Weekly market news · 26 Oct to 1 Nov 2026
The week in AI training work: privacy complaints against Scale AI, Innodata cuts jobs, agentic roles grow
Week of 26 Oct to 1 Nov 2026: open roles fell 2.1% on the platforms read the same way all week, agentic roles grew by nearly half, and noyb filed GDPR complaints against Scale AI.
AITraining.jobs research team · Published
On the 56 platforms we read the same way all week, open AI training roles fell 2.1%, from 4,419 at the start of 26 Oct to 4,324 at the end of 1 Nov 2026, as 534 removals outran 447 new listings. Law, medicine and finance roles fell 5.3% and AI rating roles 14.4%, while agentic tasks and RL environments grew from 84 to 124. The board as a whole went from 4,381 to 4,374 open roles on 62 platforms.
Outside our data, a privacy group filed 5 GDPR complaints over how Scale AI's Outlier watches its European workers, Innodata let go of nearly 550 people in India after a customer ramped down, and Appen said seven of the eight largest US AI model builders are now its customers. micro1 said it pays up to $2 million for one company's internal data, and Plato raised $16 million to pay professionals to rebuild their jobs as simulations for AI to train on.
- roles open on the board at the end of 1 Nov 2026
- 4,374
- change in open roles on the 56 platforms read the same way all week
- -2.1%
- new listings on those platforms
- 447
- platforms whose own job board was read for the first time
- 6
Where the open roles moved
Law, Medicine & Finance remained the largest kind of AI training work, with 1,532 open roles on the board at the end of 1 Nov. On the comparable platforms it fell from 1,614 to 1,529 (-5.3%), as 272 removals outran 195 new listings. AI evaluation and rating (RLHF) fell furthest among the larger kinds, from 181 to 155 (-14.4%), with 7 new listings against 33 removals. Coding and software evaluation went from 325 to 308 with only 9 new listings, multilingual work from 913 to 878, and math and STEM from 257 to 248.
Three kinds grew. Agentic tasks and RL environments rose from 84 to 124 open roles (+47.6%), with 46 new listings and 6 removals; AfterQuery's first 12 roles of that kind, opened on 29 Oct, are part of it. Robotics and physical-AI data went from 77 to 97, and 30 of its new listings were Mindrift's of 28 Oct, which, read closely, were one paid in-person study in the Los Angeles area, filming household tasks with a wearable camera at $200 a session, posted under many titles: a count of listings is not a count of openings. Transcription, audio and speech went from 607 to 638, led by micro1's 23 new roles of 27 Oct.
Among the comparable platforms, Weekday moved most: no new listings and 115 removals took it from 223 to 108 open roles. Alignerr went from 419 to 363 (52 new, 108 removed) and Mercor from 259 to 232. micro1 grew from 271 to 297 with 103 new listings, Mindrift from 66 to 90, and Handshake AI from 167 to 184 with 17 new listings and none removed. No country went from no open roles to some, or from some to none, on these platforms.
Pay
The median advertised hourly rate of the open roles on the comparable platforms held at $40. Within the kinds, RLHF evaluation eased from $50 to $43.75, coding from $62.50 to $60, writing from $60 to $55 and research studies from $100 to $75, while Law, Medicine & Finance held at $75, multilingual work at $32.50, transcription and audio at $30 and math and STEM at $40. A median moves when the mix of open roles moves, so these say which roles were open, not that anyone got a raise.
Some of the week's new programs advertised more than the same platform's earlier roles of the same kind. micro1's new transcription and audio roles of 27 Oct advertise a typical $50/hr, against $25/hr on its roles of that kind already open; Mercor's 9 new multilingual roles of 26 Oct a typical $50/hr, against $10/hr on those open on 22 Oct; and RWS TrainAI Community's 5 new multilingual roles a typical $25/hr, against $11/hr. AfterQuery's 12 agentic roles and 13 software roles of 29 Oct advertise a typical $80/hr. These are advertised rates, not guaranteed earnings.
New on our board this week
On the platforms we already read, the hourly detection caught new kinds of work: Mindrift's first 30 robotics data listings (28 Oct), AfterQuery's first 12 agentic and RL environment roles (29 Oct), Sovrano AI's 5 math and STEM roles (28 Oct), Volga Partners' 3 AI evaluation roles (30 Oct) and micro1's first 2 studies and surveys. Braintrust opened 5 roles in medicine and nursing on 30 Oct, at a typical $40/hr. New languages appeared too, 2 roles each: Burmese and Persian at Welo Data, Malayalam at RWS TrainAI Community and Bhojpuri at TransPerfect DataForce.
Some of the week's changes were in how we read the market, not in the market, and those roles count on neither side of the comparison. From 30 Oct we read Scale AI's Outlier postings carried on the Joveo job-ad network beside Outlier's own job board, 44 roles more; Appen's board was read more narrowly from 27 Oct, 14 roles fewer. Six platforms appeared in our feeds for the first time on 29 Oct with 1 open role each: Edison Scientific, EvolutionIQ, Inworld AI, JazzX AI, Kastle and Trellis.
