Abnormal was founded in 2018 by CEO Evan Reiser, CTO Sanjay Jeyakumar and head of machine learning Jeshua Bratman. The company renamed itself Abnormal AI on April 16, 2025; that release describes deployment by API into Microsoft 365 or Google Workspace. Its August 2024 Series D raised $250M at a $5.1B valuation, led by Wellington Management, when it said it had passed $200M in ARR; this guide found no later priced round. In July 2026 it reported 4,500 customers, including more than 25% of the Fortune 500. In March 2026 it shipped Attune 1.0, a behavioral foundation model it says produces approximately 85% of malicious detections.
The careers page counts 1,300+ people and calls the company “remote-first and async by default.” A Bangalore detection posting on the job board asks for Python or Go, Spark, Airflow, AWS and Postgres. San Francisco roles are hybrid (a tech lead manager posting cites “25% in-office presence”), Bangalore roles are hybrid, and the remote openings are application security and threat research in the US, two software engineering roles in the UK and an ML role in Singapore.
Process
Abnormal does not publish its engineering loop. Postings say only that recruiters use AI tools to “suggest interview questions” and that “every decision about a candidacy is made by a person” (example). The rest comes from candidate reports.
A July 2023 Blind thread describes “coding + Past Project Interviews,” with one reply calling the India bar “LC medium.” A November 2023 Bangalore offer post lists “1 OA + 1 Pair Programming + 2 Technical (Managerial) + 1 HR.” A January 2025 report describes a security-focused take-home with a one-week deadline that took about two days, then a rejection without feedback.
In May 2025 LeadDev reported that Abnormal had moved AI coding tools from take-homes into live coding interviews, where candidates are asked to go through the edge cases; the task barely changed, but everyone is now expected to reach what used to be the stretch question. Shrivu Shankar, a staff ML engineer who leads its GenAI team, contrasted the “naive candidate” who relays the AI’s answer with one who questions it. The 2026 reports fit: a September 2026 Software Engineer II report and an October 2026 senior report each describe six rounds. Two PracHub write-ups of July 2026 New York interviews, posted the same day and possibly from one candidate, say AI agents were allowed throughout (one) and that the hiring manager asked how the candidate uses AI day to day (the other). A July 2026 Bengaluru report describes the same AI-assisted codebase screen, so the format is not US-only; the LeetCode reports give no location.
| Stage (2026 reports) | What candidates describe | Reported in |
|---|---|---|
| Recruiter and hiring manager screen | Past projects, impact, why you are moving, how you use AI at work | LeetCode Sep 2026, PracHub Jul 2026 |
| Codebase round | Read an unfamiliar repo (AI allowed in most reports), add a feature, write tests | LeetCode Jun, Sep, Oct 2026; PracHub Jul 2026 |
| Incident response plus system design | Investigate a production incident, in AWS in one report, then scale an existing system | LeetCode Jul, Sep, Oct 2026 |
| Code review | Rank issues in a small repo, propose fixes, extend it under load | LeetCode Sep, Oct 2026 |
| Past projects or manager round | Technical decisions, challenges, impact | LeetCode Sep, Oct 2026 |
| Extra round (sometimes) | More technical signal, or a from-scratch system design | LeetCode Sep, Oct 2026 |
The September 2026 candidate cleared the scheduled rounds, took an added one “because the team wanted some additional signal,” did not clear it, and was put on hold with feedback citing a lack of “judgment in a few areas.” Judgment is also what security engineer loops weigh most. Extern’s August 2026 guide, citing Glassdoor, puts the process at about 31 days with a difficulty of 2.8 out of 5.
What they actually ask
- A 60-minute phone screen problem, per its 1point3acres listing (four related discussions, no dates): find duplicate images in a directory using
os.walkand a provided_calculate_hashhelper. Follow-ups cover hash collisions, limited memory, transformed near-duplicates, and scaling with Redis and map-reduce. - The AI-allowed codebase screen in a June 2026 report: a repo that ingests security events, assigns threat levels with rules, raises alerts and serves APIs. The task was to make a hard-coded enrichment layer (geo-IP, history) pluggable so customers configure it without touching platform code. Claude was allowed.
- From the September 2026 report: investigate an incident in an AWS environment, name the logs and signals you would check, and separate immediate mitigation from the long-term fix; then scale the reads and writes of an existing system. A July 2026 SDE II report describes the same pairing.
- Also September 2026: review a small repository, rank issues as P0 or P1, justify the ranking, then extend the system. The discussion went into concurrency, worker and queue scaling, and failure scenarios, and the candidate advised senior-level candidates to go beyond “add more workers” or “use a queue.”
- A July 2025 site reliability loop: an hour on DNS, TCP versus UDP, TLS, and debugging one of two identical database servers that runs slower, then a Python CodeSignal problem computing average session time per user from a CSV of open and close events.
- A 2024 ML engineer phone screen on 1point3acres: CSV parsing and an NLP sentiment task.
