We Read All 19 Anthropic PM Job Descriptions. Here Is What They Ask For

As of September 2026, Anthropic has 19 product manager job descriptions in our catalog, and we read every one of them in full. The short version: the biggest cluster is Safeguards, trust and access (6 roles), followed by Research and Labs (5). Every posting states a US annual salary: $305,000 to $385,000 (10 roles), $385,000 to $460,000 (8 roles), or $305,000 to $460,000 (1 role). Every role is listed in San Francisco, New York, Seattle or Washington, DC. What the postings ask for most is comfort with ambiguity (15 of 19), working with research teams (14), safety and risk judgment (14) and evals (10). The classic PM toolkit barely appears: SQL is named in 3 postings and A/B testing in 1. If you want an Anthropic PM job, build proof that you can ship into the unknown with researchers and make safety calls, then practise on the posting itself.

allthingspm.app is an AI PM course and interview prep platform with AI mock interviews built from any job description. Every Anthropic role below links to its page in our jobs catalog, where a mock interview is built from that posting's own text.

How did we analyse Anthropic's PM job descriptions?

We pulled every Anthropic posting from the allthingspm.app jobs catalog on 25 September 2026. The catalog read them from Anthropic's own Greenhouse careers board on 22 September 2026. There were 19. We then:

1. Read each body in full, not just the bullet lists, and separated the role-specific text from the boilerplate every posting shares (the "About Anthropic" paragraph, logistics, and "How we're different"). That boilerplate is identical across all 19 apart from the salary line.
2. Hand-coded each posting for team, locations, stated years, salary band, and 14 requirement themes, recording whether each appears as a minimum or a preferred qualification.
3. Checked for duplicates. Four Safeguards postings share just two texts (Generalist and Account Integrity & Abuse are word for word the same; so are Cyber and Rare Harms, and the two pairs differ only in formatting). The two Research postings differ only in formatting and one subheading. We still count 19 postings, because Anthropic lists 19 separate jobs, but keep in mind that some requirements are counted four times because one team wrote them once.
4. Checked the live board on 25 September 2026. It listed 624 open jobs, 17 of them product manager roles, and all 17 are in our set. The other two (Growth and Beneficial Deployments (Labs)) had come down from the board by then. We kept them in the study because they were posted this month and show what Anthropic was hiring for.
5. Compared with the market by running the same literal text matches over the 370 other PM postings in our latest job-description harvest, from 85 other AI companies, read on 22 September 2026.

Product managers are a small slice of Anthropic's hiring: 17 of 624 open jobs, under 3 percent, on the day we checked.

What PM roles is Anthropic hiring for?

Here are all 19, with the salary each posting states and the experience figure it names. Every role title links to its page in our jobs catalog, which has the full posting and a mock interview built from it.

Salary, location and years are exactly as each posting states them on Anthropic's careers board. The team families are our grouping, not Anthropic's.

Which Anthropic teams are hiring the most product managers?

Safeguards, trust and access: 6 roles

This is the largest group, and it is the clearest signal of what makes Anthropic different. The four Safeguards postings (Generalist, Account Integrity & Abuse, Cyber, Rare Harms) describe a PM who works "with our research and product teams to develop detections, evals, interventions, and tools to measure and mitigate deployment and user risks" across Claude.ai, the first-party API and external cloud providers. One responsibility reads, in full: "Ability to write safety evals and communicate externally about safety."

The other two are platform roles with a safety core. Safe Access owns identity, verification and access primitives that "sits on the critical path of every model launch and every access program". Multi-Cloud Trust & Safety owns how safeguards, fraud defences and compliance "show up on every cloud" (Amazon Bedrock, Google Cloud's Agent Platform and Microsoft Foundry) and wants someone "equally fluent in policy nuance and systems detail". A good practice prompt for this family: a new Claude capability unlocks major user value but creates a rare, high-severity misuse risk; how would you design the detection, evaluation and intervention system?

Research, Labs and human data: 5 roles

The Research and Research (Code) postings are the most technical: "working proficiency in Python and SQL" and "a strong grasp of AI/ML concepts". Their job is to "partner with research to define, improve, and ship model capabilities within a research area" and to "synthesize user insights into actionable requirements and evaluations".

