$3,850 per week plus about $15,000 per month in compute funding is a serious budget. The Anthropic Fellows Program 2026 is a four-month, full-time, funded research fellowship focused on Anthropic’s highest-priority AI safety and AI security questions.
Engineers switching into safety work, security-minded developers, and researchers who want publishable output tend to get the most out of this. If you’re searching for “Anthropic free credits” or trying to figure out how to get Anthropic credits for research, this program is one of the rare options where you don’t just get credits. You get paid.
This guide covers Anthropic Fellows Program 2026 eligibility, the exact signup flow, limitations (including the visa rule), and practical ways to stretch the compute budget.
Program at a Glance
| Provider | Anthropic |
| Credit Amount | $3,850/week stipend + ~$15,000/month compute |
| Duration | 4 months (full-time) |
| Eligibility | US/UK/Canada work authorization + strong technical background |
| Credit Card Required? | No (fully funded; you receive money) |
| Difficulty | Competitive (selective screening, assessments, interviews) |
| Best For | AI safety research, AI security research, publishable work |
| Official Page | Anthropic Program Page |
What You Actually Get
The fellowship is a structured, full-time research program (not a casual internship) designed to produce publishable AI safety or AI security work. Compensation includes a $3,850/week stipend (or equivalent in GBP/CAD), plus about $15,000/month in compute funding for research expenses. You also get direct 1:1 mentorship from senior Anthropic researchers, and optional access to shared workspaces in Berkeley, California or London, UK. If you qualify by location, you can also participate remotely from anywhere within the US, UK, or Canada.
In real-world terms, the weekly stipend makes it feasible to treat this as your primary job for four months, not a side project. The compute budget is sized for serious experimentation and evaluation work, including projects that require repeated runs, tooling, and infrastructure. You are expected to aim for a publishable paper, and Anthropic notes that a large majority of previous fellows did in fact produce publishable research.
Who Qualifies (and Who Doesn’t)
Anthropic is unusually explicit about what they want: people who can execute on research, even if they do not have a PhD or prior ML research credentials. That said, the program is constrained by work authorization and by the expectation that you can commit full-time for four months.
- You must be fluent in Python programming, since the work is research-heavy and code-first.
- Full-time availability for the full four-month fellowship is required.
- You need full-time work authorization in the US, UK, or Canada because there is no visa sponsorship.
- A strong technical background in computer science, mathematics, physics, cybersecurity, or a related field is part of the required profile.
If you need visa sponsorship, you are out of luck for this one. Also, the program asks you to apply to only one track (AI Safety or AI Security), so “I’ll apply to both and see what happens” is not the intended path.
How to Sign Up
Expect a real recruiting process, not a simple “claim credits” button.
- Go to the Anthropic Fellows Program announcement page and read the program details.
- Decide which track fits you best: AI Safety Fellow or AI Security Fellow (apply to only one).
- Click “Apply here” on the announcement page, which redirects you to Constellation’s application portal (Constellation is Anthropic’s official recruiting partner).
- Fill out the application form with your background, research interests, and motivation.
- Go through the process stages: initial screening, then technical assessments and interviews, then a research discussion round.
- If accepted, Anthropic mentors pitch project ideas to you, and you choose and shape your project in collaboration with your mentor.
- Start with the cohort start date (May or July 2026, depending on intake).
One practical note: communication and logistics are handled by Constellation, so you should expect outreach from them rather than directly from Anthropic during the application process. Also, May 2026 is closed; July 2026 and beyond are rolling.
What the Credits Cover
The “credits” piece here is compute funding: about $15,000/month earmarked for research expenses during the fellowship. Anthropic doesn’t publish a public line-item list in the announcement, but the intent is clear: fund the compute and infrastructure needed to execute your project, alongside mentorship and workspace access.
| Service / Feature | What It Does | Included? |
|---|---|---|
| Weekly stipend | Personal compensation during the fellowship period. | ✓ |
| Compute funding | Budget for research compute and related expenses. | ✓ |
| 1:1 mentorship | Guidance from senior Anthropic researchers. | ✓ |
| Workspace access | Shared offices in Berkeley or London (optional). | ✓ |
Notable exclusion: there’s no promise of visa sponsorship, and the program is not positioned as a generic “Anthropic API free credits” offer for anyone worldwide. It’s a funded fellowship with location and authorization constraints.
Limitations to Know About
Every funded program has catches. With this one, the constraints are mostly about eligibility, time commitment, and selectivity.
- No visa sponsorship is provided, so you must already have (or independently obtain) work authorization in the US, UK, or Canada.
- You can participate remotely, but only from within the US, UK, or Canada.
- May 2026 cohort applications are closed; July 2026 and future cohorts are rolling.
- You are expected to treat this as full-time research for four months, with publishable output as the goal.
What happens when the credits run out? The fellowship is time-boxed to four months, so the stipend and compute funding are tied to that term. After the fellowship ends, there is no implied continuation of funding; the “pipeline” outcome Anthropic cites is that some fellows receive full-time offers, but that is not guaranteed.
Have Unused Anthropic Credits?
It’s surprisingly common for teams to accumulate Anthropic credits they cannot use before the clock runs out. Startup programs and enterprise agreements can come with large allocations, but roadmaps change, and credits expire quietly. If you’re sitting on surplus, AI Credit Mart lets you sell unused credits (often up to about 70% of face value) instead of watching them become worthless.
Need More Anthropic Credits?
Once you’ve burned through free promos or you’re between funded programs, paying list price isn’t your only option. AI Credit Mart lists discounted Anthropic credits from companies with surplus allocations, which can be a clean way to extend your runway. Discounts typically land around 30–70% below retail, depending on the credit type and expiry window.
Tips for Getting the Most Out of Your Credits
- Apply with the mindset that you are pitching four months of publishable research, because Anthropic explicitly says this is not a typical internship.
- Pick one track (Safety or Security) and align everything you write to its focus areas, since you are asked to apply to only one.
- Plan around the rolling process for July 2026 and beyond, and do not anchor on the earlier January deadline if you missed it.
- Expect Constellation to run logistics and communication, so check your inbox for them (and keep an eye on spam filters).
- If the Fellows program is not the right fit, Anthropic explicitly suggests comparing similar programs like Constellation’s Astra Fellowship.
Frequently Asked Questions
The program includes a $3,850/week stipend plus about $15,000/month in compute funding for research expenses during a four-month, full-time fellowship. In practice, that compute budget is meant to cover the infrastructure needed to execute your project, while the stipend covers your time. It’s not a coupon code for the public; it’s tied to being accepted as a fellow. Anthropic also frames the expected “value” as publishable output, with many prior fellows producing papers and some converting into full-time roles.
No.
For this program, the compute funding runs for the four-month fellowship term, and then it ends.
Yes. If you have Anthropic credits you won’t use before they expire, you can list them on AI Credit Mart and sell them at up to 70% of face value. Companies regularly list surplus credits from startup programs and enterprise agreements.
AI Credit Mart has discounted Anthropic credits available from companies with surplus allocations. Prices are typically 30-70% below retail.
In the Fellows Program, the compute funding is tied to the fellowship period, so once the four months are over, you should assume the budget ends with it rather than rolling forward.
No. Anthropic says to apply to only one track, so pick the one that best matches your interests and background.
Yes, but only if you are located within the US, UK, or Canada, and you still need work authorization in one of those countries because there is no visa sponsorship.
This is one of the better-funded on-ramps into serious AI safety or security research: real stipend, real compute, real expectations. If you get in, treat the four months like a focused sprint and ship something publishable.
Your AI credits are losing value every day
Join the marketplace and start trading unused credits today.