Anthropic STEM Fellow

Anthropic•San Francisco, CA

12d•$100,000 - $167,000

About The Position

The Anthropic STEM Fellows Program brings science and engineering researchers and experts to Anthropic for a four-month fellowship. Fellows work in small pods with colleagues in related fields and Anthropic researchers to use Claude and internal tools to tackle important open problems in their disciplines. The work includes defining grand challenges, building datasets and simulations, evaluating model capabilities, and identifying where Claude falls short of expert judgment. Machine learning experience is helpful but not required; scientific judgment and willingness to learn quickly are emphasized. The program runs full-time from January 12 to May 8, 2027. Fellows attend a one-week in-person orientation in San Francisco. The default location is San Francisco, though remote engagements within the United States may be considered in certain circumstances. This is a fixed-term contract position, with Fellows engaged through an employer of record rather than employed directly by Anthropic. Some fellowships may be considered for extension. The four-month compensation is $100,000–$167,000 USD, equivalent to $300,000–$500,000 USD annualized, plus other benefits.

Requirements

Be excited about AI’s potential to reason about difficult questions in a field where expert judgment matters.
Have hands-on experience using Claude and Claude Code in a day-to-day workflow.
Hold a PhD in a STEM field and have several years of experience, such as postdoctoral or faculty experience.
Be comfortable working in a fast-paced, high-agency environment for a few months and delivering concrete results.
Have existing authorization to work in the United States for the duration of the program.

Nice To Haves

Prior machine learning experience is helpful but not required.

Responsibilities

Identify long-horizon open problems in the Fellow’s field where Claude falls short of expert judgment, and define what meaningful progress would look like.
Build comprehensive datasets, simulations, and rigorous, hard-to-game evaluations to test whether models can plan experiments, interpret data, and reason about mechanisms.
Systematically identify cases where models are confidently wrong.
Apply Claude to open problems in the Fellow’s domain, using test-time compute strategies and integrating the scientific tools and software used in that field.
Work with a small pod of Fellows on related problems and meet weekly with Anthropic researchers to share methods, pressure-test ideas, and build on one another’s work.