Interview Playbook / High-Growth Tech & AI

How Anthropic Interviews in 2026: Process, Questions & What They Score

HOW TO READ THIS PLAYBOOK

Compiled from 17 public sources: candidate interview reports, coaching guides, and Anthropic's own hiring pages. Interview processes change and vary by role, team, level, and region. This is one well-documented shape of Anthropic's interviews to prepare against, not a script of what your interview will be. Confirm specifics with your recruiter. SupaCV is not affiliated with or endorsed by Anthropic.

High confidenceLast verified JUL 202617 sourcesSources

Anthropic (AI safety/research lab, maker of Claude) is best known for hiring software engineers, research engineers, and research scientists, alongside growing GTM/PM functions. Its process is documented both by the company itself (careers page, "candidate AI guidance" page) and by third-party sources (IGotAnOffer, Glassdoor, Exponent, interviewing.io, InterviewQuery, and candidate blogs). Two features are distinctive and consistent across independent sources: (1) a standalone "values/culture" interview that appears in every loop regardless of role or level and is repeatedly described as unusually personal and probing rather than a typical behavioral round; and (2) an explicit, published, stage-dependent policy on AI-tool use (allowed for polishing applications and for interview prep, generally disallowed on take-homes, and strictly banned in all live interviews). Reported total timeline varies: Glassdoor's aggregate average across all roles is about 19 days, but engineering-focused coaching guides more commonly cite 3-6 weeks, and PM candidates report anywhere from roughly 3 weeks up to 4-8 weeks depending on the source. Glassdoor's difficulty rating is moderate-to-high (~3.26/5), with AI/ML Engineer roles cited as among the hardest. Exact stage count, question content, and pacing vary meaningfully by team, level, and role (SWE vs. Research Engineer vs. Research Scientist vs. PM have materially different loops); the sequence below is the commonly reported pattern, not a fixed universal script.

The Process

  1. 1

    Recruiter / talent screen · ~30 minutes

    Phone or video call with a recruiter covering background, experience, role interest, logistics, and often an early "why Anthropic" / mission-alignment question. Sometimes split into two shorter calls.

  2. 2

    Technical assessment (coding OA or take-home) · 60-90 minutes

    For engineering roles: a live or async coding exercise, commonly on CodeSignal, in a shared Python environment (Colab/Replit also reported). Usually structured as one problem split into 3-4 progressively harder levels that add requirements as you go (e.g., building out a banking system with multiple transaction types, or converting profiler stack samples into start/end trace events); it rewards clean, extensible code under changing specs rather than classic LeetCode algorithm recall. Research Engineer variants can involve implementing a Transformer component (e.g., multi-head attention) in PyTorch or debugging training code.

  3. 3

    Hiring manager screen · 45-60 minutes

    Conversation focused on engineering/product judgment: walking through past projects, architectural decisions and tradeoffs, and sometimes a lightweight code-review discussion. For PM roles this stage centers on products led, KPIs, and a past-failure question.

  4. 4

    Onsite / virtual onsite loop · Roughly 4-5 hours total across the loop

    A batch of interviews, commonly reported as split into two gated mini-loops (not every candidate report describes this hard-gate structure, so treat it as commonly reported, not universal): the first half typically covers one or two coding rounds (Python, concurrency-heavy) and a system design round (frequently framed around LLM inference infrastructure: request batching, GPU memory/KV cache, high-throughput token-generation services). Candidates generally need to clear this first half before the second half, usually a project/architecture deep-dive plus the standalone values/culture round (see the dedicated stage below), is scheduled at all. PM loops instead include a product/business case and a cross-functional panel in place of the coding/system-design rounds.

  5. 5

    Culture / values interview · ~45-60 minutes

    Typically the final round within the onsite loop (often gated behind the coding/system-design rounds described above) rather than a separate day tacked on after the loop concludes. It gets its own entry here because independent sources consistently single it out as the highest-stakes, most distinctive round in the whole process. Run by a non-technical interviewer, it appears in essentially every Anthropic loop regardless of role or seniority. It's described by many candidates as closer to a probing, reflective conversation than a standard behavioral interview: exploring how the candidate reasons about risk, ambiguity, moral tradeoffs, and long-term consequences, and testing for genuine, considered engagement with AI safety rather than generic enthusiasm. Candidates consistently report that rehearsed, polished STAR-format answers read poorly here; honesty about uncertainty and having changed one's mind is valued.

