Do you want to move into AI safety research but have no big research record yet? The Anthropic Fellows Program in AI Safety and Security funds promising technical people to do full-time research for four months. You get a weekly stipend and mentors. You also get a large compute budget. The next cohort is expected to start in January 2027.
Information Checked: 15 October 2026
At a Glance
| Item | Details |
|---|---|
| Programme | Anthropic Fellows Program AI Safety and Security |
| Length | 4 months full time about 40 hours a week |
| Start | January 2027 cohort |
| Weekly stipend | 3850 US dollars or 2310 pounds or 4300 Canadian dollars |
| Research funding | About 15000 US dollars a month for compute and approved costs |
| Who can apply | Technical candidates with work authorisation in the US UK or Canada |
| Visa support | None |
| Deadline in your source | 18 October 2026 |
What the Programme Is
The Fellows Program is meant to grow research talent in AI safety and security. Fellows work on projects that match Anthropic’s priorities. The goal is to produce public work such as academic papers.
Fellows mostly use outside research tools. These include open source models and public APIs. Anthropic reports that more than 80 percent of fellows in one earlier cohort produced papers.
The programme also welcomes people who want to switch into empirical AI research. You do not need to meet every listed preference. Anthropic says it wants promising technical people from many backgrounds.
Fellowship Locations
- London
- Berkeley
- California
- Remote in the
- United States
- United Kingdom
- Canada
Fellowship Benefits
- Weekly stipend of 3850 US dollars or 2310 pounds or 4300 Canadian dollars depending on your location
- Mentorship with direct guidance from experienced Anthropic researchers
- Research funding of about 15000 US dollars a month for compute plus other approved research costs
- Workspace in shared offices in Berkeley or London
- Community links to the wider AI safety and security research world
- Public output with a chance to publish papers and other research work
- Career paths with support towards future research and engineering roles. Anthropic says 25 to 50 percent of fellows in earlier cohorts received full-time offers
- Extension may be possible. Ask about it during recruitment
The Two Tracks
| Track | Focus | Example work |
|---|---|---|
| AI Safety | Reducing risks from more capable AI systems | Scalable oversight, adversarial robustness and AI control, model organisms, mechanistic interpretability, AI welfare |
| AI Security | Finding and fixing weaknesses in AI systems and related technology | Penetration testing, vulnerability research, security evaluations, methods to assess advanced AI systems |
Strong signs for the Safety track include experience with empirical machine learning research or large language models. Open source contributions also help.
Strong signs for the Security track include open source work on language model or security projects. Reported CVEs. Bug bounties. Experience with deep learning tools and experiment tracking.
Who Can Apply
You need a strong technical base in computer science mathematics physics or a related field. People with skills in economics social science or cybersecurity may fit some workstreams.
All applicants must:
- Be fluent in Python
- Be free to work full time during the programme
- Have full time work authorisation in the United States or the United Kingdom or Canada
- Be able to live in that country for the whole programme
Anthropic does not sponsor visas for fellows. You must already have or get your own work authorisation.
If you cannot give all four months explain your situation in your application. Cases are looked at one by one.
Key Dates
- 18 October 2026: Deadline given in your source. No time zone was given.
- Rolling basis: Applications are reviewed as they arrive according to the listing.
- January 2027: Expected cohort start. Say in your application if this date does not work for you.
How to Apply for the Anthropic Fellows Program
- Read the research areas and choose the track that fits you best.
- Update your CV. Show your Python skill and your technical projects. Add research work and open source contributions.
- Fill in the application form run by Constellation. Constellation is Anthropic’s recruiting and programme partner.
- Do not worry about the separate Greenhouse application. Your source says it is not required.
- Prepare for reference checks and technical tests. Expect interviews and a research discussion too.
You must finish the Constellation form to be considered.