The two Labs roles (Research PM, Labs and Beneficial Deployments (Labs)) are 0-to-1 roles. Both name Claude Code and MCP as examples of what Labs produces, and both ask you to "build prototypes yourself to validate ideas before committing resources". Anthropic's own Labs announcement from January 2026 says Claude Code went "from a research preview to a billion-dollar product in six months". Beneficial Deployments adds an unusual profile: someone "who has founded something, worked inside a civic institution or beneficial organization, and shipped inside a large technology company".

Human Data Platform is the least glamorous title and one of the most interesting jobs: tooling for the data that trains Claude. It wants a PM who will "sit in on crowd worker and vendor sessions" and track KPIs such as "time-to-launch for new data collection projects". The practice question we built from it: design a KPI framework for Anthropic's Human Data Platform that connects platform health to research outcomes.

Claude products and growth: 4 roles

These are the named product lines. Claude Science is the most demanding on domain: people who can "hold their own in a conversation with a structural biologist or a computational chemist, get in the weeds of model transcripts and evals, and still zoom out to help define a product category". Claude Tag brings a multiplayer agent from Slack to new collaboration surfaces; the posting says "our internal version of Claude Tag writes 65% of merged PRs for our products". Cybersecurity leads the security portfolio, including vulnerability discovery in Claude Code and Claude Security, which trade press reported went into public beta for Claude Enterprise customers in May 2026. Growth is the most conventional posting of the 19, and the only one that names A/B testing.

Platform, cloud and public sector: 3 roles

Multi-Cloud Growth, Google and New Markets and Monetization are commercial roles. Both want someone who has owned a number: "a revenue number, a margin target, or a channel P&L" in the first, and "revenue or activation" in the second. New Markets is hiring its founding PMs and asks for a view on "what changes about billing, identity, or distribution when the customer is sometimes an agent". Public Sector is the only role based outside the three main offices (remote-friendly or Washington, DC). It requires that you "are a US person", and some of its portfolios need security clearance eligibility.

Internal tools: 1 role

Business Technology is the first PM for Anthropic's home-grown internal collaboration platform. It is the longest posting and the most explicit about judgment: "Design for where models will be in 3,6,12 months, not where they are today. Assume Claude is a participant in collaboration, not a feature bolted onto it." It also asks for someone who has "slowed down or changed course for safety reasons before".

What does an Anthropic product manager get paid?

Every one of the 19 postings states its pay. Each says "The annual compensation range for this role is listed below" and then gives an "Annual Salary" in USD:

Three things are worth knowing about these numbers.

They are base salary, not total compensation. California requires employers with 15 or more employees to put a pay scale in every job posting, and since January 2026 (SB 642) that scale must be "a good faith estimate of the salary or hourly wage range that the employer reasonably expects to pay for the position upon hire". Bonuses and equity are not required. New York's Labor Law 194-B has a similar rule for jobs performed in the state. The postings mention equity only indirectly ("optional equity donation matching" in the benefits line). Levels.fyi shows self-reported Anthropic PM total compensation of $349K to $486K a year, but on the day we checked its page said it needed more submissions to unlock the full breakdown, so treat that as a thin sample.

The band does not follow the years. Claude Science prefers 8+ years and pays $305,000 to $385,000. Safeguards (Generalist) prefers 5+ and pays $385,000 to $460,000.

Identical text, different bands. Safeguards Generalist and Account Integrity & Abuse pay $385,000 to $460,000. Safeguards Cyber and Rare Harms, with essentially the same text, pay $305,000 to $385,000. The postings do not explain why. Our reading is that the band reflects an internal level, which fits the logistics line in every posting: "Years of experience required will correlate with the internal job level requirements for the position." That is an inference, not something Anthropic states, so ask your recruiter which level a role is scoped at.

Where are Anthropic's PM roles based?

Most postings list two locations (11 of 19). The shared logistics text says: "Currently, we expect all staff to be in one of our offices at least 25% of the time. However, some roles may require more time in our offices." So "remote-friendly" appears on only two roles: Public Sector and Business Technology.

