The week in AI training work: open roles fall 5.6%, Handshake AI's wage lawsuit is paused for mediation

Law, Medicine & Finance remained the largest kind of AI training work, with 1,575 open roles on the board at the end of 25 Oct. On the comparable platforms it eased from 1,490 to 1,443 (-3.2%), as 176 new listings almost kept pace with 188 removals. The steepest falls among the larger kinds were in multilingual work, down from 930 to 843 (-9.4%) with 21 new listings against 108 removals, and in AI evaluation and rating (RLHF), down from 167 to 143 (-14.4%). Transcription, audio and speech went from 575 to 545, and coding and software evaluation from 327 to 309.

Only two kinds grew, both small: agentic tasks and RL environments, from 74 to 81 open roles with 14 new listings, and robotics and physical-AI data, from 64 to 69 with none removed. Red-teaming and AI safety fell from 57 to 49 and writing from 46 to 38.

Among the comparable platforms, Welo Data moved most: no new listings and 58 removals took it from 237 to 179 open roles. Alignerr went from 432 to 378 (39 new, 93 removed), Turing from 208 to 168, and Mindrift, besides the batch left out of the comparison, from 101 to 56. Of the biggest movers, only micro1 grew, from 243 to 271 with 70 new listings. No country went from no open roles to some, or from some to none, on these platforms.

The median advertised hourly rate of the open roles on the comparable platforms went from $42 to $43. Within the kinds, Law, Medicine & Finance rose from $70 to $75, while RLHF evaluation eased from $55 to $50.50, data annotation from $30 to $25 and writing from $60 to $52.50. Multilingual work held at about $35, transcription and audio at $30, coding at $62.50 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.

The week's new programs on platforms we already read carried some of the highest rates on the board. Mercor opened 13 roles in medicine and nursing on 19 Oct at a typical $200/hr, and micro1's 4 new agentic and RL environment roles of 23 Oct advertise a typical $65/hr, up from $35/hr on its earlier roles of that kind. These are advertised rates, not guaranteed earnings.

20 platforms' own job boards were read for the first time this week. The largest were RWS TrainAI Community with 127 open roles, such as voice recording and speech evaluation in many languages, at a typical $16/hr; DataAnnotation with 112 at a typical $80/hr; and TELUS Digital AI with 66. Surge AI's own board added 13 roles at a typical $300/hr, for journalists, university professors and constitutional lawyers among others. Three AI labs now appear with AI training roles of their own: Anthropic (2, fellowships in conceptual reasoning and STEM), Mistral AI (3) and Cohere (3). Being new to our count does not make any of them new to the market, which is why they stay out of every rise and fall above.

On the platforms we already read, the hourly detection caught new kinds of work. micro1 opened 42 domain-expert roles on 19 Oct (business operations, procurement, transportation and warehousing, finance and accounting among them) and 16 law roles on 20 Oct, including cite-checking work for law students. Mindrift and Handshake AI each opened 3 new roles in medicine and nursing, and TransPerfect DataForce listed its first 2 robotics data roles on 23 Oct.

Two boards were read differently on single days, and those roles count on neither side of the comparison: Mindrift's dropped 220 roles on 22 Oct and Weekday's 101 on 21 Oct, each a batch out of line with the platform's own week. Six boards first read on 18 Oct, between this post's week and the last one's, were read all week and are in the comparison: Weekday, Rise Data Labs, Volga Partners, Lifted (Upwork), Innodata and the Scale AI Human Frontier Collective.

On 26 Oct 2026, 16 of the 25 AI labs whose public job boards we read had 90 roles open on the teams that buy, run and evaluate human data, 2.6% of their 3,515 open roles, against 101 on 17 Oct. xAI still had the most, 15 of its 298, down from 24.

Evaluations are now the most common of these jobs, 22 roles at 10 labs, ahead of human-data operations (20) and AI tutors and raters (17, at Cohere, Mistral AI, Skild AI and xAI, against 25 on 17 Oct). 6 roles buy or license data, at Mistral AI, OpenAI, Reflection AI and Thinking Machines Lab. 37 of the staff roles state a yearly salary range, the middle of the median range being $284K, and the 22 contract roles that state an hourly rate pay $18 to $100 an hour. This week's podcast episode reads what those postings ask of the people who take them.