How universities and research institutions can license data to AI companies

Universities produce some of the best documented data there is. A research group that built a speech corpus, an annotated text collection or a benchmark usually holds collection methods, consent records and a readme, which is the provenance AI buyers check first.

The catch is that a university rarely holds every right on its own. Students, research participants, funders, publishers and faculty each have a claim on parts of the archive, and the federal rules that protect them are specific.

What university and research data do AI companies buy?

AI companies buy university and research data that is expert-made, well documented and cleared for commercial use. The most marketable groups are language corpora, scientific data with results, and teaching materials with expected answers.

Assessment material is close to what buyers already name. An evaluation dataset is a set of tasks with expected answers and scoring rules, which is what a retired exam with its marking scheme already is. Corpora fit buyers of multilingual text data and speech and audio data.

Who buys data from universities?

The Linguistic Data Consortium is the most established outlet for university language data. LDC, hosted by the University of Pennsylvania, publishes contributed speech and text corpora in its catalog, asks providers for ownership and copyright information, a sample and documentation, and does not require exclusive distribution rights.

LDC's data sharing guide does not describe payments to providers, so ask before you submit. No buyer whose terms are public publishes a university price range, so any figure quoted for a research archive is a negotiation. A free data valuation tests what yours could fetch.

Can a university license student data under FERPA?

A university can license student data only with written consent or after all personally identifiable information is removed. Under 34 CFR 99.31(b)(1), an institution may release records without consent once it has made a reasonable determination that a student's identity is not personally identifiable, whether through single or multiple releases, taking into account other reasonably available information.

FERPA's studies exception does not fit a commercial license. Section 99.31(a)(6) allows disclosure to organizations conducting studies for, or on behalf of, the institution, for purposes such as improving instruction, under a written agreement that requires the data to be destroyed when no longer needed. A perpetual training license to an AI developer is a different arrangement.

Coursework and discussion posts raise copyright as well as privacy. Under 17 U.S.C. section 201(a), copyright vests in the author, and students are not usually writing as employees, so their essays and code are theirs unless they assigned them.

IRB consent forms allow reuse for AI training only if what participants were told covers it. Under the Common Rule, at 45 CFR 46.116(b)(9), consent for research with identifiable information must include one of two statements: that identifiers might be removed and the data used for future research without additional consent, or that the data will not be used or distributed for future research even with identifiers removed.

Where a form used the second statement, that study is out. Where it used the first, de-identified data can be shared for future research, but a commercial license for model training may not count as research, so read the protocol and ask your IRB. Broad consent under section 46.116(d) must describe the types of research that might be conducted and the types of institutions or researchers that might use the data.

Who owns data produced under a federal grant?

Data produced under a federal grant usually sits with the recipient institution, subject to a government license. Under 2 CFR 200.315(b), a recipient may copyright work developed under a federal award, and the agency reserves a royalty-free, nonexclusive and irrevocable right to reproduce, publish or otherwise use the work for federal purposes.

Section 200.315(d) also gives the government the right to obtain, reproduce, publish or otherwise use the data produced under a federal award, and to authorize others to use it for federal purposes. A commercial license can sit alongside those rights, but you cannot give a buyer exclusivity that overrides them. The agency can also require works to be made available through a designated public access repository.

Industry-sponsored research agreements set the sponsor's rights in results, so read those as well. Teaching materials and lecture recordings are a policy question: under 17 U.S.C. section 201(b) an employee's work within the scope of employment is a work made for hire, but institutional policies can assign scholarly and teaching works to faculty.

If you hold UK or EU data

UK universities apply UK GDPR. Article 8A, inserted by the Data (Use and Access) Act 2025 and fully in force from February 5, 2026, treats further processing for scientific research under Article 84B as compatible with the original purpose.

Where data was collected on the basis of consent, Article 8A(4) narrows the routes to a new purpose. Whether a commercial training license counts as scientific research is a question for your data protection officer and counsel, so we recommend relying on fresh consent or anonymized data for UK participant data.

How do you sell university data through Sell My Data To AI?

You start with a consultation about the datasets, their consent forms and their funding. Sellers pay nothing to be assessed, and we charge a success fee only when a deal completes, agreed in writing before we introduce you to any buyer.

Every dataset passes our eight-point vetting standard before a buyer sees it, including ownership, contract restrictions and the privacy notice in force at collection. We are a broker, not a law firm, so involve your general counsel, IRB and technology transfer office before you sign. Book a consultation to start.