in @anthropic.com. You should never pay a recruitment fee.
Mistakes To Avoid
- Applying without US work authorisation
- Expecting Anthropic to sponsor your visa
- Writing research ideas that are too wide to finish in four months
- Skipping your Claude and Claude Code experience in the form
- Forgetting that you must attend the one week orientation in San Francisco
- Replying to anyone who asks for money or uses an email that does not end in @anthropic.com
Key Dates
- 28 October 2026: Applications close
- By 20 November 2026: Offers expected
- 12 January 2027: Fellowship starts
- 8 May 2027: Fellowship ends