Yes—some highly educated professionals are already being paid to improve AI systems that could automate parts of their former work. Lawyers, scientists, engineers, writers, consultants and other specialists are creating model answers, grading AI responses and translating professional judgment into data. That does not prove mass white-collar unemployment, or that entire occupations are about to disappear. It does show a more immediate shift: expert knowledge is becoming a contract input for AI companies, often through unstable project work.
For workers, the financial contradiction is sharp. An AI-training assignment can offer a high hourly rate while providing no guaranteed hours, benefits or long-term security. The same work that pays the bills today may also help reduce the value of some tasks—and eventually some career paths—tomorrow.
As an Amazon Associate I earn from qualifying purchases.
The job is not usually building an AI model
Most professionals recruited for this work are not training neural networks or writing machine-learning infrastructure. They are producing and evaluating the human data used to make AI systems more capable and reliable.
Depending on the project, an expert might:
- Write realistic prompts based on professional tasks.
- Produce an ideal or “golden” answer.
- Create a rubric or checklist for judging quality.
- Compare several AI-generated answers.
- Identify factual, logical, stylistic, ethical or procedural errors.
- Test whether a model follows complex instructions.
- Design difficult examples intended to expose weaknesses.
- Review contracts, code, financial analyses, medical reasoning or scientific explanations.
- Annotate audio, video, images or conversations.
- Record speech or act out realistic interactions.
A typical pipeline can divide the work among several people: one creates the task, another drafts or approves the ideal answer, another grades model outputs and another checks whether reviewers applied the standards consistently. Mercor describes its expert contributors as people who review model outputs, create examples of expert reasoning and assess quality against defined standards. Mercor’s expert-network description is the company’s account of how that work is organized.
#1 Best Overall
What a lawyer or scientist may actually contribute
Legal work
A legal expert may be asked to interpret contract language, assess risk allocation, review a transaction, analyze an employment-law scenario or decide whether an answer is merely plausible or legally defensible.
For example, a project might ask an AI system to analyze an employment contract. The lawyer could then:
- Mark errors, omissions and unsupported conclusions.
- Write a model analysis that addresses the important legal and commercial risks.
- Turn professional standards into a scoring rubric.
- Grade several alternative answers.
- Explain why a response received a high or low score.
- Identify recurring mistakes for engineers and researchers to address.
Mercor currently markets legal-expert work involving contract interpretation, risk allocation, transaction analysis and high-stakes legal or regulatory reasoning. Its lawyer and legal-expert page should be read as a description of advertised opportunities, not proof that every lawyer receives those assignments.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Fix the driver behind crashes, sound loss and screen glitches3Clear out junk files and repair common Windows errorsScientific work
Scientists may test whether a model understands specialized literature, evaluate experimental designs, generate difficult domain questions or check whether an explanation separates established findings from speculation. They may also identify hallucinated citations, incorrect technical reasoning and subtle misuse of specialist terminology.
The reported work extends beyond computer science into fields including chemistry, biology, conservation, linguistics and medicine. In these areas, a generic web search cannot reliably tell an AI whether an answer is scientifically sound. Expert judgment supplies the distinction between an answer that sounds convincing and one that meets professional standards.
The central contradiction: expertise can become automation data
The reported investigation by The Verge and New York Magazine describes professionals doing work intended to improve systems that may perform parts of professional jobs. That creates a replacement feedback loop:
Rank #2
- A professional performs a task and explains what good work looks like.
- The explanation becomes an example, label, rubric or evaluation benchmark.
- An AI system learns to produce or recognize similar work.
- Employers can use the improved system to handle more of the task with fewer people, lower costs or less junior support.
- Experts may then be hired to refine the next, more advanced version.
Calling this “stealing careers” captures a real fear, but it is broader than the evidence proves. The most defensible interpretation is about task substitution, wage pressure, deskilling and control of expertise—not a claim that every lawyer or scientist will lose an entire career.
Free tools Windows power users keep installed
One-click scans. No signup required.
Five different effects are often confused
- Task substitution: AI performs individual activities such as drafting, summarizing, coding or classification.
- Job redesign: Professionals remain employed but are expected to supervise more automated work or produce more output.
- Price pressure: Employers shift work toward cheaper contractors or use AI to justify smaller teams.
