Which AI labs are hiring for the teams that buy, run and evaluate human data, and for what roles?
On 18 Oct 2026, 16 of 25 AI labs with a public job board had 101 roles open on the teams that buy, run and evaluate human data, xAI the most (24).
- human-data roles open at AI labs on 18 Oct 2026
- 101
- labs read with human-data roles open
- 16 of 25
18 Oct 2026 to 18 Oct 2026
Which AI labs and data vendors have people whose job is to source, license or acquire training data right now?
On 18 Oct 2026, 15 roles at AI labs and human-data vendors had the job of sourcing, licensing or acquiring training data, the most at OpenAI, and 20 more were building RL environments: the desks data owners and brokers sell to.
- data-acquisition and partnership roles open
- 15
- companies hiring for them
- 8
18 Oct 2026 to 18 Oct 2026
How fast are the human-data vendors hiring their own staff, and for which functions?
On 18 Oct 2026, 14 human-data vendors had 671 staff roles open on their own boards, Scale AI the most (182); engineering and product is the largest function.
- open staff roles at 14 human-data vendors on 18 Oct 2026
- 671
18 Oct 2026 to 18 Oct 2026
What do AI labs spend on human data, what do vendors bill, what do the experts receive, and what do RL environments and tasks cost?
59 published figures on what AI labs pay for human data, from 2021 to 11 Oct 2026: lab spend, vendor bill rates against worker pay, vendors' take, RL-environment prices, prices per task and contract sizes; the largest spend figure recorded: Surge reportedly generated $1.2 billion in revenue in 2024 from work with AI labs (Surge AI, reported).
- price points on human data for AI
- 59
- Surge AI (OpenAI, Google, Anthropic, Meta), per year (vendor revenue)
- $1.2B
1 Jan 2021 to 11 Oct 2026
Which kinds of training data are in rising demand, and which are cooling?
To 11 Oct 2026, the evidence shows labs buying more coding (production-like environments), computer-use app and website clones and egocentric and robotics demonstration data and less generalist annotation and microtasks and static maths and verifiable coding tasks.
- kinds of data in rising demand
- 12
20 Jun 2023 to 11 Oct 2026
Who has licensed content and data to which AI companies, on what terms and for how much?
From 11 Jul 2023 to 11 Oct 2026, 89 content and data licensing deals between AI companies and data owners are recorded, 22 with a published price; the largest is News Corp with OpenAI ("Could be" close to $250 million over five years, per The Wall Street Journal (as reported by VentureBeat); terms not disclosed by the parties).
- licensing deals recorded
- 89
- with a stated or reported price
- 22
11 Jul 2023 to 11 Oct 2026
Who buys companies' records and the archives of defunct companies to train AI, at what prices and on what terms, and what do brokers earn?
Companies' records (email, chat, tickets, code, CRM) and the archives of defunct companies now sell to AI labs and principal vendors as raw material for RL environments: 62 prices, programs and terms recorded to 11 Oct 2026, up to $12,500,000 for a single archive and a $1 billion over the next 12 months buying program at micro1.
- prices, programs and terms recorded
- 62
- largest recorded price (Spirit Airlines (debtor))
- $12.5M
12 Jan 2026 to 11 Oct 2026
Which copyright, privacy and data-rights disputes over AI training data have been filed, ruled on or settled, and what did they cost?
From 20 Sep 2023 to 11 Oct 2026, 68 lawsuits, rulings, settlements and regulatory actions over AI training data are recorded; the largest amount is Bartz v. Anthropic ($1.5 billion, approximately $3,000 per work).
- cases, rulings and settlements over training data
- 68
- Bartz v. Anthropic, the largest amount
- $1.5B
20 Sep 2023 to 11 Oct 2026
How often do workers use AI to do human-data tasks, how do fake and resold accounts get in, and which vendors have had quality failures or breaches?
From 13 Jun 2023 to 11 Oct 2026, 41 studies, incidents and vendor statements document the quality risks of human data: workers using AI models to do the tasks (33-46% of crowd workers (range estimate) in the first measurement), fake and resold accounts, quality failures on lab projects and breaches at vendors.
- studies, incidents and vendor statements
- 41
- of crowd workers used an AI model on a task, in the first measurement
- 33-46%
13 Jun 2023 to 11 Oct 2026