Anthropic Fellows Program: January 2027 cohort
Anthropic Fellows get a weekly stipend of $3,850 and about $15k a month for compute for four months of AI safety or security research. Applications for the January cohort are open.
The Anthropic Fellows Program is now taking applications for its January cohort. Fellows spend four months on full time empirical research in AI safety or AI security, with a possible extension. They mostly use external infrastructure such as open models and public APIs, and aim for a public output such as a paper.
Fellows are paid a weekly stipend of 3,850 USD, 2,310 GBP or 4,300 CAD plus benefits. Anthropic also gives about $15k a month for compute and other research costs. Each fellow gets direct mentorship from Anthropic researchers and works from a shared space in Berkeley or London, or remotely from the US, UK or Canada.
You need work authorization in the US, UK or Canada and must be located there during the program. Anthropic does not sponsor visas. It looks for a strong background in computer science, maths or physics and fluency in Python. A PhD, prior ML experience or published papers are not required.
There are two workstreams. AI Safety covers scalable oversight, resistance to adversarial attacks, AI control, model organisms, interpretability and AI welfare. AI Security values experience with pentesting, CVEs and bug bounties.
Anthropic runs several cohorts a year and reviews applications on a rolling basis. Selection starts with the application and a reference check, then technical assessments and interviews, then a research discussion. In earlier cohorts, 25 to 50% of fellows received a full time offer.