Updated 8 Nov 2026

AI red teaming and security data companies in 2026, ranked by revenue and funding.

The largest AI red teaming and security companies by revenue on record are Proximal ($200M, run rate, Sep 2026) and Toloka ($26.4M, FY2024). Next, by valuation: Vals AI ($400M). Then, by money raised: Gray Swan AI ($40M) and Sapien ($15.5M). They attack models and agents the way an adversary would, and sell the security tasks, exploits and evaluations that make them safer.

23 companies on our register say on their own pages that they work on security. 10 have revenue, a valuation or funding on record and are ranked below, every figure linked to the filing or article it comes from. Also called AI safety testing companies, AI security evaluation providers.

Hold data these companies buy? AIData.Broker values it free and introduces it to every buyer that takes it, at no cost to you: the buyers pay us. Where it earns more, we run the sale ourselves, for a fee only out of a sale.

  • No data upload
  • About two minutes
  • No email needed

23

AI red teaming and security companies on the list

2

with revenue on record

$226M

a year between them, on record

3

take data from companies

The latest money moves

The newest rounds, valuations and revenue figures among these companies, each linked to its record.

Best for each kind of work

For each thing they sell, the highest-ranked company on this list that names it on its own pages: the largest by the money on record, and how many here do the same work.

  • Best for evaluations and benchmarksProximal, $200M (run rate, Sep 2026): the largest of the 10 here that sell evaluations and benchmarks.
  • Best for datasetsProximal, $200M (run rate, Sep 2026): the largest of the 7 here that sell datasets.
  • Best for RL environmentsToloka, $26.4M (FY2024): the largest of the 5 here that sell RL environments.
  • Best for annotationToloka, $26.4M (FY2024): the largest of the 4 here that sell annotation.

The ranking: 10 AI red teaming and security companies

Each company is ranked on the first of these it has on record: its revenue over twelve months (a run rate or a year as published, or its last four quarters added up; an AI segment where the rest of its business is not AI data), then its newest valuation, then the money it disclosed raising. No company pays for a place.

AI red teaming and security data companies, ranked by revenue, valuation and funding on record
#CompanyRanked onMeasureValuationSellsPays companies for data
1Proximal$200MRevenue, run rate, Sep 2026$300MEvaluations and benchmarks, Datasets
2Toloka$26.4MRevenue, FY2024RL environments, Datasets, Expert and human data, Evaluations and benchmarks, Annotation
3Vals AI$400MValuation, 2026Evaluations and benchmarks, Datasets
4Gray Swan AI$40MMoney raised, disclosedEvaluations and benchmarks, Datasets
5Sapien$15.5MMoney raised, disclosedEvaluations and benchmarks, Annotation
6Datacurve$15MMoney raised, disclosedRL environments, Datasets, Expert and human data, Evaluations and benchmarks, AnnotationIts own programme
7Idler$9MMoney raised, disclosedRL environments, Datasets, Evaluations and benchmarks, SimulationIts own programme
8Veris AI$8.5MMoney raised, disclosedEvaluations and benchmarks, RL environments
9Latch$5MMoney raised, disclosedEvaluations and benchmarks, Datasets, Annotation
10Quesma$2MMoney raised, disclosedRL environments, Evaluations and benchmarks
Twelve months.
A run rate or ARR is the company's own annualised figure, often its best month times twelve; a year is the year it published. Where a company reports quarters or halves, the last four quarters or two halves are added up. Gross revenue of a human data company includes what it pays its experts.
Who is on it.
A company is on this list when its own pages say it does this work, read word for word. It is not on it because we chose it, and none pays to be here.
Your place, on your site.
Every ranked company has a badge of its place, kept current with this page: the address is /badges/ai-red-teaming-companies/ followed by the company's id in its page's address, as in Proximal's. Link it to this page.
Pays companies for data.
"Through our introductions": it takes companies' data through AIData.Broker's introductions and pays us a referral fee when one goes ahead. That never moves its place: the order is the money on record. "Its own programme": a published programme for buying data. "Says it buys data": its own words, linked.

More answers

Which are the largest AI red teaming and security companies?
By the money on record: Proximal, $200M (run rate, Sep 2026); Toloka, $26.4M (FY2024); Vals AI, $400M valuation (2026). 2 of the 23 on the list have revenue on record; the others are ranked on their valuation or the money they raised.
Is Vals AI the biggest of the AI red teaming and security companies?
By valuation, yes: Vals AI was valued at $400M (Aug 2026). By revenue on record, Proximal is larger: $200M (run rate, Sep 2026), against no revenue on record for Vals AI.
Which AI red teaming and security companies buy data from companies?
3 of them take data from companies that hold it, on a published programme or in their own words: Datacurve, Idler and Collinear. A company holding such data can value it free and be introduced to every one whose rules take it.
How many AI red teaming and security companies are there?
23 companies on our register say on their own pages that they work on security. 10 of them have revenue, a valuation or funding on record, and are ranked; the rest are listed by name.
How is this list ranked?
Each company is ranked on the first of these it has on record: its revenue over twelve months (a run rate or a year as published, or its last four quarters added up; an AI segment where the rest of its business is not AI data), then its newest valuation, then the money it disclosed raising. No company pays for a place.

Cite this

AIData.Broker, “AI red teaming and security data companies in 2026, ranked by revenue and funding”, reviewed 8 Nov 2026. https://aidata.broker/best/ai-red-teaming-companies

Free to reuse under CC BY 4.0 if you credit AIData.Broker and link to this page. Each row names the publication that stated it, and words quoted from a source keep that source's own terms.

We ask once, before your first download, so we know who uses the data. The files stay free to reuse under CC BY 4.0. Downloading means you accept our terms and privacy policy.

Your is downloading.

These companies buy and build the data AI is trained on. See what yours is worth to them.