Anthropic Fellows Program — Fellowship

The Anthropic Fellows Program funds engineers and researchers to work full time for four months on empirical AI research questions that Anthropic has prioritised, each paired with an Anthropic researcher as mentor. Anthropic is an AI company; the programme is announced and its results are published on the company’s alignment science blog. Mentors pitch project ideas at the start of a cohort, and fellows choose and shape a project in close collaboration with their mentor. One application covers five workstreams: AI safety, AI security, ML systems and performance, reinforcement learning, and economics and societal impacts. Inside the AI safety workstream the research areas are scalable oversight, adversarial robustness and AI control, model organisms, model internals and mechanistic interpretability, and AI welfare. Anthropic reports that over 80% of fellows produced papers and that over 40% later joined the company full time, while stating plainly that it does not guarantee any full-time offer.

Benefits & funding

Weekly stipend of 3,850 USD, 2,310 GBP or 4,300 CAD depending on where the fellow is based, plus benefits that vary by country.
Funding for compute of roughly 15,000 USD a month, plus other research expenses.
Close mentorship from an Anthropic researcher across the four months, on a project drawn from Anthropic’s own research agenda. Named AI safety mentors include Ethan Perez, Sam Bowman, Trenton Bricken, Collin Burns, Kyle Fish, Samuel Marks and Alex Tamkin.
Access to a shared workspace in Berkeley, California or London, UK, which mentors visit. Remote fellows based in the United States, the United Kingdom or Canada are also accepted.
Four months of full-time research, with a possible extension.
Past fellow projects include work on agentic misalignment, subliminal learning, rapid response to new ASL3 jailbreaks, open-source circuit tracing for interpretability, and finding vulnerabilities in blockchain smart contracts.

Eligibility

Work authorisation in the United States, the United Kingdom or Canada, held already or obtained independently, and presence in that country for the length of the programme. Anthropic does not sponsor visas for fellows.
Fluent in Python.
Available to work full time, an expected 40 hours a week for four months.
No PhD, prior machine-learning experience or published papers required.
Able to make concrete progress on ambiguous problems, pick up new skills quickly and debug when things break; motivated by reducing catastrophic risks from advanced AI.
Helpful but not required: a strong background in a discipline tied to one workstream, such as economics, social sciences or cybersecurity, and research or engineering experience in that area. Past fellows have come from physics, mathematics, computer science and cybersecurity.

Can you apply?

Worth a closer look

Residence required
USA · UK · Canada
Not in your profile yet

We only check the criteria we have data for. Always confirm full eligibility on the official page before applying.

Application process

1
Apply through the programme posting on Anthropic’s job board. Applications and interviews are managed by Constellation, Anthropic’s recruiting partner.
2
One application is read across all five workstreams; state a workstream preference if you have one.
3
Say in the application if the published start date does not work, since another timeline can sometimes be arranged.
4
Initial application review and reference check.
5
Technical assessments and interviews.
6
A research discussion.
7
Selected fellows start on the cohort date and work 40 hours a week for four months.

Program timeline

26 July 2026, 11:59pm PT
applications closed for the November 2026 cohort
2 November 2026
the November 2026 cohort begins
Four months of full-time research follow the start date, with a possible extension
Anthropic runs multiple cohorts each year and reads applications on a rolling basis; no date is published for the next call