- Deskilling: Junior workers lose foundational assignments that once helped them develop into senior practitioners.
- Occupational elimination: The profession itself largely disappears. Current evidence does not establish that outcome for lawyers, scientists or white-collar workers generally.
Why would professionals help automate their own work?
Workers are not necessarily naïve about the contradiction. Many accept these assignments because the immediate alternative is worse.
Possible incentives include:
- Layoffs or reduced demand in the original profession.
- A weak hiring market for educated workers.
- Higher short-term pay than available conventional work.
- Remote access and flexible scheduling.
- The ability to use existing expertise without retraining.
- The hope that AI-related work will lead to a more stable role.
- Immediate expenses such as rent, debt, health costs or family obligations.
That is a financial decision, not necessarily an endorsement of replacing human workers. A person can believe the work may damage their profession over time and still need income this month.
High hourly pay does not mean secure income
Mercor’s current pages advertise examples ranging from roughly $60 to $250 an hour depending on profession and specialization. Examples include writing roles at $75–$100 an hour, cybersecurity roles at $70–$90, a backend-engineering role at $85 and a lawyer occupation-study role at $73. Its engineering page advertises some opportunities at approximately $80–$200 an hour.
These are advertised rates, not guaranteed earnings or salaries. Actual income depends on selection, geography, credentials, project demand, hours offered, acceptance criteria and whether work continues. Mercor says projects may last weeks to months and may be extended, shortened or ended early. A current legal listing describes some independent-contractor work of approximately six to 15 hours per week.
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteThe difference between an hourly rate and dependable income is substantial:
Rank #3
Illustrative example—not a reported typical outcome: $100 an hour for 10 hours in one week produces $1,000 gross. If there is no work during the following three weeks, that is still only $1,000 gross for the month before taxes and expenses—about $250 per week on average, or approximately $4,333 a month if the same pattern repeats.
A contractor also generally has to account for taxes, health insurance, retirement saving, unpaid assessments, equipment, gaps between projects and time spent waiting for tasks. Legal protections vary by jurisdiction, contract and worker classification; independent-contractor status does not mean a worker has no rights, but it commonly means fewer employer-provided protections than a conventional employee.
The hidden management system
Although the work may be advertised as flexible, the reported arrangement can operate like an on-demand labor market. A frontier AI laboratory identifies a weakness, a data supplier recruits specialists, workers produce evaluations, the supplier aggregates and reviews the output, and the laboratory measures whether the model improved. The project may then pause while the next assignment requires a different specialty, lower cost or faster completion.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Workers reported in the investigation described projects starting and stopping abruptly, changing rates or requirements, time monitoring and competition for limited tasks. Other features can include automated interviews, shrinking time targets, Slack-based task allocation and removal from a project with little warning.
This creates an unusual combination: the worker may have a prestigious qualification, but the platform can still treat the person as interchangeable supply for a narrow assignment.
Why confidentiality matters financially
Professional careers normally accumulate value through references, publications, recognizable projects, case studies and reputation. Confidential AI-training work may produce valuable output without giving the contributor portable credit.
Non-disclosure agreements and code names can mean a worker:
Recommended Free Tools
- Does not know the ultimate AI client.
- Cannot describe the assignment publicly.
- Cannot retain examples for a portfolio.
- Fears losing access to future projects if conditions are discussed.
- Cannot easily compare pay and treatment with other contributors.
Confidentiality can protect client information and proprietary systems, but it also changes the worker’s bargaining position. It may make collective action and public documentation more difficult, though an NDA does not automatically make organizing unlawful or impossible in every jurisdiction.
Quality and safety risks
Human expertise is not automatically high quality simply because the worker is highly educated. The conditions of collection matter. Pressure to work faster, inconsistent instructions, changing rubrics and automated monitoring can all reduce the reliability of the data.
Reported and plausible failure modes include:
- Rushing complex judgments to meet shrinking time targets.
- Managers interpreting the same rubric differently.
- Pay systems rewarding speed more than accuracy.
- Workers using AI assistance despite project rules.
- Monitoring software misclassifying legitimate work as inactivity.
- Evaluation criteria changing partway through an assignment.
