Anthropic Will Pay Staff Engineers to Switch Into AI Safety

Anthropic has opened applications for the Anthropic Residency, described on its own page as a paid, fixed-term research programme. It is aimed at people who already lead technical work somewhere else and want to move into empirical AI research full-time. The sentence doing most of the work is this one: "Background in machine learning is not required."

What it actually pays

The stated figure is $350,000 to $500,000 a year equivalent. Equivalent is the important word — it is an annualised rate, not the amount you receive. Placements run between six weeks and six months, so here is the same figure worked out per option.

Even at the bottom of the range, roughly $40,000 for six weeks is unusual for a programme that explicitly does not require experience in the field you are being paid to work in.

Who it is for

The page names three groups: professors in a quantitative discipline, technical founders, and staff-level engineers. The common thread is that you have already led ambitious technical work somewhere, rather than that you have done it in machine learning.

The stated expectations are that you are motivated by safe and beneficial AI, want to move into empirical AI research full-time, thrive in fast-paced collaborative settings, can implement ideas quickly and can communicate clearly. That last pair is worth reading as a filter: this is a programme for people who ship, not people who want to read about alignment.

What is explicitly not required

A PhD. Prior ML experience. Published papers. A machine learning background of any kind. Anthropic states all four. If you have been assuming you were not eligible for AI safety work because you lack the academic credentials, this programme is written to say otherwise.

What you would actually work on

Residents work on empirical research problems set out by Anthropic researchers and matched to the resident’s interests, with a paired mentor. Three focus areas are named: AI safety and alignment, AI security and frontier red-teaming, and ML systems and reinforcement learning.

The mentor pairing is the detail that distinguishes this from a short contract. You are not being hired to deliver something; you are being placed on an open problem with someone who knows the field.

What the page does not tell you

Three things are absent, and all three matter if you are weighing this against a job.

The selection process. No stages, interviews or timeline are described. You do not know how long you will be waiting or what you will be asked to do.

Cohort size. Not stated. There is no way to estimate your odds from the page.

What happens at the end. There is no guaranteed role. The stated aim is for residents "to continue working on AI research full-time, whether at Anthropic or elsewhere" — and those last two words are doing real work. This is a paid placement that may lead somewhere, not a hiring funnel with an implied offer.

Start dates are described as flexible rather than fixed, which helps if you are giving notice somewhere.

If you are going to apply

You have until 22 November 2026, which is eighteen days from the date on this article. Applications for this round close then.

On choosing a length. Six weeks is the lowest-risk option and the one most compatible with remote participation, which the page says is more likely for shorter placements. Six months is the one that plausibly changes your career, and it is also the one that requires you to leave what you are doing now.

On the lack of ML background. Take the page at its word but read the rest of the requirements carefully. Not requiring ML is not the same as not requiring depth — the bar being described is someone who has led hard technical work and can move fast, which is its own filter.

Frequently asked questions