Anthropic Will Pay You $500K to Learn AI: The Anthropic Residency, Explained Box by Box

Anthropic just opened a job that pays up to $500,000 a year, and the posting says, in plain words, that you don't need a background in machine learning. It is called the Anthropic Residency, and your boss for a few months could be one of the researchers who built modern AI. The catch is who it is really for. The good news is a line at the very bottom of the page that most people scroll past, and it helps people who don't fit.

We read the announcement, the job posting and every box on the application form. Here is what it pays, who should apply, and what to type in the four boxes that decide it. Applications close November 22, 2026.

What the Residency is, and the real pay math

The Anthropic Residency is a paid research program. It lasts six weeks, three months or six months, and you pick which lengths work for you. The pay is quoted per year and pro-rated, which means you get the share that matches the time you are there. So here is the real math.

Based on $350K to $500K a year. The London posting quotes £265K to £375K a year.

The job is in San Francisco or New York, with a London version too. You can do it remotely, mostly for the shorter stays. You get a mentor, access to Claude models and your own GPU budget (GPUs are the chips AI research runs on). The goal Anthropic wrote down is that every resident leaves able to do AI research full time, at Anthropic or somewhere else.

The catch: who it's really for

This is not for beginners. The posting names three kinds of people: a professor in a number-heavy field, a technical founder, or a staff-level engineer. What they are buying is someone who has already led something hard. They will teach you the AI part.

So ask yourself three things:

• Have I led a big technical project, start to finish?
• Can I point to something I built or published that people used?
• Could two people who worked with me back that up?

If that is three yeses, the form walkthrough below is for you. If not, keep reading anyway. The section on the line at the bottom of the page is for you.

The mentors

Honestly, the money isn't the best part. The mentors are. There are nine names on the list. A few of them:

• Jascha Sohl-Dickstein invented diffusion models, the idea behind most of the AI image generators you have used, and was a principal scientist at Google Brain.
• Nicholas Carlini spent years at Google Brain and DeepMind finding ways to break AI models, and has won best-paper awards at ICML three times.
• Jack Lindsey runs a team that looks inside Claude's brain for things like introspection and emotion.
• Evan Hubinger leads the team whose job is to try to make Claude misbehave before anyone else can.

People pay a lot to sit in a lecture hall with researchers like this. Here, you would get one-on-one time with one of them, and they would be paying you.

What you'd work on

The work falls into four buckets, and they match the four choices in the mentor-area dropdown on the form:

• Safety: making sure AI stays honest, even when it gets smarter than us.
• Interpretability: looking inside the model to see why it does what it does.
• Security: attacking the model on purpose. One past project had AI agents find $4.6 million worth of bugs in crypto contracts.
• Training systems: making the machine that builds the models better.

Pick the one closest to what you already know. A cryptography professor fits security. An infrastructure engineer fits training systems. That match is half your application.

The form: 26 boxes, most of them take seconds

The application has 26 boxes. Most are quick. A few tips on the easy ones:

• Resume or LinkedIn: you need at least one of the two. Give both.
• Start date: type a real month. If you are a professor who can only do summer, say so in the timeline box. They ask because they want to work around it.
• Residency length: these are checkboxes. Tick every length you can actually do. More boxes give them more ways to fit you in.
• The two visa questions: tell the truth. Three months or longer and they sponsor. If you need a visa, don't tick only six weeks.
• Arbitration boxes: legal sign-offs. Read them and click.
• AI policy box: a yes. It changes how you write the next part, so more on that below.

The 4 boxes that decide it

1. Why Anthropic

The form itself says they value this one highly, and that great answers run 200 to 400 words. Use four parts:

1. The moment you started caring about AI safety. Make it a real moment, not "I've always been passionate."
2. The hardest thing you have led, and the skill from it that carries over.
3. One paper or one team at Anthropic, by name, and why it grabbed you. Their research directions page is linked right in the posting, so read it.
4. Where you want to be in two years.

2. Your notable work

They ask for links and what you did on each. So write one line per link: what it is, what you personally did, and the result as a number. "Led the team that cut payment fraud by 40%" beats "worked on fraud."

3. The research you'd do

This is where most people go vague. Use one sentence shape: I know this. Your team works on that. Here's what I'd test.

A payments engineer might write: "I've built fraud systems for ten years. Your safeguards team catches misuse. I'd test whether fraud-style anomaly detection can flag a jailbreak before it works." It is specific, it is yours, and nobody else can write it. Then pick the mentor area that matches.

4. References

Two or three people, one per line, in the exact format the form shows. And relax: they won't contact anyone until later rounds, and they will tell you first. Just give those people a heads-up.

The additional information box is optional. Use it only if something big doesn't fit anywhere else.

Can you use Claude to write it?

Anthropic actually answers this in its AI guidance for candidates: write your first draft yourself, then use Claude to make it better. The stories have to be yours.

The prompt we would use tells Claude not to add anything you didn't write, and to ask you for real examples wherever you are vague. That is how you use AI here without breaking their rule. One more thing: the take-home test and the interviews are all you, with no AI unless they say otherwise.

The line at the bottom of the page

At the very bottom, the posting says applicants may also be considered for the Anthropic Fellows Program. That is a different program, built for strong technical people who don't have the big title yet.

Fellows pays $3,850 a week for four months, plus about $15,000 a month in compute. In past rounds, a quarter to half of fellows got full-time offers. So even if you are not a professor or a founder, a good Residency application can still end up in front of the Fellows team.

To be fair, there is a limit: Fellows don't get visa sponsorship, so you need the right to work in the US, UK or Canada.

Deadline and what happens next

Applications close November 22, 2026 at 11:59 PM Pacific. After that, it is a recruiter call, a technical test, a research conversation and references. Offers go out in January.

Get the free application kit

We put all 26 boxes in one free kit called The Residency Application Kit: what to type in each, the four-part Why Anthropic worksheet, a research question finder for your background, and the Claude prompt that follows Anthropic's AI rule. Comment RESIDENT on the video, or grab it below. And if you know a professor, a founder or a senior engineer who has been curious about AI, send them this. It might be the best email they get this year.

Not affiliated with Anthropic. All sources checked November 3, 2026.