- Sensitive legal, medical, financial or personal information entering a training pipeline.
- Workers not knowing how their contribution will ultimately be used.
The possibility that workers use AI to generate supposedly human training data also raises concerns about degraded or recursive data. “Model collapse” is a recognized technical concern involving poor synthetic data, but the existence of that risk does not prove that every platform or project is contaminated.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What the evidence does—and does not—show about mass unemployment
The existence of expert AI-training work is evidence of task-level automation and a new market for professional judgment. It is not evidence that AI has already caused universal white-collar job destruction.
Anthropic’s March 2026 labor-market study reports limited employment effects to date and warns that AI exposure is not the same as displacement. An occupation can be highly exposed yet retain demand because AI increases productivity, creates new work, shifts responsibilities or makes services cheaper and therefore more widely used. The study also reports that occupations with greater observed AI exposure are projected by the U.S. Bureau of Labor Statistics to grow more slowly through 2034.
Those findings support two statements at once:
- Near-term mass unemployment among educated professionals is not established by the cited evidence.
- Pressure on specific tasks, wages, entry-level pathways and contractor conditions is increasingly visible.
Company-reported scale also needs context. The investigation said Mercor claimed approximately 30,000 professionals worked on its platform each week, while Scale AI claimed more than 700,000 people in its broader expert network. Those are company claims reported by the investigation, not independently audited totals, and they do not show how many people worked full time, how much they earned or how long assignments lasted.
Who benefits when expertise becomes model capability?
- AI laboratories obtain specialized data needed to improve systems in high-value fields.
- Data vendors recruit, screen, coordinate and aggregate expert labor.
- Workers receive possible short-term income and exposure to frontier AI work.
- Employers and clients may gain cheaper or faster access to legal, scientific, financial or technical reasoning.
- The public may receive more capable tools, but also bears risks involving privacy, accountability, job quality and concentrated ownership.
The unresolved financial question is who owns the value created when a professional’s judgment is converted into reusable model capability. A worker may be paid once for an evaluation that helps a company sell an automated service repeatedly. That does not by itself prove the contract is unfair, but it explains why compensation, attribution, disclosure and bargaining rights matter.
Should you take an AI-training contract?
For someone between jobs, the work may be a rational bridge. For someone who needs predictable income, benefits or a public portfolio, it may be a poor substitute for employment.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Ask these questions before accepting
- Am I paid for logged time, completed tasks or accepted work?
- Are assessments, onboarding or corrections unpaid?
- Is there a minimum number of hours?
- Who can pause or terminate the project, and how much notice is provided?
- Can the client change the rubric, time limit or rate?
- Is monitoring software required, and what counts as productive time?
- Is outside AI assistance prohibited, and how is that rule enforced?
- Does the agreement restrict portfolio use or public discussion?
- Are taxes withheld, or must I make my own payments?
- Are there geographic, licensing, conflict-of-interest or employer restrictions?
- Will the work involve confidential client, patient, legal or personal data?
- Can I keep copies of my contributions or receive any attribution?
- Are disputes subject to arbitration or a particular jurisdiction’s law?
Calculate the income conservatively
Do not multiply the advertised rate by 40 hours a week unless the contract guarantees those hours. Instead, estimate income under three scenarios: the minimum hours promised, the hours you realistically expect and a worst case in which the project pauses. Subtract taxes, insurance, equipment, unpaid screening time and gaps between assignments. Treat any unguaranteed work as variable income rather than as a replacement for a salary.
The larger career risk may be the missing training ladder
Professional work is not only about today’s tasks. Junior drafting, research, coding and analysis often teach workers how to become senior practitioners. If AI absorbs those entry-level assignments, firms may still employ experienced lawyers or scientists while hiring fewer beginners who would have developed into them.
That is why a profession can remain intact in the short term while becoming harder to enter and less secure over time. The danger is not necessarily a sudden emptying of every office. It may be a gradual transfer of judgment, bargaining power and career development from professionals to platforms that capture and scale their expertise.
At the same time, expert contributors have genuine agency. Some want to improve model reliability, influence how systems handle specialized questions or earn money during a difficult transition. The work can complement a career rather than replace it. Its consequences depend on the task, contract, model capability, employer behavior and how the resulting value is distributed.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