What the AI labs are hiring for
On 2 Nov 2026, 16 of the 25 AI labs whose public job boards we read had 102 roles open on the teams that buy, run and evaluate human data, 2.9% of their 3,559 open roles, against 90 on 26 Oct. xAI had the most, 26 of its 308, up from 15 a week earlier.
AI tutors and raters are again the most common of these jobs, 28 roles at 4 labs, against 17 a week earlier, when evaluations led. 6 roles buy or license data, at Mistral AI, OpenAI, Reflection AI and Thinking Machines Lab. 38 of the staff roles state a yearly salary range, the middle of the median range being $284K, and the 32 contract roles that state an hourly rate pay $18 to $100 an hour. This week's podcast episode is about what happens after an expert is accepted by a platform.
Platform news
What was reported about the platforms this week, each item with the outlet that reported it.
A privacy group files GDPR complaints in four countries over how Scale AI's Outlier watches its workers
noyb, with Intérêt à Agir, DiPLab and Reversing.Works, filed 5 GDPR complaints against Scale AI in France, Italy, the Netherlands and Austria on 27 Oct. They allege that Outlier, which noyb says has more than 900,000 freelance workers, puts trackers on their devices without consent, makes them repeat face-video and ID checks through Persona, and left their requests for their own data unanswered, and they ask the authorities to investigate and fine the company. Scale AI told The Register the claims are unfounded and inaccurate and that Outlier's practices are disclosed to contributors and require their consent. These are complaints; no authority has ruled. They follow the DiPLab report in our last post.
Sources: noyb, 27 Oct 2026; The Register, 27 Oct 2026
· Innodata
Innodata lets go of nearly 550 people in India after a customer ramps down, and its shares fall a quarter
Innodata ended the jobs of nearly 550 employees in India effective 26 Oct, its termination email citing "the ramp-down of the business assignment for which you were engaged"; cuts in the Philippines were reported only from employees' own posts. Its shares fell about 25% on 27 Oct. Innodata has not said which customer ramped down; its report for the quarter to 30 Jun 2026 showed two customers supplying 37% and 34% of its revenue, which is how one contract can move a whole workforce. It reports third-quarter results on 5 Nov.
Sources: People Matters, 27 Oct 2026; Protos, 28 Oct 2026
· Appen
Appen's quarterly revenue rises 34% and it now counts seven of the eight largest US model builders as customers
Appen reported revenue of $73.9 million for the quarter to 30 Sep 2026, up 34% on a year earlier, of which $46.1 million came from Appen China. It said it added two of the leading US frontier model builders as customers in the quarter, including a former large customer, and that it now counts "seven of the eight largest US-based AI model builders as customers". It kept its full-year guidance of $270 million to $300 million in revenue, likely in the lower half. The figures are unaudited.
Source: Appen, ASX quarterly activity report, 29 Oct 2026
micro1 says it pays up to $2 million for one company's internal data, to build training environments for AI
Business Insider reported that micro1 is buying companies' internal Slack messages, Google documents, Confluence pages, CRM records and design files, which labs use to build realistic training simulations, and that micro1 "says it pays $100,000 to $2 million for a single midsize company's data", within the $1 billion it plans to spend over the next year. The chief executive of the recruiting startup VanHack said he also had an offer from Mercor before selling to micro1. The prices are micro1's own figures, and we found the story only at Business Insider.
Source: Business Insider, 26 Oct 2026
· iMerit
iMerit launches iMerit Scholars, a platform to recruit and pay domain experts for AI labs
iMerit, bought by EXL this summer, launched iMerit Scholars on 26 Oct to recruit, assess, assign and pay specialists who evaluate and fine-tune AI models, with hourly, per-deliverable and per-outcome pricing. It says its community already counts more than 30,000 vetted professionals in 118 countries and that it expects to grow to hundreds of thousands; one customer, it said, asked for about 400 extra workers within three or four days. The figures are the company's own.
Source: SiliconANGLE, 26 Oct 2026
Plato raises $16 million to pay professionals to rebuild their jobs as simulations for AI
Plato, which builds simulated workplaces where AI models practise jobs in customer support, people operations, finance, accounting and healthcare administration, raised a $16 million Series A led by Nava Ventures, with Microsoft AI's chief executive Mustafa Suleyman among its investors, bringing its total to $19 million. It says it works with domain experts to recreate those jobs and will spend the money partly on "paying professionals to recreate their jobs in simulated environments". The figures are from the company's own release.
Source: Plato release (PR Newswire), via Yahoo Finance, 28 Oct 2026
A US rule that makes it easier to classify workers as contractors clears White House review
The Labor Department's rule that would make it easier for employers to classify workers as independent contractors cleared White House review on 30 Oct, Bloomberg Tax reported, which sets the department up to finalize it. Most AI training work is offered as contract work, and the wage lawsuits against Handshake AI and Scale AI turn on exactly that classification. The story is paywalled after its first paragraph, and we found it at one outlet only.
Source: Bloomberg Tax, 30 Oct 2026
Two US senators ask Meta what smart-glasses footage human reviewers see when it trains its AI
Senators Ron Wyden and Ed Markey wrote to Meta on 28 Oct after reports that contractors reviewing footage from its smart glasses saw intimate scenes. As 404 Media quotes it, their letter says the imagery may be used for AI training "including with the assistance of human reviewers" and asks what safeguards exist and what Meta trains on. We could not open the letter itself, only 404 Media's report of it.