Two details to check with a recruiter. The Claude Tag posting lists San Francisco and New York, but its last line reads "San Francisco and Seattle only". And although Anthropic's board lists jobs in cities such as London, Dublin, Tokyo and Zürich, none of the PM roles is based outside the US. On visas, every posting says: "We do sponsor visas! However, we aren't able to successfully sponsor visas for every role and every candidate." Public Sector is the exception, with its US person requirement.

How senior do you need to be for an Anthropic PM role?

None of the 19 titles says Senior, Staff, Principal or Lead. They are all "Product Manager" (or "Product Management"), and the seniority shows up in the years and the pay band instead.

The lowest figure anywhere is five years. There is no associate or entry-level PM posting in the set. But look at where the years sit:

• 13 postings put the years only under preferred qualifications.
• 3 postings put years in the minimum: Multi-Cloud Trust & Safety ("Have 10+ years of experience"), and Safe Access and New Markets, which both say five or more years is what "most people" need but add "we care about the ownership, not the years".
• Human Data Platform puts 5+ years in an "ideal candidate" list, and Claude Tag has a single "good fit" list with 8+ years in it.
• Business Technology names no years at all and says it hires "on demonstrated impact", listing what you have shipped and moved rather than "scope owned, team size, or tenure".

Every posting also carries this line: "We encourage you to apply even if you do not believe you meet every single qualification." The pattern is consistent. Anthropic sets the bar with ownership and judgment, and treats years as a proxy it is willing to waive.

What skills and experience do Anthropic PM postings ask for?

We coded the role-specific text of each posting for 14 themes. A posting counts once per theme, whether the theme appears in responsibilities, minimum or preferred qualifications.

Three findings stand out.

Evals are a PM skill here, not an engineering one. Ten postings put evaluation in the PM's hands. Safeguards PMs write safety evals. Claude Science PMs "build and shape evals grounded in real scientific workflows". Cybersecurity and Public Sector both want "Fluency with AI products: evaluations, model behavior as part of the product surface". If you have never designed an eval set or a launch gate, that is the single most useful gap to close. Our evals chapter is built around exactly that.

Research is your main partner. Fourteen postings name research teams as partners or users. The Human Data posting even lists "Researcher EQ: Deep understanding of how researchers think and what motivates them" as a virtue. This is a different muscle from working with engineers, and we cover it in the lesson on working with researchers.

The classic PM analytics stack is nearly absent. SQL appears in 3 postings, Python in 2, A/B testing in 1. That does not mean Anthropic PMs do not use data. New Markets asks you to "Pull your own data: SQL or equivalent, to define a metric, build a funnel, or size a market". It means the postings assume you can, and spend their words on judgment instead.

Several roles require real domain depth. Beyond the themes above, some postings want specific domain experience: security for Cybersecurity ("enough technical depth to read a finding and tell whether it's real"), science for Claude Science, government for Public Sector, hyperscaler partnerships for the two multi-cloud roles, and billing or identity systems for New Markets and Safe Access. The tools named are few: Claude Code (3 postings), MCP (2, both as examples of Labs products), SQL (3), Python (2), Slack (2), and the three clouds.

How do Anthropic's PM postings compare with other AI companies?

To see what is specific to Anthropic, we ran the same literal text matches over the other 370 PM postings in our job-description harvest. This is a cruder method than the hand-coding above (it counts a word, not a meaning), so the Anthropic numbers differ slightly from the table. For example, "evaluations" also matches "technical evaluations" in a sales sense.

Anthropic's postings are about four times as likely to talk about evals and nearly five times as likely to talk about 0-to-1 work. Pay transparency also stands out: all 19 Anthropic postings state a salary range, while roughly half of the other postings do (between 46% and 52%, depending on how strictly you match a range). For the full market picture, see our state of AI PM hiring in 2026.

What phrases and values keep coming up?

