AI · 3 November 2026 · 4 min read

Anthropic is paying up to $500,000 a year for outsiders to work on safety. A machine-learning degree is not required

The residency is a fixed term of six weeks, three months or six months, for professors, technical founders and staff engineers. The page says residents will be paid $350,000 to $500,000 a year, as an equivalent rate. Applications close on 22 November. A paid term is not a permanent research job.

A glass and stone office building on a San Francisco street, with a postal van and university banners. — A street photograph published with the San Francisco Standard’s 2025 account of Anthropic expanding its office in the city. The residency prefers San Francisco. The picture is not an applicant, and it is not a contract.

Anthropic has opened applications for a residency aimed at people who already lead technical work somewhere else. The page names professors in a quantitative subject, technical founders, and staff engineers. It says a background in machine learning is not required, and that an applicant does not need a doctorate, prior machine-learning experience, or published papers. The work is the company’s: problems set out by Anthropic researchers and matched to the resident, in alignment, in security and frontier red-teaming, and in the systems and reinforcement learning that train the models. The stated aim is that, by the end, every resident is in a position to keep doing AI research full time, at Anthropic or somewhere else. A position to continue is not an offer of a job.

The terms are on the page. Placements run for six weeks, three months or six months, with the company asking for a preferred length and saying it will try to be flexible about the start. The pay sentence is exact. Residents will be paid $350,000 to $500,000 a year, as an equivalent. Equivalent means the rate, not a cheque for half a million for a six-week stay. Each resident is paired with an Anthropic mentor for regular one-to-one guidance, and the page says residents will work with frontier models and will have a dedicated computing budget for their projects. Mentorship on a page is a promise of time. It is not a named researcher, and a budget is not a cluster a reader can see.

The geography is a preference, not a rule for the short term. Anthropic prefers residents in its San Francisco offices. Remote work is possible, especially for the shorter placements. Most of the company’s full-time research roles, it says, still require a move to San Francisco. Applications for this round close on 22 November 2026. There is a US form and a UK form. The page does not say how many places there are. A deadline without a number of seats is an invitation, not a cohort.