Source: 404 Media, 29 Oct 2026
The week in our data
26 Oct 2026 to 1 Nov 2026, read from the job boards of the 62 platforms with open roles on AITraining.jobs. Each count opens the live list, which keeps moving after this week.
Open roles by kind of work
The board's count at the end of 1 Nov 2026, then the same kind on the 56 platforms read the same way all week, from the start of 26 Oct 2026 to the end of 1 Nov 2026. Median pay is the middle advertised hourly rate, shown where at least 3 roles state one at both ends.
| Kind of work | On the board | Comparable, start to end | New listings | Median pay, start to end |
|---|---|---|---|---|
| Law, Medicine & Finance | 1,532 | 1,614 to 1,529(-5.3%) | 195 | $75 to $75 |
| Multilingual & Localization | 878 | 913 to 878(-3.8%) | 49 | $32.50 to $32.50 |
| Transcription, Audio & Speech | 640 | 607 to 638(+5.1%) | 64 | $30 to $30 |
| Coding & Software Evaluation | 308 | 325 to 308(-5.2%) | 9 | $62.50 to $60 |
| Math & STEM | 248 | 257 to 248(-3.5%) | 7 | $40 to $40 |
| Data Annotation & Labeling | 241 | 248 to 241(-2.8%) | 20 | $22 to $21 |
| AI Evaluation & Rating (RLHF) | 200 | 181 to 155(-14.4%) | 7 | $50 to $43.75 |
| Agentic Tasks & RL Environments | 124 | 84 to 124(+47.6%) | 46 | $50 to $50 |
| Robotics & Physical-AI Data | 97 | 77 to 97(+26%) | 36 | $10 to $10 |
| Red-Teaming & AI Safety | 55 | 55 to 55(0%) | 3 | $80 to $78.75 |
| Writing & Content | 33 | 37 to 33(-10.8%) | 8 | $60 to $55 |
| Research Studies & Surveys | 18 | 21 to 18(-14.3%) | 3 | $100 to $75 |
What changed in what the board reads
A platform read for the first time is left out of every "rose" or "fell" in this post; on the other days below our reading moved a platform's count, not the market, and those roles count on neither side.
- Edison Scientific: its own job board read for the first time on 29 Oct 2026, 1 roles open at the week's end.
- EvolutionIQ: its own job board read for the first time on 29 Oct 2026, 1 roles open at the week's end.
- Inworld AI: its own job board read for the first time on 29 Oct 2026, 1 roles open at the week's end.
- JazzX AI: its own job board read for the first time on 29 Oct 2026, 1 roles open at the week's end.
- Kastle: its own job board read for the first time on 29 Oct 2026, 1 roles open at the week's end.
- Trellis: its own job board read for the first time on 29 Oct 2026, 1 roles open at the week's end.
- Outlier (Scale AI): this site began reading Scale AI's Outlier campaign on the Joveo job-ad network besides Outlier's own job board, on 30 Oct 2026, 44 roles that came into or left view with it count on neither side.
- Appen: a batch out of line with its week on 27 Oct 2026, 14 roles removed that day count on neither side.
Biggest moves on the comparable platforms
- Weekday: 223 to 108 open roles, no new listings and 115 removed.
- Alignerr: 419 to 363 open roles, 52 new listings and 108 removed.
- Mercor: 259 to 232 open roles, 38 new listings and 65 removed.
- micro1: 271 to 297 open roles, 103 new listings and 77 removed.
- Mindrift: 66 to 90 open roles, 94 new listings and 63 removed.
- Handshake AI: 167 to 184 open roles, 17 new listings and 0 removed.
Countries opened or closed
No country went from no open roles to some, or from some to none, on the comparable platforms this week.
How this post is counted
- The week is the seven full days (UTC) before the day of publication. A role is open at a moment when the platform's board lists it then; a new listing is one dated in the week; a removal is one no longer listed.
- Every "rose" or "fell" is measured on the platforms read since before the week, each role dated by the day the platform says it posted it (else the day we first saw it). A platform whose own board was read for the first time is left out. A day on which our feed reached roles the platform posted earlier (its undated sightings more likely that than new roles, by a model fitted on the roles that are dated), or on which a platform's removals were out of line with the rest of its own week, is how a change in what the board reads looks, not how hiring looks: those roles count on neither side.
- Pay is the advertised hourly rate (the middle of a stated range); a median is given only where at least 3 roles state one. A country opened when no role was open to it as the week began and some are at its end, and closed the other way round.
- Platform news comes from the outlet or company named under each item, linked and dated. The post's words are fixed on its day; a correction is added below them with its date.
Cite and follow
AITraining.jobs (2026). The week in AI training work: privacy complaints against Scale AI, Innodata cuts jobs, agentic roles grow. https://aitraining.jobs/news/2026-11-02
Free to quote with credit under CC BY 4.0. The monthly view of the same market is in the monthly data stories, every figure in the research reports.