Some sentences recur almost word for word across teams, which tells you they reflect a shared hiring bar rather than one manager's taste:

• "Think creatively about the risks and benefits of new technologies ... beyond past checklists and playbooks": 9 postings (all four Safeguards, both Research, both Labs, and Claude Science in a longer form).
• Flexing into new domains (for example "flex into different domains based on the business problem at hand"): 9 postings.
• "Navigate and execute amidst ambiguity" (or "amid") with "simple, easy-to-understand solutions": 5 postings.
• "MVP vs. ideal state": 5 postings (the four Safeguards roles and Claude Science's "MVP versus ideal state").
• "A creative, hacker spirit and love solving puzzles": 4 postings.
• "Intellectual curiosity without ego": 3 postings.
• "Write clearly enough that an engineer, an executive, and an external partner can each see the tradeoff you're making": 2 postings (Safe Access and New Markets).
• Designing for future models ("You design for where the models will be rather than where they are"): 2 postings (Cybersecurity and Business Technology).

The shared boilerplate adds the company's own framing. Anthropic calls itself "a public benefit corporation", says "We view AI research as an empirical science", and ends its description of how it works with "As such, we greatly value communication skills."

What is required and what is only preferred?

Across the 19 postings, the minimum qualifications are mostly about judgment and ownership: making technical tradeoffs, owning an outcome end to end, shipping through ambiguity, and communicating across research, policy and engineering. The preferred qualifications carry most of the credentials: years of experience, founder background, advanced degrees (Claude Science), hyperscaler employers such as AWS, GCP or Azure (the multi-cloud roles), and clearances (Public Sector).

A few roles make domain experience a hard minimum:

• Cybersecurity: "You've spent meaningful time in security, whether building security products, running a security team, or doing the work yourself".
• Claude Science: "a scientific or deeply technical background".
• Multi-Cloud Growth, Google: "Have built or shipped products with a hyperscaler as a partner".
• New Markets and Safe Access: having owned a developer platform, API, billing, identity or access product "end to end".
• Public Sector: being a US person, though it also says "We weigh product judgment above any domain credential".

What does this mean for your application and prep?

If you are aiming at one of these roles, here is what the data suggests.

1. Pick one family and read its posting line by line. The roles are more different than the shared title suggests. A Safeguards PM and a Claude Tag PM are measured on different things. Start from the Anthropic jobs and choose the one where you have the strongest evidence.
2. Build an eval story. Ten of 19 postings ask for eval work. Have one concrete example: what you measured, how you built the test set, and what launch decision it changed. Our evals chapter and the trust, safety and agent security chapter cover the PM side of this.
3. Build something yourself with an AI tool. Three postings say "prototype yourself" outright and ten ask for hands-on AI fluency. A small working prototype is stronger evidence than a claim. The PM as builder chapter shows how to scope one. Note Anthropic's candidate guidance: use Claude to refine your application, but "create your first draft yourself".
4. Prepare a safety tradeoff story. Fourteen postings make safety or misuse part of the job. Business Technology asks outright whether you have "slowed down or changed course for safety reasons before". Have a real example ready.
5. Show research partnership. A time you turned vague or contradictory expert feedback into a roadmap call maps to at least 14 postings. We have a practice question for exactly this: tell me about a time you turned vague or contradictory feedback in a highly technical product area into a clear roadmap decision.
6. Tailor your resume to the posting, not the company. Run your resume against the exact job description in our JD resume review, or use Resume Job Match to see which roles your resume fits best.
7. Practise on the posting's own text. Each role page has a mock interview built from that job description, and our Anthropic company page collects 105 Anthropic questions, 90 of them written from Anthropic's own PM postings.

This post covers what the postings say. For the interview itself (rounds, the culture interview, question types, and a 2 to 4 week plan), read our Anthropic PM interview guide. For practice questions across AI companies, see AI PM interview questions, and if you are still building toward roles like these, start with how to become an AI product manager.

Frequently asked questions

How many product manager roles is Anthropic hiring for?

As of September 2026 our catalog holds 19 Anthropic PM postings, read from Anthropic's Greenhouse board on 22 September. On 25 September the live board listed 17 PM roles out of 624 open jobs. Growth and Beneficial Deployments (Labs) had come down by then. The largest group is Safeguards, trust and access, with 6 roles.

What is the salary for a product manager at Anthropic?

Every Anthropic PM posting states an annual salary range. Ten list $305,000 to $385,000, eight list $385,000 to $460,000, and one (Multi-Cloud Growth, Google) lists $305,000 to $460,000. These are the salary ranges that California and New York pay transparency laws require in a posting, so they do not include equity or bonuses.

