Patronus AI review: what it licenses and how it pays

Patronus AI publishes no minimum team size, years of records or region, and its intake form takes teams from 10 or fewer people up to more than 1,000, and companies that are still operating, winding down, closed or acquired. In our view, that makes it worth asking if your company is small or already closed.⁷

What it pays for is depth: for a codebase it looks at the code, its development history, tests and supporting material, and for company records it looks at the volume, how many years they cover and how the systems connect to each other.⁸

Patronus AI licenses private codebases used in production with their development history, company records such as emails, documents, orders, invoices and tickets, and the code, original assets and history behind released games. It uses them to build AI training and evaluation tasks and environments, says its customers may use that work, and says it does not resell your code or records themselves.^1,4

If you have a codebase on GitHub, you can let it measure the repositories you pick through a read-only connection, which it says does not run your code and deletes the temporary copy afterward. You don't need to export anything or give access to get a first answer.^9,3

After a first answer within 2 business days, Patronus AI reviews a sample under a mutual NDA and makes a written offer with the price, what's included and how it will check the full delivery. If you accept, you share the full dataset, it has up to 30 days to check it against what you agreed, and it pays you directly once it accepts, without waiting for another buyer. It gives no number of days from acceptance to payment.^3,10

Its sample terms say two named reviewers see the sample in an isolated location, it is used only to decide on a license, and if Patronus AI says no it deletes every copy and confirms that in writing. You confirm you have the right to share before the sample goes over, and proof such as an IP assignment is needed only when the license is drawn up.^10,11

Patronus AI does it itself, after it accepts the full dataset and before any use: each person gets the same pseudonym in every file and system, and you get a report of what was replaced. That means the full delivery goes over with names in it, so ask whether you can strip them out first if your records hold client or patient details.⁵

It says it follows GDPR and HIPAA rules, and that it agrees how health information or EU and UK personal data will be handled before any sample is shared.¹²

Patronus AI publishes no dollar range, and the estimate its form shows is, in its words, not an offer. Accepted data is deleted on a schedule set in each license, and the published terms give no period, so ask how long it keeps your data before you sign.^2,6

Patronus AI, Inc. was founded in 2023 by CEO Anand Kannappan and CTO Rebecca Qian, whom SiliconANGLE describes as former Meta AI researchers. It raised a $50M Series B in June 2026 led by Greenfield Partners, for $70M in total, and its main business is testing where AI models fail and building environments to improve them.^13,14,15