Resources

An FAQ on AI training work for experts

Patronus AI teamAs of November 202611 min read

Patronus AI publishes this FAQ and hires domain experts on contract. Every claim links its source, read on Nov 19, 2026.

AI training work pays professionals to write tasks, reference answers and grading rubrics, review model answers and sometimes build whole training environments in their own field. AI labs hire for it directly and through data companies and research labs. Postings read on Nov 19, 2026 listed hourly pay from $35 to $150 or more, depending on field and experience, mostly for remote contract or temporary work. The FTC warns that task scams use similar language, and it says an employer that asks you to pay to get a job is a scammer.

Snapshot as of November 2026. Last updated

What is AI training work for experts?

AI training work pays professionals to turn what they know into material a model learns from or is tested on. AI labs train models with reinforcement learning on tasks that come with a way to grade the answer, and in professional fields many of those tasks come from people who do that work. SemiAnalysis (Jan 6, 2026) writes that while software engineers build the environment, “often the workflows are created, described, and graded by domain specific contractors involved in the environment creation process”, with finance professionals defining financial tasks, doctors medical ones and lawyers legal ones.

Labs hire experts directly and through data companies. The xAI job board, whose postings name the employer SpaceXSI, lists temporary AI Tutor roles in fields from medicine to math to languages. Scale AI runs Outlier, which it describes as a platform that “connects experts with leading AI companies to provide human feedback that improves large language models (LLMs).” Research labs that build training environments hire experts too. Patronus AI is one, and on Nov 19, 2026 its Greenhouse job board listed contract roles for experts in finance, healthcare, law, sales, mechanical engineering and interactive coding. Epoch AI’s FAQ on RL environments (Jan 12, 2026) quotes one environment founder. “Finding the experts isn’t that hard, but managing them and doing quality control is hard.”

Sources [1] [7] [9] [2] [19]

What do domain experts actually do in AI training jobs?

Experts write tasks, reference answers and the rubrics or checks that grade an answer, and they review model answers or other experts’ work. SemiAnalysis writes, “Contractors design tasks, write expected solutions, specify reward signals, and in some cases, grade model solutions.” It adds that most grading rubrics are still written by humans.

Postings read on Nov 19, 2026 describe the same jobs. The AI Tutor, Medicine - Temporary posting asks clinicians to “Create and solve complex medical problems within your area of expertise”, write detailed reference answers and critically evaluate model responses. The AI Tutor, Legal & Compliance - Temporary posting covers legal research, analysis, writing and applying the law to facts. Scale AI’s Subject Matter Expert role is limited to quality work, including audits, scoring rubrics and gold standards. Outlier’s home page lists “Write a Challenging Prompt” among its common tasks, which means thinking of a difficult question in your field that would confuse an AI model and then writing the correct answer.

Patronus AI’s finance posting asks experts to show researchers what their work looks like day to day and to build tasks from a scenario, including the working files, the request, their own answer and how it should be graded. Its site reliability engineering posting asks for tasks that take an experienced engineer about a working day, with the sandboxed setup each task needs, the expert’s own solution and “checks a tester could verify without opinion.” Mechanize, which builds coding tasks with its own engineers, says quality assurance on the grader takes the most time.

Sources [1] [3] [4] [8] [9] [12] [18] [11]

Which fields hire experts for AI training?

Postings span many fields. On Nov 19, 2026, DataAnnotation listed 133 roles in categories including coding, cybersecurity, data and ML, STEM, engineering, finance, economics, accounting, legal, medical, healthcare operations, design and languages, with languages the largest group. The xAI board listed AI Tutor roles in applied math, biology, chemistry, design, humanities, legal and compliance, materials science, math, medicine, physics and writing, plus a range of languages.

Patronus AI’s open expert roles on the same date are in finance, healthcare, law, sales, mechanical engineering, and interactive coding for payments processing and site reliability engineering. Several list specific jobs. The healthcare posting lists roles such as medical coding and billing specialists, prior authorization specialists and clinicians, and the sales posting lists roles such as revenue operations, commissions and deal desk analysts.

Sources [10] [7] [12] [13] [14] [15] [16] [17] [18]

How much does AI training work pay experts?

Pay is usually hourly and varies by field and experience. These rates are as each posting stated them on Nov 19, 2026. DataAnnotation’s home page advertises “expert rates of up to $75–$125+/hr”, and its role list shows up to $50 an hour for generalists, up to $125 or more an hour for roles such as attorneys and physicians, and up to $150 or more an hour for software engineers. Outlier’s home page promises “Competitive pay + quality-based rewards, paid weekly” without a figure.

xAI board postings list ranges for US-based candidates. The Spanish tutor role lists $35 to $45 an hour, the Math role $45 to $75, the Legal & Compliance role $50 to $100 and the Medicine role $45 to $100, each depending on experience and other factors. For international candidates, the postings say pay information comes during recruitment.