  6. 6

    Reference checks and team matching · 1-2 weeks

    After the loop, Anthropic runs reference checks and works to match the candidate to an actual team; multiple sources note candidates can be turned down at this stage for lack of team fit even after passing all interviews. A reapplication cooldown is reported for candidates who don't advance, and its length varies by how far you got: roughly a few months (commonly cited around 3) for an early-stage rejection, versus 6-12 months after a later-stage or final-round rejection.

  7. 7

    Offer and negotiation · Varies

    Verbal/written offer with compensation discussion. Public aggregators disagree substantially on exact numbers, so treat any figure as directional rather than a firm band: junior/mid-level engineering base salary is commonly cited around $175K-$230K, while the $300K-$490K+ figures often quoted for senior/staff engineering levels typically represent total compensation (base + equity + bonus), not base salary alone; some aggregators (e.g., Levels.fyi) show total comp for senior/lead engineering levels running considerably higher still (roughly $550K-$840K+). Confirm with your recruiter whether any number they give you is base or total comp before anchoring on it.

Evaluation Framework

Anthropic's published values ("What we value and how we act")

Act for the global goodHold light and shadeBe good to our usersIgnite a race to the top on safetyDo the simple thing that worksBe helpful, honest, and harmlessPut the mission first

This 7-item list is publicly documented, verbatim, on Anthropic's own website, confirmed via two independent direct fetches. It appears on anthropic.com/company under the heading "What we value and how we act," and the identical seven items also appear in full on anthropic.com/careers, though that page introduces them under a different heading ("Principles that guide how we show up for each other and our mission"), not the same phrase. The content and count match exactly on both pages; only the heading wording differs. Unlike Amazon's Leadership Principles, Anthropic does not publish a formal per-value interview rubric or say each value maps to a specific question, but candidate reports consistently describe the dedicated values/culture round (present in every loop) as the place these values (especially safety/risk reasoning, honesty, and mission commitment) are explicitly probed. One secondary source (a candidate's Substack account, not an official document) describes the culture round as informally assessing four things: tolerating complexity without collapsing it into a tidy narrative, acknowledging what you don't know (epistemic humility), reasoning about second-order/systemic effects, and intellectual honesty even when it's unflattering. This framing is candidate-reported/unofficial and should not be treated as an Anthropic-published rubric the way the 7 values are.

Sample Interview Questions

Coding / hands-on implementation7 questions
  • Build out core business logic for a banking system that must handle multiple transaction types, with requirements added incrementally as the interview progresses.
  • Write a function that converts a list of profiler stack samples (timestamps + call stacks) into a list of start/end trace events for each function call.
  • Given stack trace samples, determine the longest-running function, including handling edge cases like recursive or infinite-loop-like patterns.
  • Implement a data structure or algorithm from scratch (e.g., a specific hash map or parser) and be ready to defend your time/space complexity choices under questioning.
  • (Research Engineer variant) Implement a Transformer component such as multi-head attention from scratch in PyTorch.
  • (Research Engineer variant) Debug a broken/incorrect model training loop and explain what's wrong and why.
  • Refactor your existing solution cleanly as new constraints or requirements are layered on mid-interview, without a full rewrite.
System design (LLM infrastructure-flavored)6 questions
  • Design an inference batching system for a single GPU that can process up to 100 inputs per batch, where users submit requests synchronously and wait for results.
  • Design an end-to-end request pipeline for LLM queries: intake, batching, GPU routing/load balancing, and returning responses to the correct caller.
  • Design a token-generation service that must handle very high request volume: reason about throughput, horizontal scaling, and request routing.
  • How would you manage GPU memory (e.g., KV cache) under many concurrent, variable-length generation requests?
  • How would you support streaming responses and a priority-based request queue in an LLM-serving system?
  • Design the backend architecture for a consumer-facing chat product like Claude.ai.
AI safety / mission alignment6 questions
  • Why do you want to work at Anthropic specifically, as opposed to another AI lab (e.g., OpenAI, Google DeepMind)?
  • What are your thoughts on AI safety and the risks posed by advanced AI systems?
  • Which of Anthropic's stated values resonates most with you, and why?
  • What are the ethical risks of deploying agentic AI systems in high-stakes environments?
  • How would you handle being assigned to work on a project you personally believed was unsafe or premature to ship?
  • What's a specific, considered (not generic) point of view you hold about how AI development should be governed or paced?
Values / behavioral (culture round)7 questions
  • Tell me about a time you did something that conflicted with your own values.
  • Tell me about a time you pushed back on a decision and lost: how did you handle it afterward?
  • Describe a time you changed your mind about something you previously felt strongly about.
  • Tell me about a time you worked on something you had moral or ethical reservations about.
  • Walk me through a project you're proud of and the tradeoffs you navigated to get there.
  • Who do you respect but disagree with on values, and why?
  • Describe a time you made a decision under significant uncertainty with no clearly right answer.

