Anthropic’s Staff+ Application Security Engineer post says “We use Claude as our primary tool across every part of the job,” and that “the human work is the judgment layer.” That tells you more about the Anthropic security engineer interview than the forums do. Glassdoor lists about three Senior Security Engineer interviews, and our searches found no Reddit thread (Reddit served a login wall, so one may exist).
Everything below is about the security roles. The general engineering loop is in the Anthropic interview guide and in how the Anthropic software engineer interview works, and other company loops sit in the security engineer interview guides.
The Anthropic security engineer interview process, stage by stage
No source lays out the full security loop, and order varies; the general loop in our SWE post puts the hiring-manager call before the final rounds. What each stage has behind it:
- Recruiter screen. No security report describes it. A Safeguards software engineer on Aced was asked “Why do you want to work at Anthropic?” (one report, adjacent role).
- Online assessment. The careers page names CodeSignal for technical roles, and the Aced candidate took a CodeSignal-style pre-assessment they recall as “pretty hard” (not confirmed for security).
- Technical screen. A London Senior Security Engineer phone screen on 1point3acres was a multithreaded Python web crawler (one post-interview report). Two Blind posts expected coding or code review (pre-interview expectations).
- Coding rounds (screen or onsite not stated). Glassdoor snippets mention an async web crawler and a code-review task (snippets only).
- Design. One Glassdoor interviewer asked the candidate not to go deep on security (single report). The Aced candidate designed an LLM sampling and batching API (adjacent role).
- Values. Detailed only for the Aced role: seven questions from non-technical interviewers, and a recruiter who said this is where people usually fail (one report, adjacent role).
- Hiring manager and past projects. A Glassdoor snippet describes a manager pushing on past projects’ metrics and technical decisions; the Aced candidate got a conflict question and “biggest project” (snippets, adjacent role).
- References. Anthropic wanted the Aced candidate’s references “from day one” after hearing of competing offers (one report, adjacent role).
What Anthropic says about its own interviews
The “How we hire” section of Anthropic’s careers page says all interviews run on Google Meet, scheduled across time zones. Technical roles use live coding tools such as Colab and CodeSignal, and “You can look things up,” if you know basic syntax and the standard library. Anthropic gives no feedback on resumes or interviews, and you can re-apply after 12 months, or sooner if your experience has materially changed. Offers carry no fixed deadline, and visas and green cards are sponsored for eligible roles. The job posts expect staff in an office at least 25% of the time.
The candidate AI guidance bans AI assistance in live interviews and take-homes unless Anthropic explicitly allows it; Claude is fine for preparing or polishing your own draft. The page dates from July 2025 but is still the one Anthropic links, and security applications ask you to confirm it. See AI-era interviewing for other employers’ rules. Anthropic’s only documented take-home, for performance engineering, allows AI, per its post on AI-resistant technical evaluations. We found none for security.
The application form is the first screen
The AppSec, D&R and Offensive Security forms all require “Why do you want to work at Anthropic?” and note that great answers are often 200 to 400 words. AppSec also asks for a paragraph on “something meaningful that you have done in line with your values.”
The Security Engineer, Detection & Response form adds required screening questions whose answer options climb a ladder:
- Your Kubernetes security experience, from “never used” up to having “configured security controls or investigated security incidents.”
- How you would investigate a suspicious DNS request, from a basic search up to having “developed custom tooling.”
- “When was the last time you wrote code professionally or for a personal project?”
- Whether you regularly read and understand code written by other engineers.
Pick the rung you can defend, because claiming custom DNS tooling invites the question of how you built it.
What Anthropic’s application security engineer post asks for
The Staff+ Application Security Engineer post wants “people who can find the vulnerability but would rather build the system that finds them all,” with production-quality code in Python, Go, Rust or TypeScript. Builders beat reviewers here. Both it and the M&A variant are Staff+: the minimum “will correlate with the internal job level,” and 7+ years is preferred.
The team hires engineers who “ship production systems and clear a hands-on threat-modeling bar.” No published loop includes a threat-modeling round, but prepare as if that line implies one. The post names the new systems: multi-agent orchestration, sandboxed code execution, agents holding delegated credentials, and “untrusted tool output crossing trust boundaries.” Practice with the worked threat modeling interview and the agent questions in AI security interview questions.
The preferred qualifications ask for engineers who already use LLMs heavily, “with opinions about where they help and where they don’t,” and the post wants someone “comfortable being handed a problem and a lot of latitude rather than a spec.” Have two stories ready: a bug an LLM found for you, and a time its output was noise. The team shaped Claude Code’s sandboxing and auto mode; read Anthropic’s posts on sandboxing, the auto mode threat model and how it contains Claude.
Detection and response, software security and offensive roles
The D&R post asks for “3+ years of software engineering… and/or 5+ years of detection engineering, incident response, or threat hunting,” plus “A high level of proficiency in Python and query languages such as SQL.” It includes on-call. The Senior and Staff+ Software Security Engineer posts (5+ and 8+ years) center on identity and secrets, Kubernetes RBAC, supply chain, and cloud IAM, and list “passion for AI safety” as a qualification. For them, a security system design interview is closer to the job than any puzzle. The Offensive Security post asks for 5+ years and “deep expertise in at least two of” macOS, Linux, cloud, Kubernetes and CI/CD security.
