Anthropic STEM Fellowship 2027

The Anthropic STEM Fellows Program pays US$100,000–$167,000 for four months of full-time work inside Anthropic, pairing your domain expertise with frontier Claude models to attack the hardest open problems in your field.

Applications close 28 October 2026. Offers go out by 20 November, and the programme runs 12 January to 8 May 2027.

Read this before you spend time on it

Two things rule most Indian applicants out, and both are stated in the posting:

"Existing authorization to work in the United States for the duration of the program."

"We are not currently able to sponsor visas for fellows. To participate in the Fellows program, you need to have or independently obtain full-time work authorization in the United States."

This is not a condition you sort out after being selected — it is a minimum qualification, screened at application. If you are in India without an existing US work visa, you cannot take this up. Anthropic does sponsor visas for its regular full-time roles, and the posting says explicitly that those policies do not apply to the Fellows Program.

Second: this is a job, not a grant. It is a fixed-term contract position, and Fellows are engaged through an employer of record rather than employed directly by Anthropic. The money is salary for four months of full-time work. Nothing goes to a company, no equity changes hands, and there is nothing to repay. If you are looking for capital for a startup, this is not that.

It is listed here because founders keep asking about it. The people it genuinely fits are PhD researchers already in the US — including Indian researchers on an existing visa, for whom it is one of the better-paid short fellowships going.

What it actually is

Anthropic is building toward "an AI scientist" — a system with the long-horizon reasoning and experimental judgement to push the frontier in real disciplines. Getting there needs people who have spent years at the frontier of those fields, which is what the fellowship buys.

You work in a small pod of Fellows from related fields. Your pod meets weekly with Anthropic researchers to share methods and pressure-test ideas. You get a grand challenge — an ambitious, open-ended problem in your field, scoped with their research team so four months is enough to make real progress — plus access to frontier Claude models and internal evaluation tooling.

Three responsibilities:

1. Set the grand challenge for your field — identify long-horizon open problems where Claude falls short of expert judgement, and define what real progress looks like.
2. Build datasets, simulations and hard-to-game evaluations that test whether models can plan experiments, interpret data and reason about mechanisms — systematically finding where they are confidently wrong.
3. Apply Claude to open problems in your domain, using test-time compute strategies and integrating the scientific tools your field relies on.

The first cohort was 20+ Fellows across mathematics, physics, the life sciences and engineering, running June to September 2026. Anthropic says the problems they built now feed Claude's training and one of its internal science evaluations — so the work is used, not shelved.

Who they want

• A PhD in a STEM field, plus several years of experience — post-doc or faculty level.
• Hands-on experience using Claude and Claude Code in your day-to-day workflow. This is a stated minimum qualification, not a nice-to-have.
• Scientific judgement over ML credentials. Prior machine learning experience is "helpful but not required" — they say judgement and a willingness to learn quickly matter more. If you are a strong domain scientist who has never trained a model, you are still in scope.
• Comfort in a fast-paced, high-agency environment, delivering concrete results in a few months.
• Existing US work authorisation.

The money and the logistics

How to apply

Apply through the Greenhouse posting before 28 October 2026. Offers are planned by 20 November.

Later stages typically include a take-home exercise and conversations with the research team about your research interests and possible project directions — so expect to defend a direction, not just a CV.

Anthropic publishes guidance on using AI in its application process; worth reading before you draft anything, given what the programme is.

One safety note from Anthropic itself: their recruiters only contact candidates from @anthropic.com addresses, and will never ask for money or banking details before your first day.