How this registry works

The goal is a source of truth on AI training data and RL environment vendors that buyers, founders and AI assistants can check for themselves.

Who is listed

A company is listed when it has a live website and public evidence that it sells RL environments, training or human data, evals and benchmarks, simulation, or data annotation to AI developers. Providers that offer only RL infrastructure are outside this core list. A vendor with both infrastructure and an eligible offering stays listed. Offering details we can't confirm are marked as unknown rather than treated as infrastructure-only. Listing is free. We find companies through public directories (we started from Pavlov's List and RL List), launch announcements and our own monitoring. Companies that stop meeting the criteria are marked or removed. Acquired companies stay listed with their acquirer noted.

Where facts come from

Ordinary company facts, such as what a company sells, where it's based, when it was founded, its size and its funding, come from credible public sources: the company's own site, documentation and job posts, its Y Combinator or LinkedIn company page, reputable press, company registries, and funding databases when nothing better exists. Each fact links to its source and shows the date it was checked. Descriptions are written by the registry from those sources.

Claims a reader may act on are held to a stricter standard. Security and privacy details, where a vendor's data comes from, named customers, and everything on Who's buying (what a program pays and who qualifies) are published only from the company's or program's own published words, with the quote one tap away.

When we can't find a value it reads Not publicly listed, which is not a claim that the company lacks it. When sources don't settle a value it reads Not confirmed. A fact the company confirmed directly is marked Vendor-confirmed.

The trust & data layer

For each vendor we record how it protects a buyer's confidential material (SOC 2, trust center, stated practices), where its environments and data come from (built from a client's own data, licensed real-world data, expert-created, synthetic or product replicas, each backed by a short quote from the vendor's own site), whether work can be exclusive, and whether it can run in a customer's cloud or on-prem. These are the questions procurement and security teams tend to ask first, and they are rarely collected in one place.

SOC 2 Type I/II entries are vendor statements, not independently verified audits. “Unspecified” means the vendor claims SOC 2 without stating a type; “Pending” means work is in progress. “Not publicly listed” means we found no public statement. None of these counts as a completed Type I/II report.

Domains describe task areas such as coding, finance or computer use, not a company's website domain. “Multi-domain” means work across several areas; it does not imply coverage of every area. A vendor appears in a specific domain filter only when that area has supporting evidence.

What we don't do

• Rank vendors or score their quality. Public samples and benchmarks are linked so you can judge them yourself.
• Accept payment for listing, placement or edits.
• Publish personal contact details.

Corrections

Anyone can add a missing company or fix a listing; we ask whether you work at the company. A person reviews every submission and checks it before anything changes. Today we publish a change when it is supported by a public source, such as the company's website, documentation or trust center. If the information isn't public yet, we follow up. Past values stay in the public fact history.

Disclosure

Who Sells Data is maintained by Secludy AI, which makes sensitive data safe to use for AI. Some listed companies may be Secludy customers or prospects. That relationship never affects whether a company is listed, the order it appears in, or what its page says. The maintainer note above the register and on vendor pages is the only Secludy promotional content on the site. Names and emails sent through the form are used only to follow up on that submission. They are never published or added to a mailing list.

Using the data

The full dataset is free to download as CSV or JSON. Reuse terms are not yet approved. Third-party material, including quoted passages and linked samples, retains its own terms. AI assistants and tools can read /llms.txt and /data/directory.json directly.