Security roles at Abnormal AI
No public candidate report covers an Abnormal application security, threat research or detection engineering loop; what follows comes from postings.
- Application Security Engineer II (remote US): secure “LLM-integrated features, agentic workflows, MCP connectors, and the model supply chain” against threats like prompt injection, and run threat models and architecture reviews. It wants production code in Python, Go, Java or TypeScript and reports to the Director of Security Engineering, whose posting describes a commercial cloud plus “a separate FedRAMP-authorized environment.”
- Threat Researcher (remote US): investigates email and identity campaigns, writes briefs and presents to customers. It asks for MITRE ATT&CK, BEC, credential harvesting and account takeover knowledge, plus basic SQL.
- The ADEX team (Software Engineer II, Bangalore) owns impersonation detection, studies attacks that got through, tunes detection logic built on scored signals, and builds LLM detection agents measured on precision and recall. A Machine Learning Engineer 1 role on misdirected email (Singapore) covers thresholds, A/B tests, drift and on-call for false positives.
Detection engineers are hired as software engineers (the ADEX opening is a Software Engineer II posting), so expect the codebase, incident and code review rounds above; no public report describes a 2026 machine learning engineer loop. For application security, read the posting’s must-haves as the syllabus: threat-model an LLM feature with tool access, explain where prompt injection gets in, and review code for OWASP Top 10 issues. For threat research, prepare to defend an assessment built on incomplete evidence to a specific audience, the measure the posting names. For detection and ML roles, be ready to set a threshold for a detector given the cost of a false positive, read precision and recall on imbalanced data, and say how you would notice drift, the work the ADEX and misdirected-email postings describe. These are inferred preparation targets, not reported questions.
Levels and comp (2026)
Levels.fyi shows a median US software engineer package of $261K from 90 submissions (updated November 5, 2026): $195K base, $47K stock as annualized grant value, $19K bonus. Level medians are $178.9K at P1, $203.2K at P2, $228.4K at P3, $301.6K at P4 and $351.4K at P5; this guide found no public mapping from posted titles to P-levels, so ask the recruiter which level the role is. The machine learning engineer page shows $190K to $296K+.
Posted US base ranges as of November 2026, each noting possible bonus and equity: Software Engineer II, Insider Risk (San Francisco) $149,200 to $214,500; Tech Lead Manager, Identity Security $187,400 to $269,500; Application Security Engineer II $130,100 to $187,000; Threat Researcher $122,400 to $176,000; Director of Security Engineering $214,200 to $308,000.
For Bangalore, LeetCode posts report an April 2025 SDE II offer at 48L base, 12% variable and $30K of RSUs over four years, and a September 2025 Software Engineer 1 offer at 32L base, a 0 to 20% performance bonus and $11,500 of RSUs over four years. The stock is private, so ask for the latest 409A price, the vesting schedule and the post-termination exercise window.
Prep priorities
Practice the format before the algorithms. Take an unfamiliar open-source repository in Python or Go, give yourself ten minutes to map it with an AI assistant, then add a feature with tests while explaining each decision. Rehearse an incident: from logs and metrics, state the mitigation separately from the root-cause fix. For design, go past “add a queue” and explain what happens to ordering, retries and throughput when workers saturate. Entry-level candidates should still drill medium algorithm problems, since a June 2026 SDE-1 candidate was scheduled for a DSA round.
Spend an evening on the domain: business email compromise, vendor fraud, account takeover, lookalike domains, and what an API-connected product sees that a mail gateway cannot. Security candidates should work through the security engineer interview guides here and compare the CrowdStrike and Okta guides.
Frequently Asked Questions
Is Abnormal Security the same company as Abnormal AI?
Yes. It became Abnormal AI on April 16, 2025, and its legal name is now Abnormal AI, Inc. Older Blind and LeetCode posts use Abnormal Security, so search both names.
Can you use AI tools in the Abnormal AI coding interview?
In the codebase rounds, reports say yes. A June 2026 candidate used Claude, and a July 2026 SDE II round “explicitly allowed the use of AI tools.” Rules per round are not published, so confirm with the recruiter in writing.
Is the Abnormal AI interview LeetCode-style?
Less than it was. Reports from 2023 to 2025 describe medium-level coding, and 1point3acres lists a duplicate-image hashing problem; 2026 reports describe codebase work, incident investigation, code review and system design.
Does Abnormal AI hire remote engineers?
Some. In November 2026, US product engineering postings are hybrid in San Francisco, while the application security and threat research roles are remote in the US, two software engineering roles are remote in the UK and one ML role is remote in Singapore. An April 2025 Bangalore offer specified two office days a week.
How much does Abnormal AI pay?
Levels.fyi’s US software engineer median is $261K in total compensation; posted US base ranges are $130,100 to $187,000 for Application Security Engineer II and $122,400 to $176,000 for Threat Researcher, with bonus and equity possible on top.
Similar company guides
Prepping for Abnormal AI? Put it to work:
