Key findings
- 1
Between the week of 7 Sep 2026 and the week of 21 Sep 2026, open roles in Red-Teaming & AI Safety on the platforms tracked since 1 Sep 2026 went down 15.7%, from about 58 to 49, against -2.1% for all AI training work on the same platforms; in the part week before (1 to 6 Sep) the panel averaged about 29.
- 2
3 platforms listed the 38 open roles in Red-Teaming & AI Safety; Mercor held 81.6% (HHI 6,911, highly concentrated, against 1,680 for all AI training work).
- 3
34 of the 38 open roles in Red-Teaming & AI Safety (89.5%) stated an hourly rate, with a median of $39/hr (middle half $19 to $55), 33% below the $58/hr of all AI training work.
- 4
73.7% of the 38 open roles in Red-Teaming & AI Safety accepted applicants from any country, against 37.7% of all AI training roles; of the 20 countries compared, residents of Germany and the United Kingdom (34 each), then the United States (31) could apply to the most, and residents of 11 of the countries (among them Argentina and Bangladesh) to the fewest (28 each).
- 5
The words used most in the titles of open roles in Red-Teaming & AI Safety were Expert or SME (28), Red teamer or tester (7) and Analyst or researcher (2), of 38 open roles.
- 6
Of the 32 open roles in Red-Teaming & AI Safety with a full description, 3.1% mention a PhD or master's degree, none a professional credential and none an assessment or test.
- 7
Over the 30 days from 1 Sep 2026 to 30 Sep 2026, open roles in Red-Teaming & AI Safety on the platforms tracked at its start went up 100.0% (16 to 32): platforms listed 71 new roles and removed 75, and a removed role had been listed for a median of 17 days (32 across all AI training work).
Figure 1
Open roles, week by week
Is the number of open roles in Red-Teaming & AI Safety rising or falling?
| Week | Open roles (fixed panel) | Index |
|---|---|---|
| 1 to 6 SepPartial week | 29 | 50.0 |
| 7 to 13 Sep | 58 | 100.0 |
| 14 to 20 Sep | 54 | 92.6 |
| 21 to 27 Sep | 49 | 84.3 |
| 28 to 30 SepPartial week | 45 | 77.8 |
Sample. Average open roles in Red-Teaming & AI Safety over each week's days, on the 15 platforms tracked since 1 Sep 2026 (the fixed panel; 2 of them list this work), 1 Sep 2026 to 30 Sep 2026.
The index is the panel's open roles against the week of 7 Sep 2026 (100), beside the same index for all AI training work on the same platforms.
OpenTrain AI (first tracked 5 Sep 2026) lists this work but was added to the dataset later, so it is left out of the panel and counted only in the last column: a platform's arrival never reads as growth.
Weeks with fewer than 10 open roles on the panel, or cut by the span, are marked.
Source. AITraining.jobs Research, The Market for Red-Teaming & AI Safety Work, September 2026 edition.
Figure 2
Listings added and removed
How fast do roles in Red-Teaming & AI Safety turn over, and is the stock growing on the same platforms?
| Work | Open on 1 Sep 2026 | Open on 30 Sep 2026, same platforms | Change | New listings |
|---|---|---|---|---|
| Red-Teaming & AI Safety | 16 | 32 | +100.0% | 71 |
| All AI training work | 1,591 | 1,656 | +4.1% | 1,585 |
Sample. Roles listed from 1 Sep 2026 to 30 Sep 2026: 113 in Red-Teaming & AI Safety, 4,472 across all AI training work.
Change compares the roles open on the first and the last day of the window on the platforms tracked when it opened, so a platform added during the window does not read as growth.
A new listing first appeared in the window on a platform already tracked before that week; a platform's first week (its back catalogue) is not counted.
Median days listed runs from a removed role's listing date to its last sighting, for removed roles with a known listing date (at least 10; 75 here).
Source. AITraining.jobs Research, The Market for Red-Teaming & AI Safety Work, September 2026 edition.