Do Anthropic PM roles require a technical background?

Most ask for technical depth, but few ask for coding. Only the two Research postings name Python, and only three name SQL. Claude Tag says outright that "An engineering background is not required". What the postings want is enough depth to make technical tradeoffs, work with researchers, and reason about model behavior and evals.

How many years of experience does Anthropic want for PM roles?

The lowest figure in any of the 19 postings is 5+ years, and the highest is 12+ (Cybersecurity). In 13 postings the years appear only as a preferred qualification. Two postings say they care about "the ownership, not the years", and Business Technology names no years at all.

Can you work remotely as a PM at Anthropic?

Mostly not. The postings say staff are expected in an office at least 25% of the time. Only Public Sector and Business Technology are listed as remote-friendly, and both say travel is required. All other roles are in San Francisco, New York or Seattle.

Does Anthropic sponsor visas for product managers?

The postings say Anthropic does sponsor visas, but "aren't able to successfully sponsor visas for every role and every candidate", and keeps an immigration lawyer to help. The Public Sector role is an exception, because it requires candidates to be a US person.

Can I use Claude when applying to Anthropic?

Anthropic's candidate guidance says to draft your application yourself and then use Claude to refine it. It says to use Claude to research the company and practise answers, but not during live interviews unless told otherwise. Our Anthropic interview guide covers this in more detail.

Sources

1. allthingspm.app jobs catalog, 19 Anthropic PM postings (catalog_jds, read from Anthropic's Greenhouse board on 22 September 2026, analysed 25 September 2026): jobs
2. Anthropic careers board on Greenhouse, live job list checked 25 September 2026 (624 jobs, 17 PM roles): job-boards.greenhouse.io/anthropic and boards-api.greenhouse.io/v1/boards/anthropic/jobs
3. Anthropic, Product Manager, Safeguards (Generalist): greenhouse
4. Anthropic, Product Manager, Safeguards (Account Integrity & Abuse): greenhouse
5. Anthropic, Product Manager, Safeguards (Cyber): greenhouse
6. Anthropic, Product Manager, Safeguards Rare Harms: greenhouse
7. Anthropic, Product Manager, Safe Access: greenhouse
8. Anthropic, Product Manager, Multi-Cloud Trust & Safety: greenhouse
9. Anthropic, Product Management, Research: greenhouse
10. Anthropic, Product Manager, Research (Code): greenhouse
11. Anthropic, Research Product Manager, Labs: greenhouse
12. Anthropic, Product Manager, Beneficial Deployments (Labs), no longer on the board as of 25 September 2026: greenhouse
13. Anthropic, Product Management, Human Data Platform: greenhouse
14. Anthropic, Product Manager, Claude Science: greenhouse
15. Anthropic, Product Manager, Claude Tag: greenhouse
16. Anthropic, Product Manager, Cybersecurity: greenhouse
17. Anthropic, Product Manager, Growth, no longer on the board as of 25 September 2026: greenhouse
18. Anthropic, Product Manager, Multi-Cloud Growth, Google: greenhouse
19. Anthropic, Product Manager, New Markets and Monetization: greenhouse
20. Anthropic, Product Manager, Public Sector: greenhouse
21. Anthropic, Product Manager, Business Technology: greenhouse
22. allthingspm.app JD corpus, 389 PM postings from 86 AI companies (370 non-Anthropic postings from 85 companies used for comparison), read 22 September 2026
23. Anthropic, "Introducing Labs", 13 January 2026: anthropic.com
24. Anthropic, guidance on candidates' AI usage: anthropic.com
25. Infosecurity Magazine, "Anthropic Rolls Out Claude Security for AI Vulnerability Scanning", 1 May 2026: infosecurity-magazine.com
26. California Employment Law Report, "What California Law Requires in Your Job Postings", April 2026 (Labor Code 432.3, SB 1162, SB 642): californiaemploymentlawreport.com
27. New York State Department of Labor, Pay Transparency (Labor Law 194-B): dol.ny.gov
28. Levels.fyi, Anthropic Product Manager salaries, checked 25 September 2026: levels.fyi
29. allthingspm.app question bank, Anthropic company page (105 questions, 90 written from Anthropic PM postings): question-bank/companies/anthropic