Patronus AI’s expert postings list $50 to $150 per hour, depending on experience, for finance, healthcare, legal, sales and mechanical engineering, and $150 per hour for the two interactive coding roles, each for 10 to 30 hours per week, as of Nov 19, 2026.

Sources [10] [9] [6] [5] [4] [3] [12] [13] [14] [15] [16] [17] [18]

Do you need AI or coding experience to do AI training work?

Usually not, but you need deep experience in your own field. Outlier says “No AI experience needed.” The xAI Legal & Compliance posting requires a J.D., active bar admission in good standing and 2 or more years of practice, and lists familiarity with AI tools or data annotation only as a preference. The Medicine posting requires one of several named clinical degrees (MD, DO, MBBS, PharmD and others) or one of several named master’s or PhD degrees, plus 2 or more years in a clinical, research or health-related role after the degree.

Patronus AI’s postings ask for 3 or more years doing the job yourself, “Still in the role, or left recently”, and call prior AI data labeling, annotation or RL task creation experience a bonus. The exception is its two interactive coding roles, which require that experience. Epoch AI’s interviewees made the same point about task design. One founder said, “Domain knowledge and expert level prompting is more important than ML skills for creating tasks.”

Sources [9] [4] [3] [12] [17] [18] [2]

Is AI training work a full-time job or a contract?

Most of the postings read for this FAQ are contract or temporary work. Scale AI’s Subject Matter Expert role is part-time and hourly, and says you can work as many or as few hours as you like. DataAnnotation describes its work as a flexible, task-based contractor role with no long-term commitment. Patronus AI’s expert roles are independent contractor roles, remote, on your own schedule, at 10 to 30 hours per week, with rolling start dates.

Some postings are closer to employment. xAI’s AI Tutor postings are temporary positions with a stated length of 6 months, with possible extensions or conversion to a permanent role. They may be full-time or part-time, with full-time preferred, and list benefits for eligible US-based positions including health insurance, a 401(k) plan and paid sick leave. SemiAnalysis reports that these expert contracts, often arranged through human data contractor firms, typically last at least a quarter and can be part time or full time.

Sources [12] [8] [10] [4] [3] [1]

How do you get hired for AI training work?

Outlier’s home page lists a step to “Verify skills and start earning.” The xAI Medicine posting asks for English proficiency at CEFR C1 level or higher, and all four xAI tutor postings read for this FAQ say candidates must be willing to travel to a SpaceXSI site. Scale AI’s Subject Matter Expert posting says it waits 90 days before reconsidering a candidate for the same role. Patronus AI’s postings list a short AI interview and resume review, one paid trial task “scored against a fixed quality checklist”, and then onboarding and a first project. Applicants to its finance, healthcare, legal and sales roles rank the specific jobs they are interested in.

Who can apply depends on location. On Nov 19, 2026, the xAI board’s Legal & Compliance and Spanish tutor postings were open to remote candidates in the United States, United Kingdom, Canada and Australia, and the Math and Medicine postings added India. Those postings also said they could not hire US-based candidates in Wyoming or Illinois at that time.

Patronus AI’s healthcare and legal expert postings say “Remote, US only” in the posting text. Its finance, sales, mechanical engineering and interactive coding postings say “Remote” there, while the Greenhouse location field for each of these seven expert postings reads “Remote - US”. DataAnnotation and the xAI board both list many language roles.

Sources [12] [8] [9] [3] [4] [6] [5] [13] [14] [15] [16] [17] [18] [10] [7] [19]

Can I use my employer’s or clients’ documents in AI training tasks?

Postings set rules for both AI use and source material. Epoch AI quotes a neolab researcher on the challenge of managing contractors. “But how do you make sure they’re not just using LLMs? How do you make sure they’re actually verified?” Patronus AI’s finance posting says “Nothing copied from published sources, and no real client, employer or personal data. Every scenario is fictional.” Its healthcare posting excludes real patient data “even de-identified”. Its legal posting excludes anything privileged or confidential, its mechanical engineering posting excludes anything under NDA or export control, and its coding postings exclude code, configs or data from a current or former employer or client.

The same postings allow AI tools as an aid but say you are responsible and evaluated for your own work. For your own protection, read the confidentiality terms of your current job before you sign any contract that draws on it.

Sources [12] [13] [14] [16] [17] [2]

What makes a good AI training task?

A good task is hard for current models and can be graded the same way every time. Mechanize, which builds coding tasks for AI labs, says it is often more efficient to identify specific capability issues in existing models and build the task to target those, and that “The ideal task has multiple different sources of difficulty, not just one.” It also says the most common failure mode for new hires is underestimating what coding agents can already do, and that without extensive experience with these agents, first task ideas will likely be too easy.

