Interviewing at Anthropic: what’s official, what’s rumor, and where AI is banned
In May 2025 Anthropic told job applicants not to use AI anywhere in its hiring process. By July it had reversed part of that, as Fortune reported on 21 July 2025, and published a stage-by-stage policy that is still the clearest rulebook any AI lab has given candidates. If you’re researching the Anthropic interview process, that policy matters more than any round-by-round leak, so this guide starts there.
A note on method. We read Anthropic’s careers page and its candidate AI guidance (marked “Last updated Jul 10, 2025”) on 31 October 2026. Anything below that comes from those pages is labelled official. Anything that comes from candidate reports is labelled as such, because Anthropic hasn’t confirmed it and a lot of what circulates online is guesswork.
Can you use AI in the Anthropic interview process?
Only in some stages. Anthropic’s guidance says you may use Claude to refine an application after writing the first draft yourself, and to research the company and practice answers. Take-home assessments are done without Claude unless Anthropic says otherwise. Live interviews allow no AI assistance unless the interviewer tells you it’s permitted.
Laid out as a table:
| Stage | AI allowed? | Anthropic’s wording, paraphrased |
|---|---|---|
| Application (resume, written answers) | Yes, after your own first draft | Write the first draft yourself, then use Claude to refine it |
| Interview prep | Yes, encouraged | Use Claude to research Anthropic, practice answers and prepare questions |
| Take-home assessment | No, unless stated | Complete it without Claude unless told otherwise |
| Live interviews | No, unless stated | It’s all you, with no AI assistance unless told otherwise |
Anthropic also lists how it uses Claude on its side: writing job descriptions, developing interview questions, drafting candidate communications, analyzing hiring metrics, transcribing interviews and identifying candidates to source. It says it doesn’t use your data to train Claude and doesn’t let Claude make hiring decisions. We think publishing that list is the right call (and it’s more than most employers tell you), because it makes the “be transparent with us” request feel like a two-way deal.
For a site that makes a live interview assistant, this is the awkward section, so we’ll be direct. At Anthropic, using a real-time assistant in a live round breaks a rule you were told about in writing. Don’t. Use AI as hard as you like in prep, which Anthropic explicitly invites.
What the official pages say about the interviews themselves
Not as much as you’d hope, but more than nothing. From the careers page:
- All interviews run over Google Meet.
- Technical roles use live coding tools, and Anthropic names two: Colab and CodeSignal.
- On technical rounds you’re allowed to look things up; the page asks that you be comfortable enough with basic syntax and standard libraries that lookups don’t eat your time.
- Non-technical interviews are described as conversational.
- Recruiters only contact candidates from @anthropic.com addresses (worth knowing, given how many recruiting scams use AI-lab names).
- On offers, the page simply says to take your time.
a quick word on fake recruiters
The @anthropic.com detail deserves more than a bullet. Anthropic calls this out on its own careers page, in a section about scams, which is a decent hint that fake offers using its name are a real and recurring problem. If someone contacts you about the Anthropic interview process from a Gmail address, a Telegram account or a lookalike domain, stop there. Real recruiters won’t ask you to pay for equipment, install unusual software or move to a messaging app before a single interview has happened. When in doubt, find the role on the official careers page and apply there directly; if the message was real, the recruiter will find your application.
The same page lists what Anthropic cares about. It says it cares “about what you can do, not where you learned to do it,” and suggests putting independent research, blog posts and open-source work at the top of your resume. It also lists principles including “Do the simple thing that works,” “Be helpful, honest, and harmless” and “Put the mission first.”
The “look things up” line is a bigger deal than it sounds. Plenty of companies still run coding rounds as closed-book memory tests. Anthropic is telling you it wants to watch how you work, not whether you’ve memorized the argument order of a standard library call.
What candidates report (unverified)
A candidate-written guide published by Aced describes one or two recruiter screens, a remote technical screen of about 60 to 90 minutes, and onsite loops spread over two separate days that cover coding, system design, behavioral and culture questions, plus a deep dive into a past project. The same guide says most candidates report one to four months from application to offer. We can’t verify any of it, and Anthropic’s own pages don’t list a fixed number of rounds, so treat those specifics as “possible” rather than “expected.”
Three things we’d take from the reports anyway. Expect at least one coding round in CodeSignal or Colab, since Anthropic names both. Expect a project deep dive, which rewards having one piece of work you can talk about for 40 minutes without notes. And expect to be asked why you want to work on AI safety specifically, with a follow-up if your answer sounds generic.
the project deep dive is where prep pays off most
Pick the project now, not the night before. Write down what you built, what broke, what you’d change, and one decision you made that a reasonable colleague would have argued against. Then have someone who doesn’t know the project ask “why?” five times in a row. If the fifth answer still holds up, you’re ready. If it falls apart at the third, that’s the part an interviewer will find too. It’s an uncomfortable exercise, which is the point.
It also helps to know the numbers in your own story. Latency before and after, how many users, how long the migration took. Interviewers notice when someone has to guess at figures from their own work.
How to prepare, stage by stage
The application is the one place AI help is officially fine, with a condition. Write your own draft first. If Claude rewrites it into something that no longer sounds like you, that defeats the point; use it to tighten, not to author.
For coding, practice inside the tools Anthropic names. Colab and CodeSignal both behave differently from a local IDE (autocomplete, run buttons, how output shows). An hour in each before the interview is cheap insurance. Our coding interview patterns guide covers the problem shapes worth drilling, and since “do the simple thing that works” is a stated principle, we’d lean toward clear, working code before clever code.
For senior roles, prepare for design conversations too. Our system design interview guide is the general version. Anthropic doesn’t publish a system design round format, so don’t over-index on a specific template.
For the mission and values conversation, read Anthropic’s published principles and decide which ones you actually agree with. That sounds obvious. It’s also the part people skip. If you disagree with something, saying so thoughtfully is probably better than reciting the list back. (That’s our opinion, and we can’t prove it, but “Be helpful, honest, and harmless” is an odd value to fake your way through.) The STAR method guide helps with structuring the story-based questions.
For live practice, mock rounds are where a real-time tool earns its keep. Craqly runs on Mac and Windows and gives you 15 free minutes, no card; use them on a practice call with a friend asking you Anthropic-style questions, then switch it off for the real thing. If you’re unsure where the line sits in general, we wrote an honest answer on whether AI interview assistants count as cheating, and Anthropic’s policy is about as clear an example of the line as you’ll find.
How it compares to other big-company loops
Most large tech employers say far less about AI in writing. Google is the useful contrast. In March 2025 CNBC reported that Sundar Pichai had suggested at a February town hall that hiring managers consider bringing back some in-person interviews, with employee concerns about AI-assisted cheating as part of the backdrop. That’s a response to the problem through format, not through a published candidate policy. Our Google interview process guide covers that loop in detail.
Anthropic went the other way: keep interviews remote on Google Meet, allow lookups on technical rounds, and write down exactly where AI is and isn’t allowed. Our view, and it’s only a view, is that the written rule is better for candidates. You know where the line is, and “I didn’t know” stops being a defense for anyone.
Anthropic says it plans to review the guidance regularly “to reflect evolving AI capabilities.” Given that it reversed its own blanket ban within roughly two months in 2025, we wouldn’t be surprised to see live rounds that allow Claude on purpose before long. Check the guidance page the week of your interview, not the week you applied.