Figure 3
Who lists it
Which platforms list roles in Red-Teaming & AI Safety, and what does each advertise?
| Platform | Open roles | Share of open roles | Median per hour |
|---|---|---|---|
| Mercor | 31 | 81.6% | $33 |
| OpenTrain AI | 6 | 15.8% | n/a |
| AfterQuery | 1 | 2.6% | n/a |
Sample. 38 roles in Red-Teaming & AI Safety open at the end of 30 Sep 2026, by the platform that lists them: HHI 6,911 (highly concentrated), against 1,680 for all AI training work.
HHI: the sum of every platform's squared percentage share of the open roles, 0 to 10,000; under 1,500 is unconcentrated, 1,500 to 2,500 moderately and above 2,500 highly concentrated (US merger-guideline thresholds).
Shares are of listings, not of people hired or money paid: a platform posting one role per language or country weighs more here than its hiring may.
A median needs at least 8 roles stating an hourly rate.
Source. AITraining.jobs Research, The Market for Red-Teaming & AI Safety Work, September 2026 edition.
Figure 4
What it pays
What do roles in Red-Teaming & AI Safety advertise per hour, against AI training work as a whole?
| Work | Median per hour | 25th percentile | 75th percentile | Stating an hourly rate |
|---|---|---|---|---|
| Red-Teaming & AI Safety | $39 | $19 | $55 | 89.5% |
| All AI training work | $58 | $30 | $75 | 72.2% |
Sample. Roles open on 30 Sep 2026: 38 in Red-Teaming & AI Safety, 34 of them stating an hourly rate; 2,726 across all AI training work.
Pay is the hourly figure the posting advertises (the middle of a stated range). The middle half runs from the 25th to the 75th percentile.
Roles paid per task or per project, or that state no figure, do not state an hourly rate. A median needs at least 8 roles stating one.
Source. AITraining.jobs Research, The Market for Red-Teaming & AI Safety Work, September 2026 edition.
Figure 5
How its rates are distributed
What share of roles in Red-Teaming & AI Safety pay $50 an hour or more?
| Advertised rate | Open roles | Share of priced roles | All AI training work |
|---|---|---|---|
| Under $15/hr | 0 | 0.0% | 7.3% |
| $15 to $25/hr | 13 | 38.2% | 10.1% |
| $25 to $35/hr | 3 | 8.8% | 10.8% |
| $35 to $50/hr | 5 | 14.7% | 17.8% |
| $50 to $75/hr | 10 | 29.4% | 28.2% |
| $75 to $100/hr | 1 | 2.9% | 10.8% |
| $100 to $150/hr | 2 | 5.9% | 8.0% |
| $150/hr or more | 0 | 0.0% | 6.9% |
Sample. 34 roles in Red-Teaming & AI Safety open on 30 Sep 2026 that state an hourly rate, against 1,967 across all AI training work.
Each role counts once, at the middle of its advertised range. The bands are those of the pay-transparency report.
Source. AITraining.jobs Research, The Market for Red-Teaming & AI Safety Work, September 2026 edition.
Figure 6
Applicant demand per open role
Do roles in Red-Teaming & AI Safety draw more or fewer applicants per opening than the average AI training role?
| Roles | Clicks per role a week | Index |
|---|---|---|
| Red-Teaming & AI SafetySmall sample | 0.11 | 0.53x |
| All AI training work | 0.21 | 1.00x |
| Its roles stating an hourly rateSmall sample | 0.10 | 0.45x |
| Its roles stating no hourly rateSmall sample | 0.60 | 2.83x |
| Its roles open worldwideSmall sample | 0.12 | 0.57x |
| Its roles limited by countrySmall sample | 0.10 | 0.47x |
Sample. Apply clicks on a specific role on AITraining.jobs, from 1 Sep 2026 to 30 Sep 2026: 28 on roles in Red-Teaming & AI Safety, against the days those roles were open.
Clicks per role a week = apply clicks / (open role-days / 7). Index 1.00 is the rate of every role on the site.
Rows with fewer than 30 clicks or 3 roles open on an average day are marked as a small sample.
Apply clicks measure interest among visitors to AITraining.jobs, who see roles open in their country and recent, priced roles first. Clicks are never split by platform.
Source. AITraining.jobs Research, The Market for Red-Teaming & AI Safety Work, September 2026 edition.
Figure 7
Applicant demand by listing age
Do newer roles in Red-Teaming & AI Safety draw more applicants than older ones?
| Listing age | Clicks per role a week | Index (this work = 1) |
|---|---|---|
| Listed during the windowSmall sample | 0.08 | 0.75x |
| Listed up to 4 weeks before itSmall sample | 0.00 | 0.00x |
| Listed 4 to 12 weeks before itSmall sample | 0.18 | 1.59x |
| Listed more than 12 weeks before itSmall sample | 0.12 | 1.07x |
Sample. Apply clicks on roles in Red-Teaming & AI Safety, from 1 Sep 2026 to 30 Sep 2026, by when each role was listed relative to the window's first day (1 Sep 2026).
Listing date: the date the job board gives for the posting, or else the first day the role appeared in the feed. A role that arrived with a platform's back catalogue and gives no date has none.
The index divides each row's rate by the rate of all roles in this kind of work, so it reads within the kind.
Rows with fewer than 30 clicks or 3 roles open on an average day are marked as a small sample.
Source. AITraining.jobs Research, The Market for Red-Teaming & AI Safety Work, September 2026 edition.
Figure 8
Who may apply: location rules
Can people apply to roles in Red-Teaming & AI Safety from anywhere?
| Location rule | Open roles | Share of its open roles | All AI training work |
|---|---|---|---|
| Open worldwide | 28 | 73.7% | 37.7% |
| Worldwide except listed countries | 1 | 2.6% | 7.3% |
| Only listed countries | 9 | 23.7% | 55.0% |
| No rule stated | 0 | 0.0% | 0.0% |
Sample. 38 roles in Red-Teaming & AI Safety open at the end of 30 Sep 2026, against 2,726 across all AI training work.
The rule is read from each posting's location text and, where the posting is silent, from what the platform states about where it hires.
Source. AITraining.jobs Research, The Market for Red-Teaming & AI Safety Work, September 2026 edition.
Figure 9
Who may apply: roles open to each country
How many roles in Red-Teaming & AI Safety can a resident of each country apply to?
| Country | Open roles | Share of its open roles |
|---|---|---|
| Any country (worldwide roles only) | 28 | 73.7% |
| Germany | 34 | 89.5% |
| United Kingdom | 34 | 89.5% |
| United States | 31 | 81.6% |
| Egypt | 29 | 76.3% |
| Indonesia | 29 | 76.3% |
| India | 29 | 76.3% |
| Philippines | 29 | 76.3% |
| Pakistan | 29 | 76.3% |
| Saudi Arabia | 29 | 76.3% |
| Argentina | 28 | 73.7% |
| Bangladesh | 28 | 73.7% |
| Brazil | 28 | 73.7% |
| Chile | 28 | 73.7% |
| Colombia | 28 | 73.7% |
| Ghana | 28 | 73.7% |
| Japan | 28 | 73.7% |
| Cambodia | 28 | 73.7% |
| South Korea | 28 | 73.7% |
| Nigeria | 28 | 73.7% |
| Palestine | 28 | 73.7% |
Sample. 38 roles in Red-Teaming & AI Safety open at the end of 30 Sep 2026, for residents of a fixed set of 20 countries.
A country's count is the roles open worldwide, plus roles whose list names it, plus roles whose exclusion list does not. Roles with no rule stated count for no country.
The 20 countries are the same in every report of this series: those named most often by the location rules of open AI training roles on 9 Oct 2026, either as the only countries accepted or as excluded.
Median pay is over the roles open to that country that state an hourly rate (at least 8).
Source. AITraining.jobs Research, The Market for Red-Teaming & AI Safety Work, September 2026 edition.
Figure 10
What the roles are called
What are roles in Red-Teaming & AI Safety called: annotator, evaluator, expert or trainer?
| Title word | Open roles | Share of its open roles |
|---|---|---|
| Expert or SME | 28 | 73.7% |
| Red teamer or tester | 7 | 18.4% |
| Analyst or researcher | 2 | 5.3% |
| Engineer or developer | 2 | 5.3% |
| Evaluator or reviewer | 2 | 5.3% |
| Lead or manager | 1 | 2.6% |
| Quality assurance | 1 | 2.6% |
| Specialist | 1 | 2.6% |
Sample. 37 of the 38 roles in Red-Teaming & AI Safety open on 30 Sep 2026 name one of these words in their title (97.4%); 8 title words are named.
A title can use several of these words ('Expert AI Trainer') and counts for each. The vocabularies are those of the job-titles, languages and expert-fields reports.
Median pay needs at least 8 roles stating an hourly rate.
Source. AITraining.jobs Research, The Market for Red-Teaming & AI Safety Work, September 2026 edition.
Figure 11
What the descriptions ask for
Do roles in Red-Teaming & AI Safety ask for degrees, credentials, experience or tests?
| Requirement | Roles | Share of described roles |
|---|---|---|
| Mentions a PhD | 1 | 3.1% |
| Mentions a PhD or master's | 1 | 3.1% |
| Mentions any degree | 10 | 31.3% |
| Mentions a professional credential | 0 | 0.0% |
| States years of experience | 3 | 9.4% |
| Asks for 5 years or more | 2 | 6.3% |
| States weekly hours | 0 | 0.0% |
| Mentions an assessment or test | 0 | 0.0% |
| Mentions an AI-led or video interview | 0 | 0.0% |
| Mentions a background check | 0 | 0.0% |
Sample. 32 roles in Red-Teaming & AI Safety open on 30 Sep 2026 whose job board publishes a full description (AfterQuery and Mercor), 84.2% of its open roles; 850 described roles across all AI training work.
A mention can be a requirement or a preference ('a PhD is a plus'): the figure counts both. The patterns are those of the requirements report.
Only some platforms publish full descriptions, so this figure describes their roles, not every platform's.
Source. AITraining.jobs Research, The Market for Red-Teaming & AI Safety Work, September 2026 edition.
For labs, data vendors and researchers
Buy the data behind this report
The figures above are a summary. Behind them are records we collect ourselves, every day:
9,572 roles tracked, each with its history
Every role on 29 job boards, read daily since 13 Jul 2026: its pay and place, and the first and last day it was open.
6,200 apply clicks in 28 days
Which roles people actually apply to, by field, pay and kind of work. It happens on this site, so nobody else can count it.
39 lab and vendor hiring boards
The job boards of 25 AI labs and 14 human-data vendors, read every day: who is hiring for human data, and who is hiring to buy it.
What you can buy
The records
The rows behind any figure, with their history, not only the totals.
A regular feed
The same figures every week or month, as files.
A custom cut or benchmark
Your countries, languages or fields, or your own platform against the market.
Method and definitions
The public job boards of the AI training platforms this site tracks, read every day. Apply clicks are made on AITraining.jobs; full descriptions come from the platforms whose boards publish them. Every figure is computed by one database query over the window shown, the same query for the live report and for every edition. Each figure states its sample beneath it.
- Kind of work
- Each role is classified from its title and description into one of the twelve kinds of AI training work this site tracks; this report counts the roles classified as Red-Teaming & AI Safety.
- Fixed panel
- The platforms already tracked when the weekly series opens. The index follows their open roles, so a platform added later (such as OpenTrain AI and Welo Data in September 2026) never reads as growth.
- Index
- The panel's average open roles in a week, divided by the same in the base week, times 100; shown beside the same index for all AI training work on the same platforms.
- Open role
- A role listed by a given day and still in its platform's feed that day.
- New listing
- A role that first appeared during the window on a platform already tracked before that week. A platform's first week in this dataset is its back catalogue and is not counted as new.
- Removed listing
- A role that left its platform's feed during the window.
- Days listed
- From a removed role's listing date to the last day it was seen in its feed. A median needs at least 10 removed roles.
- HHI
- Herfindahl-Hirschman Index: the sum of every platform's squared percentage share of open roles, 0 to 10,000. Under 1,500 unconcentrated, 1,500 to 2,500 moderately, above 2,500 highly concentrated.