Coach's Tips

The values/culture round is reported as the hardest stage to pass and shows up in every loop: prepare emotionally honest, specific stories about moral gray areas, being wrong, or losing an argument rather than polished, pre-packaged STAR narratives; multiple sources explicitly call rehearsed STAR answers a top failure mode here.

Go beyond generic AI-safety enthusiasm: read Anthropic's own public materials (e.g., its published research/safety posts and leadership essays) before interviewing, and be ready to articulate a specific, considered opinion on AI risk and governance; interviewers are reported to probe for genuine engagement, not applicant excitement.

For coding rounds, practice building and incrementally refactoring working software against a spec that changes mid-exercise (the CodeSignal-style multi-level format), rather than only grinding isolated LeetCode-style algorithm problems.

For system design, expect prompts framed around LLM-specific infrastructure (batching, GPU/KV-cache memory, high-throughput token generation, streaming); if the terminology is unfamiliar, translate it into general distributed-systems concepts (queuing, load balancing, async processing) you already know.

Follow Anthropic's own published AI-usage policy precisely: it's fine to use Claude to refine an application draft you wrote yourself and to prep for interviews, but take-homes should generally be done without AI assistance unless told otherwise, and live interviews are reported to have a strict no-AI-assistance rule with real consequences for violations.

Treat the hiring-manager conversation and post-loop period as part of the evaluation, not a formality: reference checks and team-matching happen after the interview loop, and candidates have reportedly been turned down at that stage for lack of team fit even after clearing all interviews.

If you're given a specific compensation number during the process, ask explicitly whether it's base salary or total compensation before you anchor on it; public sources conflate the two constantly, and the gap between a senior-level base and total comp figure can be well over $100K.

Common questions

How many stages are in Anthropic's interview process?+

Anthropic's process has 7 stages, in order: Recruiter / talent screen, Technical assessment (coding OA or take-home), Hiring manager screen, Onsite / virtual onsite loop, Culture / values interview, Reference checks and team matching, Offer and negotiation.

What framework does Anthropic use to evaluate candidates?+

Anthropic evaluates candidates against Anthropic's published values ("What we value and how we act"): Act for the global good, Hold light and shade, Be good to our users, Ignite a race to the top on safety, Do the simple thing that works, Be helpful, honest, and harmless, Put the mission first.

What kinds of questions does Anthropic ask?+

Anthropic's question bank spans 4 categories: Coding / hands-on implementation; System design (LLM infrastructure-flavored); AI safety / mission alignment; Values / behavioral (culture round).

How reliable is this Anthropic interview playbook?+

This playbook is high confidence, compiled from 17 public sources, and last verified July 7, 2026. It describes one well-documented shape of Anthropic's interviews, not a guarantee of what any individual loop will look like.

Sources

Interviewing at Anthropic?

Add it as a Target Role and tailor your resume against the actual job description.

Add Anthropic as a Target Role