Grading is the slow part. Mechanize writes graders as a mix of procedural tests and a rubric for an LLM grader, and says it is very rare for a first grader to be good enough, because graders can be nondeterministic, unfair, incomplete or wrong. It asks for a breakdown of why frontier models fail each task, so that failures reflect a gap in the model rather than a flaw in the task. Patronus AI’s interactive coding postings ask for checks “a tester could verify without opinion”.

Graders also have to resist shortcuts. Epoch AI’s interviewees named robustness against reward hacking, where a model raises its score without doing the task, as the top quality criterion, and one founder said, “It takes many many iterations to check against reward hacking.” For the vocabulary behind this, see our FAQ on RL environments and the verifier glossary entry.

Sources [11] [18] [2]

How can you tell a real AI training job from a scam?

The FTC warns about a scam that uses the same vocabulary as AI training work. Its December 2024 Data Spotlight on task scams says about 20,000 people reported them in the first half of 2024, compared with about 5,000 in all of 2023, and that reported losses to job scams topped $220 million in the first half of 2024. These scams usually start with an unexpected text or WhatsApp message offering vague online work, use the word “tasks” and buzzwords like “product boosting” or “app optimization”, pay small amounts at first, and then ask for a deposit to unlock supposed earnings. The FTC says crypto is the currency of choice for these scams.

The FTC’s advice is to ignore generic and unexpected job texts or WhatsApp messages, and never pay anyone to get paid. Its Job Scams guide adds, “Honest employers, including the federal government, will never ask you to pay to get a job. Anyone who does is a scammer.” It also says to search the company or recruiter’s name with words like “scam”, “review” or “complaint”, and never to deposit a check from an employer and send part of the money back.

A legitimate posting can be checked at its source. Look for the role on the company’s own careers page or applicant tracking system, apply there rather than through a message, and expect the company to pay you, not the reverse. Patronus AI lists its open roles at /join, and each role page links to its Greenhouse application. Report scams to the FTC at ReportFraud.ftc.gov.

Sources [20] [21] [12]

About this FAQ

Patronus AI publishes this FAQ. It hires domain experts on contract, so its postings appear as one example among the employers described. The format follows Epoch AI’s FAQ on RL environments.

Pay and open roles change often. Every posting cited here was read on Nov 19, 2026, and the Sources list gives each one. To report an error, write to us through the contact page.

Sources

  1. 1 AJ and Dylan Patel, “RL Environments and RL for Science: Data Foundries and Multi-Agent Architectures”, SemiAnalysis, Jan 6, 2026 (free portion read)
  2. 2 Jean-Stanislas Denain and Chris Barber, “An FAQ on reinforcement learning environments”, Epoch AI Gradient Updates, Jan 12, 2026
  3. 3 SpaceXSI (xAI job board), “AI Tutor, Medicine - Temporary”, job posting, read Nov 19, 2026
  4. 4 SpaceXSI (xAI job board), “AI Tutor, Legal & Compliance - Temporary”, job posting, read Nov 19, 2026
  5. 5 SpaceXSI (xAI job board), “AI Tutor, Math - Temporary”, job posting, read Nov 19, 2026
  6. 6 SpaceXSI (xAI job board), “AI Tutor, Spanish - Temporary”, job posting, read Nov 19, 2026
  7. 7 xAI job board (postings name the employer SpaceXSI), AI Tutor listings, read Nov 19, 2026
  8. 8 Scale AI, “Subject Matter Expert”, job posting, first published Jun 9, 2026, read Nov 19, 2026
  9. 9 Outlier, home page, read Nov 19, 2026
  10. 10 DataAnnotation, home page and role list, read Nov 19, 2026
  11. 11 Mechanize, “What working at Mechanize is like”, read Nov 19, 2026
  12. 12 Patronus AI, “Expert: Finance (Contract, Remote)”, job posting, read Nov 19, 2026
  13. 13 Patronus AI, “Expert: Healthcare (Contract, Remote)”, job posting, read Nov 19, 2026
  14. 14 Patronus AI, “Expert: Legal (Contract, Remote)”, job posting, read Nov 19, 2026
  15. 15 Patronus AI, “Expert: Sales (Contract, Remote)”, job posting, read Nov 19, 2026
  16. 16 Patronus AI, “Expert: Mechanical Engineering (Contract, Remote)”, job posting, read Nov 19, 2026
  17. 17 Patronus AI, “Expert: Interactive Coding, Payments Processing (Contract, Remote)”, job posting, read Nov 19, 2026
  18. 18 Patronus AI, “Expert: Interactive Coding, Site Reliability Engineering (Contract, Remote)”, job posting, read Nov 19, 2026
  19. 19 Patronus AI open roles, Greenhouse job board API, read Nov 19, 2026
  20. 20 Federal Trade Commission, “Paying to get paid: gamified job scams drive record losses”, Data Spotlight, Dec 2024
  21. 21 Federal Trade Commission, “Job Scams”, Consumer Advice, March 2023

Patronus AI · November 2026 snapshot · More resources