xAI is the one large employer that hires under the title AI Tutor: 32 open roles on 3 October 2026, mostly language audio work at $35–45 an hour in the US, with specialist roles up to $100. This guide covers the roles and terms, the 2025 layoff of generalist tutors, and the same work at other platforms under other names.
Multimango is not a company you apply to. It is a web tool where contractors rate and annotate AI outputs, and its terms say you are engaged and paid by a separate administrator: in practice the vendor you already work through, such as Outlier, TELUS Digital, RWS, Mercor or CrowdGen. Here is what its terms say, how people get access, and what changed in 2026.
A drought is what DataAnnotation workers call weeks or months with an empty dashboard on an open account. DataAnnotation does not acknowledge droughts; its FAQ says work depends on demand for your skills and your performance. What workers believe causes them, how long 2026 droughts lasted by their accounts, and what to do meanwhile.
Learn how to test AI models with clear tasks, scoring rules, safety checks, and evidence that help evaluators deliver reliable paid work across projects.
Outlier is a real platform operated by Scale AI through a subsidiary called Smart Ecosystem Inc., and it pays weekly through PayPal, Airtm or ACH. Its degree rule, unpaid onboarding and uneven task supply are what catch people out. Here is what Outlier publishes about itself, what happened at Scale AI in 2025, and what contributors say.
Data science and machine learning AI training jobs pay practitioners to catch the statistical and modelling mistakes AI makes and to build the tasks that expose them. From 36 listings on this site: the four kinds of work, what they pay (median midpoint $85 an hour) and who qualifies.
Writing and education AI training jobs pay journalists, editors, writers and teachers to rewrite, mark and teach on the record for AI labs. From 16 listings on this site: the three kinds of task, what they pay (median midpoint $67.50 an hour) and what counts as a qualification.
Video, image and design AI training jobs range from filming your own chores to judging generated video frame by frame and critiquing design work. From 38 listings on this site: the five kinds of task, what each pays ($6 to $350 an hour) and who qualifies.
Audio and voice AI jobs cover recording, transcription, speech data auditing, listening QA and audio engineering. From 107 listings on this site: what each type involves, what it pays (median midpoint $28 an hour, voice cloning up to $150) and what you need to qualify.
Psychology and mental health AI training jobs pay psychologists, psychiatrists and specialist clinicians to review cases, write scenarios and judge how AI responds to distress. From 20 listings on this site: the tasks, what they pay (16 of 20 at $100 to $200 an hour) and who qualifies.
Engineering AI training jobs pay mechanical, electrical, manufacturing and nuclear engineers to build and check the technical work AI systems are learning. From 98 engineering listings on this site: the work by discipline, what each pays (median midpoint $75 an hour) and who qualifies.
STEM AI training jobs pay mathematicians, physicists, chemists and biologists to write, solve and check the problems AI models are tested on. From 64 science and maths listings on this site: the four kinds of task, what each field pays (median midpoint $80 an hour) and who qualifies.
Finance AI training jobs pay bankers, analysts, accountants and underwriters to build and mark the financial work AI models attempt. From 56 finance listings on this site: the kinds of task, what they pay (median midpoint $85 an hour), which experience counts and who hires.
Medical AI training jobs pay doctors, nurses and other clinicians to check and stress-test how AI handles medicine. From 45 healthcare listings on this site: the main task types, what they pay (physician roles median $180 an hour), which credentials count and where the lower doors are.
Legal AI training jobs pay lawyers to write, grade and review the legal work AI models produce. From the 56 legal listings on this site: the three kinds of task, what they pay (median midpoint $110 an hour), which credentials matter and who is hiring.
Data annotation is adding labels to audio, video, documents and text so a model can learn from them. What one task involves, who hires, what the 27 annotation roles on this site pay, and how to start.
RL environment jobs pay software engineers to build the tasks, reference solutions and verifiers AI agents are trained and graded on. What one environment contains, who hires, and what the 14 listings on this site pay.
AI safety and policy review jobs pay you to judge whether a model handled a sensitive request correctly under a written policy. What one task involves, who hires, and what the listings on this site pay.
Prompt writing and prompt evaluation jobs pay you to write hard, realistic questions for AI models and to check whether a question is good enough to test one. What a task involves and what 36 listings on this site pay.
Side by side comparison tasks ask you to pick the better of two or more AI outputs and say why. What one task involves, the 24 listings on this site that describe it, and what they pay.
AI response evaluation jobs pay you to judge a model's answer against a rubric and explain why. What one task involves, who hires (133 listings on this site) and what it pays, from $14 to $400 an hour.
What RLHF is, what a worker actually does in an RLHF task (comparing, rating and rewriting model answers), who hires for it and what the 26 listings on this site that mention it pay.
AI red teaming jobs pay people to test where a model's safeguards fail and write up what they find. What one task involves, the three kinds of listing, and what the 56 red teaming listings on this site pay.
For most people, no: most AI training roles are contract work sized for 20 hours a week or fewer, with no benefits and pay that often depends on output. A few are genuine full-time placements. What the 699 listings on this site say about hours, contracts and the arithmetic.
You do not need to be a native English speaker for most AI training work. Of the 699 listings on this site, 552 name no country restriction, 172 require another language, and only a handful ask for native-level English. What that means for speakers in Africa, Asia and elsewhere, and the payment check to make first.
AI training platforms pay through Stripe, Wise, PayPal and a few others, and each of those services has its own list of countries it serves. Which platform on this site pays through what, and where each payout service says it is unavailable or limited, from the services' own pages.
Plenty of data annotation jobs are open without a degree, but not the best-paid ones. Across the 699 listings on this site, 126 state that no degree is asked for and 216 name a degree or professional qualification. Here is where each group sits and what it pays.
The United States has income tax treaties with many countries, and they can reduce or remove US tax on some US source income paid to residents of those countries. How the IRS describes the mechanism, how a treaty claim is made on the W-8BEN or Form 8233, and where to read your own country's treaty.
Some non-US contractors see US tax withheld from AI training payments; many see none. How the IRS rules on withholding for foreign persons work, what decides whether income counts as US source, where the W-8BEN fits, and what the IRS says about services performed outside the United States.
Contractors doing AI training work for US platforms may receive a year-end tax form, or none. What Form 1099-NEC and Form 1042-S each report, who tends to get which, the thresholds and dates in the current IRS instructions, and why some contractors receive neither.
DataAnnotation is a real platform run by Surge Labs, Inc., the company behind Surge AI, and it pays through PayPal. It is also the platform people most often describe as silent: no word after the assessment, weeks without tasks, and accounts closed without explanation. Here is what it states, what workers report, and what it leaves unsaid.
The category is real: Scale AI, Labelbox and Handshake all run one, and micro1 publishes peer-reviewed research on its own interviewer. That is exactly why the imitations work. Here is what a genuine platform does, the things that feel wrong but are normal, and the signals that mean stop, drawn from what the five platforms on this board actually publish.
"AI training" covers three different jobs: labeling data, judging what a model produced, and doing your own professional work so a model can learn from it. They pay differently, screen differently, and want different people. Here is what separates them, and which one the listings on this board are actually offering.
Every listing shows its pay and who it is open to.