An engineer with about four years meets the D&R post’s 3+ years of software engineering minimum and sits a year short of Senior Software Security’s 5+; the Staff+ AppSec posts are a reach. The Frontier Safety Roadmap explains the demand: “we will need substantially more security capacity staffing than we currently have.”
Posted salary ranges for Anthropic security roles
| Anthropic security role (Greenhouse job post) | Posted annual base salary, US unless noted, Nov 2026 (equity not stated) |
|---|---|
| Staff+ Application Security Engineer | $320,000–$485,000 |
| Staff+ Application Security Engineer, M&A | $320,000–$485,000 |
| Security Engineer, Detection & Response | $300,000–$405,000 |
| Security Software Engineer, D&R Platform | $320,000–$405,000 |
| Incident Manager, D&R | $290,000–$365,000 |
| Security Engineer, Threat Intel | $320,000–$405,000 |
| Senior Software Security Engineer | $320,000–$405,000 |
| Staff+ Software Security Engineer | $405,000–$485,000 |
| Staff+ Software Security Engineer, Secure Frameworks or Labs | $320,000–$485,000 |
| Security Engineer, Offensive Security | $300,000–$320,000 |
| Security Engineer, Corporate Security | $320,000–$405,000 |
| Staff Security Engineer, Risk Engineering | $320,000–$405,000 |
| Staff Software Security Engineer, London | £255,000–£325,000 |
| Software Engineering Manager, Network Security | $405,000–$485,000 |
| Engineering Manager, Hardware Platform Security | $485,000–$625,000 |
Levels.fyi had no security-specific Anthropic data we could verify, and one prep site’s “$125,000 average base” contradicts every posted range. Run an offer through the total comp calculator once equity is named, then use the salary negotiation guide.
Interview questions Anthropic security candidates report
The two Blind posts that describe the screen format were written before the interviews, and neither poster reported back. One said the “1st round seems to be a coding / code review kind of interview”; the other said a “security family” screen needed Python and SQL. That’s what they were told, likely by recruiters. The post-interview record is Glassdoor snippets (the pages blocked our tools) and one login-walled 1point3acres summary.
Those reports agree that security candidates write real code and that the problems are practical. One Glassdoor reviewer was “tested on coding skills and design skills” despite applying for security. A Glassdoor overview snippet describes a Security Engineer code-review task: find the bugs, suggest improvements. The secure code review interview guide works that format through.
The crawler is the most specific signal, in two separate Senior Security Engineer reports: async on Glassdoor, with advice to “start small with a synchronous solution,” and multithreaded Python in the London screen on 1point3acres. Given the design report that steered away from security, keep plain distributed-systems answers ready from the system design interview guides.
How to prep, in priority order
Write the application answers first
Anchor “Why Anthropic?” in named Anthropic security work, such as the sandboxing post or a roadmap goal, and say what you’d build toward it.
Practical concurrent Python, with no assistant open
Build a crawler that respects a concurrency limit and skips URLs it has already seen. Start synchronous, as the Glassdoor advice suggests, then move to threads and asyncio. Practice with docs open and no AI, as in the live rounds. D&R candidates should add SQL interview questions.
Code review and threat modeling reps
Read unfamiliar code often and write findings ranked by severity; the D&R form asks whether you do this. Then threat model one agent system end to end: an agent holding delegated credentials that reads untrusted tool output. Why the security engineer loop rewards judgment over trivia explains what interviewers listen for.
Values stories and one project you can defend
The detailed values questions come from the adjacent Safeguards report, so treat them as types, not a script: a time you built something against your values, a conflict, tough feedback given and received, being wrong, and your view of Anthropic’s mission, each built with the behavioral interview and STAR method. Also pick one project you can defend on its metrics and technical decisions, where the manager round reportedly pushes.
What we couldn’t confirm
- The number of rounds. Prep sites say three to five and cite nothing; no source describes a fixed security panel.
- A take-home or online assessment for security. The one Blind reply claiming a security “multi-stage OA” reads as promotion for a question-bank site.
- Whether the AppSec, D&R and infrastructure loops differ. No candidate report compares them.
- A Blind reply saying ChatGPT is allowed. It contradicts Anthropic’s written guidance; follow the guidance.
- Whether AppSec coding is LeetCode-style. On Blind, the only answer was “Leetcode,” the poster’s follow-up, “Have you done the app sec interview? What level was the code?”, went unanswered, and every problem described elsewhere is practical. A few standard data-structure problems are a cheap hedge.
- Timelines. Glassdoor snippets say 30 days on average, about four weeks, and 14 days; the Aced candidate’s process took two weeks.
- A file-path traversal task in Glassdoor snippets, which disagree on whether it exists.
Anthropic gives no interview feedback, so if you go through this loop, write down what each round covered the same day. That record will beat anything on this list.
Practice the behavioral round:
