Anthropic Fellows Program: What It Is and How to Get In

If you've been hunting for a real way into AI safety research without a PhD or a stack of published papers, the Anthropic Fellows Program is probably already on your radar. It's a paid, four-month fellowship that pairs newcomers and experienced engineers alike with Anthropic researchers, then sets them loose on real empirical questions about how to keep advanced AI systems safe. No prior machine learning career required, just strong technical fundamentals and genuine motivation to work on the problem.

The idea behind the program is simple. AI safety research needs more hands than any single company can hire directly, and there are plenty of capable engineers who want to work on the problem but don't know how to break in. The fellowship gives them a structured, funded way to try it out, work alongside real researchers, and produce something they can point to afterward, whether or not they end up staying in the field long term.

What Is the Anthropic Fellows Program?

At its core, the program is Anthropic's answer to a talent bottleneck. There are plenty of sharp engineers and researchers who care about AI safety but have no clear path into the field. The fellowship closes that gap by offering funding, mentorship, and compute resources so fellows can spend four months working full-time on a single empirical research question.

Fellows aren't left to figure things out alone. Each one is matched with an Anthropic mentor who helps shape the project, unblock technical roadblocks, and push the work toward a publishable result, whether that's a paper, an open-source tool, or a public write-up. In the program's first cohort, more than 80 percent of fellows produced a public research output, and over 40 percent went on to join Anthropic full-time afterward.

It's worth noting that this isn't a classroom program or a set of assigned readings. Fellows pick their own project within a broad research area, shape the direction with their mentor, and are judged largely on what they actually produce. That structure tends to favor people who are comfortable with ambiguity and don't need a detailed syllabus to make progress.

The program has also grown since its first pilot cohort. Early rounds accepted a small group of 10 to 15 fellows working on a narrower set of safety topics. More recent cohorts have expanded both the number of fellows and the range of research areas on offer, reflecting how much the underlying safety problems have grown alongside the models themselves.

What Fellows Actually Work On

The research areas shift a bit from cohort to cohort, but they consistently sit under the broader umbrella of technical AI safety. Recent fellows have contributed to work on agentic misalignment, subliminal learning patterns, rapid response systems for new jailbreak techniques, and open-source circuit analysis.

Here's a quick breakdown of the main research tracks fellows typically choose from:

None of these tracks are set in stone. Fellows are encouraged to shape their project around a specific question within a track rather than working on something predetermined. Two fellows in the same research area, say, adversarial robustness, might end up with completely different projects depending on what gap they spot and what their mentor thinks is worth pursuing.

Pay, Compute Budget, and Location Requirements

One thing that sets this fellowship apart from a typical unpaid research internship is that it's genuinely well funded. Fellows aren't expected to cover their own living costs or scrape together compute credits from a free tier somewhere.

Because there's no visa sponsorship, applicants generally need existing work authorization in one of the three eligible countries. Remote work is allowed within those regions, but you won't be able to relocate internationally just to take part.

The compute budget in particular is worth pausing on. Fifteen thousand dollars a month in compute is enough to run genuinely serious experiments, not just small-scale tests on a laptop. That matters a lot for a program built around empirical research, since a lot of promising safety questions only become answerable once you can actually run the experiments at a meaningful scale.

Fellows who prefer working in person can use the shared workspaces in London or Berkeley, which also puts them physically close to other researchers and mentors. Remote fellows still get regular check-ins, but being in one of the hub cities can make the day-to-day mentorship feel a bit more natural.

Who Should Apply

Anthropic has been pretty clear that it cares far more about your ability to execute on research than your résumé. That said, a few traits show up again and again in strong candidates:

• Solid coding ability, especially in Python and common machine learning frameworks

• A track record of taking a vague, messy problem and making real progress on it

• Clear written communication, since public outputs are a core part of the program

• Genuine interest in reducing catastrophic risks from advanced AI, not just a resume line

• Comfort working independently for months at a time with periodic mentor check-ins

You don't need a published paper or a machine learning degree to qualify. Plenty of fellows come from adjacent fields like software engineering, physics, or applied math, and pick up the safety-specific context during the fellowship itself.