- Advertised pay
- The hourly figure a posting states (the middle of a stated range). Medians need at least 8 roles stating a rate; the middle half runs from the 25th to the 75th percentile.
- Clicks per open role a week
- Apply clicks on AITraining.jobs on a specific role, divided by the days the roles were open, times 7.
- Demand index
- Clicks per open role a week divided by the same rate across every role on the site. Above 1, the roles draw more applicants than the average role.
- Listing age
- When a role was listed relative to the window's first day: during the window, up to 4 weeks before it, 4 to 12 weeks before or longer. The listing date is the job board's posting date, or else the first day the role appeared in the feed.
- Location rule
- Read from each posting's location text and, where that names no country, from what the posting's own description states: open worldwide, worldwide except listed countries, only listed countries, or no rule stated.
- Open to a country
- A role is open to residents of a country if it is open worldwide, its list names the country, or its exclusion list does not. The 20 countries compared are the same in every report of the series.
- Window
- Live: the last 28 complete days, in UTC, and the last twelve complete ISO weeks (from the week of 3 Aug 2026 at the earliest) for the weekly series. Edition: one calendar month, in UTC, and its weeks.
What this data cannot tell you
- Platforms that recruit without a public job board are not counted. Several large ones invite workers to projects directly.
- A listing is not a hire. Some platforms keep evergreen roles open for months, and some post one listing per language or country.
- Apply clicks measure interest among the people who use AITraining.jobs, and what is shown first gets clicked more. Clicks are never published by platform.
- Full descriptions are published by a few platforms only, so the requirements figure describes their roles and appears only when at least 10 roles of this kind are described.
- A kind of work with few open roles gives figures that move with a handful of postings: rows on too little data are marked as a small sample.
How the job feeds are read and roles are classified: site methodology. How fresh, complete and consistent the data is, checked every day: data quality and coverage.
Editions
The live report covers the last 28 complete days and changes every day. Each complete month is also published as an edition, frozen two days after the month ends, so a figure you cite from an edition never moves.
- Live dataLast 28 complete days, recomputed daily
- September 2026 editionFrozen 10 Oct 2026
- August 2026 editionFrozen 10 Oct 2026
Cite and reuse
Free to quote, chart and republish under CC BY 4.0 with a link to this page. This edition is frozen, so its figures will read the same later.
Citation
AITraining.jobs (2026). The Market for Red-Teaming & AI Safety Work, September 2026 edition. https://www.aitraining.team/research/red-teaming-ai-safety-market/2026-09
BibTeX
@misc{aitrainingjobs_red_teaming_ai_safety_market_2026_09,
author = {{AITraining.jobs}},
title = {The Market for Red-Teaming & AI Safety Work, September 2026 edition},
year = {2026},
howpublished = {\url{https://www.aitraining.team/research/red-teaming-ai-safety-market/2026-09}},
}More research
The same figures for every other kind of work:
- AI Evaluation & Rating (RLHF) market
- Coding & Software Evaluation market
- Math & STEM market
- Law, Medicine & Finance market
- Agentic Tasks & RL Environments market
- Data Annotation & Labeling market
- Robotics & Physical-AI Data market
- Multilingual & Localization market
- Transcription, Audio & Speech market
- Writing & Content market
- Research Studies & Surveys market
- Which AI Labs Are Hiring for Human DataWhich AI labs are hiring for the teams that buy, run and evaluate human data, and for what roles?Read the report
- Who Is Buying AI Training DataWhich AI labs and data vendors have people whose job is to source, license or acquire training data right now?Read the report
- Where Human-Data Vendors Are BuildingHow fast are the human-data vendors hiring their own staff, and for which functions?Read the report
- What AI Labs Pay for Human DataWhat do AI labs spend on human data, what do vendors bill, what do the experts receive, and what do RL environments and tasks cost?Read the report
- What AI Labs Are BuyingWhich kinds of training data are in rising demand, and which are cooling?Read the report
- AI Training Data Licensing DealsWho has licensed content and data to which AI companies, on what terms and for how much?Read the report
- The State of AI Training PayWhat does AI training work pay, by kind of work and by platform?Read the report