Anthropic Fellows Program 2026 | Study in United States — Anthropic

A weekly stipend north of $3,800 in the US, four months of full-time research, and no requirement to have ever worked in AI before. The Anthropic Fellows Program takes people with strong technical instincts, pairs them with experienced AI researchers, and points them at real, publishable safety and security problems.

Run by Anthropic, the fellowship treats prior AI research experience as a bonus, not a prerequisite, which sets it apart from most research fellowships in the field.

About the Scholarship

The Anthropic Fellows Program exists to grow the pool of skilled AI researchers and engineers working on making AI systems safer and more reliable. Fellows spend four months full-time completing an empirical research project built on open-source models, public APIs, and other external research infrastructure, working closely with an Anthropic mentor throughout. The explicit goal is a public research output, whether that is a paper or a technical publication, and the program’s own track record backs that up: more than 80 percent of fellows in previous cohorts have gone on to produce research papers.

Fellows choose from five distinct workstreams: AI Safety, AI Security, ML Systems and Performance, Reinforcement Learning, and the Anthropic Institute track covering Economics and Policy. That range means the program pulls in far more than computer scientists. Mathematicians, physicists, economists, and policy researchers all have a defined path into the fellowship, provided they bring strong technical or analytical skills and genuine interest in AI’s trajectory.

Work happens either from shared workspaces in Berkeley, California or London, United Kingdom, or remotely for fellows based in the United States, the United Kingdom, or Canada. Participants must already hold work authorization in whichever country they are based in, since the program does not sponsor visas, and fellows are required to remain in their authorized country for the duration.

Because the fellowship explicitly welcomes candidates without prior AI research experience, it functions as a genuine entry point into the field for people with strong technical backgrounds who have not yet had the chance to work on AI-specific problems. Applicants weighing other paid, technically demanding US and UK research roles sometimes also compare this fellowship against the GovAI Research Fellowship, a much longer two-year contract role focused specifically on AI governance and policy rather than a four-month empirical research sprint spanning multiple technical workstreams. Those drawn more to policy research generally might also look at the Wilson Center Fellowship Program, a paid, one-year fellowship covering a much broader range of foreign policy topics beyond artificial intelligence specifically.

The four-month timeline is worth understanding on its own terms. It is long enough to produce a genuine, publishable piece of research, but short enough that fellows are expected to move quickly from problem selection to execution, without the extended ramp-up time typical of a PhD program or multi-year postdoc. That pace suits candidates who already have strong technical fundamentals and mainly need focused time and mentorship to direct those skills toward a specific AI research question, rather than candidates starting from scratch on core technical training.

The mentorship structure also deserves attention. Rather than a loose advising relationship, fellows work closely with an assigned Anthropic mentor throughout the entire four months, which means the quality and specificity of feedback a fellow receives tends to be considerably more hands-on than in a typical academic research assistantship, where a single advisor might be splitting attention across a much larger group of students.

Scholarship Benefits

Compensation is structured to let fellows focus entirely on research for the full four months, without the distraction of a second job or unfunded computing costs.

• A weekly stipend of USD 3,850 for US-based fellows, GBP 2,310 for UK-based fellows, or CAD 4,300 for Canada-based fellows
• Country-specific additional benefits
• Direct mentorship from experienced Anthropic researchers throughout the fellowship
• Funding for computing resources of approximately USD 15,000 per month
• Coverage of additional research-related expenses
• Access to shared workspaces in Berkeley or London, or the option to work remotely
• Membership in a growing AI safety and security research community
• The opportunity to publish public research outputs, including academic papers
• A track record of past fellows continuing into AI safety and research careers, with some receiving offers for future roles at Anthropic

Eligibility Criteria

• Applicants should be genuinely interested in AI research and engineering
• A strong technical background is required, though not necessarily in AI specifically
• Backgrounds in computer science, mathematics, physics, or related technical fields are encouraged
• Previous AI research experience is not mandatory
• Fluency in Python programming is required
• Applicants must be available to work full-time for the full four months
• Genuine interest in developing safe and beneficial AI systems is expected
• Strong problem-solving, communication, and collaboration skills are required
• Applicants must have legal work authorization in the United States, the United Kingdom, or Canada, and must remain in that country throughout the fellowship
• Anthropic actively encourages candidates from diverse and underrepresented backgrounds, and those who do not meet every preferred qualification are still encouraged to apply
• Candidates with experience in large language models, open-source contributions, or prior AI safety or security work may have an advantage in specific workstreams, though this is not a strict requirement

Fields of Study Covered

Because the fellowship spans five distinct workstreams, it draws candidates from a genuinely broad range of technical and analytical fields, including:

• Computer science and machine learning
• Mathematics and statistics
• Physics and other quantitative sciences
• Cybersecurity and distributed systems
• Reinforcement learning
• Economics and public policy, specifically for the Anthropic Institute track

Required Documents

• Completed online application form
• Updated CV or resume
• Reference information
• Technical assessment, if required for the selected workstream
• Any additional information requested during the application process

How to Apply (Step-by-Step)

1. Visit the official Anthropic Fellows Program application portal.
2. Complete the online application form with personal, educational, and professional details.
3. Select your preferred fellowship workstream based on your background and interests.
4. Upload the required information and documents, including your CV and reference details.
5. Submit your application before the deadline.
6. Complete a reference check if shortlisted.
7. Participate in technical assessments, interviews, and a research discussion during the later selection stages.

Selection Process

Selection moves through several stages: an initial application review, a reference check for shortlisted candidates, technical assessments and interviews, and a research discussion that gives the Anthropic team a chance to see how a candidate thinks through an open-ended technical problem in real time. Because the program does not require prior AI research experience, reviewers weigh raw technical strength and genuine curiosity heavily alongside whatever research background a candidate already has.

Given the program’s emphasis on diversity and its explicit encouragement of candidates who do not meet every preferred qualification, the selection process appears designed to catch strong technical talent that a more narrowly credentialed screen might filter out.

Tips to Improve Your Chances

• Be specific about which workstream fits your background and interests, rather than applying generically across all five tracks
• Highlight concrete technical projects, even outside AI specifically, that demonstrate strong problem-solving and implementation skills
• Brush up on Python fluency before applying, since this is a stated requirement rather than a preference
• If you lack AI research experience, use the application to show genuine curiosity and initiative in learning about AI safety or security independently
• Prepare for the research discussion stage by practicing talking through your technical thinking process out loud, not just presenting a finished solution
• Read a few examples of research produced by past Anthropic Fellows if available, to calibrate the scope and ambition level typical of a successful four-month project

Frequently Asked Questions

Do I need prior AI research experience to apply?

No. The program explicitly states that previous AI research experience is not mandatory, and it welcomes strong technical candidates from computer science, mathematics, physics, and related fields regardless of AI-specific background.

Which countries can fellows be based in?

Fellows must have work authorization in the United States, the United Kingdom, or Canada, and must remain in their authorized country for the full four months of the fellowship.

Can fellows work remotely, or is in-person required?

Both options exist. Fellows can work from shared workspaces in Berkeley or London, or remotely if based in the US, UK, or Canada.

What research workstreams are available?

Fellows choose from AI Safety, AI Security, ML Systems and Performance, Reinforcement Learning, or the Anthropic Institute track covering Economics and Policy.

How long does the fellowship last?

The program runs for four months, with fellows working full-time throughout, and successful applicants joining the cohort starting in November 2026.

What happens after the fellowship ends?

Fellows typically leave with a public research output, such as a paper or technical publication, and outstanding participants may receive opportunities for future roles at Anthropic, following a pattern seen in previous cohorts.