That said, the selection process is genuinely competitive. Anthropic has been open about caring more about demonstrated ability than credentials, but with a limited number of spots each cohort, most successful applicants have at least one concrete example of independent, self-directed technical work behind them, whether that's a research project, a substantial open-source contribution, or a technical writing sample that shows how they think through a hard problem.

How the Application Process Works

The process is fairly straightforward, though competition for each cohort is steep given the funding and mentorship on offer. Broadly, it breaks down into these steps:

1. Submit an application through Anthropic's official fellows page, including background and areas of interest

2. Complete any initial technical screening, which typically focuses on coding ability and problem-solving

3. Interview with Anthropic researchers to discuss potential project directions and research fit

4. Get matched with a mentor and a research area once accepted into a cohort

5. Begin the four-month fellowship, with regular check-ins and a target public output by the end

Cohorts don't run continuously. Anthropic opens applications on a rolling basis ahead of each new cohort start date, so it's worth checking the official page regularly rather than assuming a fixed annual schedule.

What Happens After the Fellowship

The four months aren't just a research sprint that ends with a polite goodbye. Anthropic has consistently converted a meaningful share of fellows into full-time hires, and many more have gone on to safety-focused roles at other organizations. Even fellows who don't land a full-time offer walk away with a public research output, direct mentorship from working researchers, and a much clearer sense of whether AI safety research is the right long-term path for them.

That combination, real funding, real mentorship, and a real shot at a full-time role, is a big part of why the program draws so much interest every cohort.

Even fellows who move on to a different company often keep working on safety research, just under a different roof. That ripple effect is arguably as important to Anthropic's mission as the direct hires, since the whole field benefits from having more trained safety researchers spread across different organizations rather than concentrated in just one place.

Common Mistakes Applicants Make

A surprising number of strong candidates get filtered out early for reasons that have nothing to do with their technical skill. Watching for a few common missteps can save you an easy rejection:

• Submitting a generic application that doesn't mention a specific research area or project idea

• Overselling credentials instead of showing concrete, verifiable work

• Skipping the technical screening prep and assuming raw talent alone will carry the interview

• Applying without confirming work authorization in the US, UK, or Canada ahead of time

• Waiting for a fixed annual deadline instead of checking the rolling cohort schedule

Most of these are easy to fix with a bit of planning. The application itself doesn't need to be long, but it does need to be specific about what you want to work on and why.

Tips to Strengthen Your Application

A few practical steps can make your application stand out before you even reach the interview stage:

• Have a couple of concrete project ideas ready rather than vague interests

• Point to any open-source contributions, side projects, or writing that shows how you think

• Keep your application materials clean and easy to skim

If you're pulling together a resume or writing sample as a PDF, our PDF tools can help you convert, compress, or merge files without losing formatting.

Applicants who want to touch up a headshot or diagram for a portfolio site often reach for our image tools to resize or clean up graphics quickly.

For polishing the writing sample itself, our text tools can help check word counts, formatting, and readability before you submit anything.

If your background is more engineering-heavy, our developer tools are handy for formatting code snippets or converting data files for a portfolio project.

And if you want to keep track of new cohort announcements and AI research trends, our latest blogs section covers updates like this on a regular basis.

Is the Program Worth Applying To

For anyone serious about a career in AI safety research, the Anthropic Fellows Program is one of the few paths that offers real funding, direct mentorship, and a genuine shot at a full-time research role, all without requiring a PhD or years of prior publications. The bar is high, but so is the payoff: a public research output, hands-on experience with frontier safety problems, and a network of researchers who can vouch for your work long after the four months are up.

If you have the technical fundamentals and the motivation, it's worth putting together an application the next time a cohort opens. Start early, pick a research area that genuinely interests you rather than one that just sounds impressive, and use the four-month structure as a real chance to test whether AI safety research is a field you want to build a career around. Whether or not you end up staying at Anthropic afterward, the experience, funding, and mentorship on offer here are hard to match anywhere else in the industry right now.