DataPacts

Universities and Research Groups Licensing Data to AI Companies

Checked

At a glance

Universities and research groups license data to AI developers mostly through access terms on datasets they build or host, and the public record holds membership fees and dataset terms rather than signed deal values.

Independent reference. Every buyer is covered on the same terms, every figure carries its source and check date, and no buyer pays to be listed or placed higher.

Request an independent introduction

Tell us what data your company holds. We reply with the buyers that fit and what their terms leave out. Your details reach a buyer only after you agree to a named introduction.

IndependentYour details stay with us until you agreeNot selling is a valid outcome

The Linguistic Data Consortium, hosted by the University of Pennsylvania, charges for-profit members USD 34,000 for Standard membership (checked November 9, 2026), and the Institutional Data Initiative at Harvard Law School Library requires contact for commercial use of Institutional Books 1.0, including AI model training (checked November 9, 2026).

Ownership inside the institution, funder and sponsor terms, ethics approval and records law decide what can be licensed at all.

Research data is one part of AI training data licensing, no university has published a direct AI license with a named buyer and a price, and deciding not to license is a valid outcome.

DataPacts Editorial Team checked every figure here against its primary source on November 9, 2026, and the buyer pages again on November 16, 2026; the page on how we source, date and check every figure sets out the rules.

What Universities and Research Groups Hold That AI Developers Ask For

A university holds six kinds of data AI developers ask for: research datasets, distributed corpora, digitized collections, lab records, student records and patient data, and each falls under a different rule. The table compares what is on record for each, as of 2026-11-16.

Holding Example on record Rule or term that decides licensing Source
Research datasets built by groups or consortia Ego4D: 3,670 hours, 923 participants, 74 locations, 9 countries (Ego4D consortium, 2022-02-17); Open X-Embodiment: 1M+ trajectories, "21 institutions" vs "34 labs" on the same page ownership across institutions; participant consent ego4d-data.org; robotics-transformer-x.github.io
Corpora distributed by a university-hosted consortium Switchboard-1 Release 2: approximately 260 hours, 543 speakers (LDC) consortium user agreements catalog.ldc.upenn.edu
Digitized library collections Institutional Books 1.0: 983,004 volumes, 242B tokens (2025-06-10) access terms; commercial use on request arxiv.org; huggingface.co
Lab and experiment records Mercor research-lab listing: lab notebooks the organization owns, 1,000+ runs, unpublished; "academic spin-outs" among industries (checked 2026-11-16) ownership; sponsor or partner terms api.data.mercor.com
Student records (US) education records 34 CFR 99.31(b)(1) ecfr.gov
Patient data at academic medical centers protected health information 45 CFR 164.502(a)(5)(ii), 164.514; EHDS from 2029-03-26 ecfr.gov; EUR-Lex

Universities are one of the types of who sells data to AI companies, and the holder is an institution with employees, funders and participants, each with a say.

Institutional Licensing Routes on Record

The routes on public record run through institutions' own terms: consortium membership, open release with commercial use on request, and buyer program listings; no university has published a direct AI license with a buyer and a price. The table compares the published routes, as of 2026-11-16, in alphabetical order by route.

Route Operator Terms as published Money terms published Source
Buyer program listing Mercor (buyer), "Biotech & Life Sciences Lab" records the organization owns; unpublished; not a fit: data owned by a sponsor or partner, data already published Mercor states USD 50,000 to 875,000, "depends on volume and depth of records" api.data.mercor.com, checked 2026-11-16
Consortium membership Linguistic Data Consortium (hosted by the University of Pennsylvania) for-profit members retain commercial rights to data acquired during membership; perpetual rights to data licensed in membership years; submissions portal for providers for-profit Standard USD 34,000, Subscription USD 40,000; not-for-profit Standard USD 2,400, Subscription USD 3,850 ldc.upenn.edu, checked 2026-11-09
Data provider terms with academic licensees Protege (provider terms) worldwide, royalty-free, non-exclusive license to Protege; access after termination 12 months for existing licensees, 3 years for academic licensees revenue share per use, percentage not public withprotege.ai, checked 2026-11-09
Direct license by a university none on public record n/a n/a n/a
Open release, commercial use on request Institutional Data Initiative (Harvard Law School Library) noncommercial early-access terms; commercial use including AI model training requires contacting the initiative not published huggingface.co, checked 2026-11-09

A University-Hosted Consortium: Membership Fees for Commercial Rights

The Linguistic Data Consortium, hosted by the University of Pennsylvania, licenses speech and text corpora to members and charges for-profit members more for commercial rights.

As of November 9, 2026, its join page lists for-profit Standard membership at USD 34,000 and Subscription at USD 40,000, against USD 2,400 and USD 3,850 for not-for-profit and US government members, without labeling the year. For-profit members retain commercial rights to data acquired during membership, and a submissions portal serves data providers.

The consortium's fee schedule and agreements are set out under Linguistic Data Consortium membership and license terms.

A Law School Library Initiative: Open Release, Commercial Use on Request

The Institutional Data Initiative at Harvard Law School Library released Institutional Books 1.0 openly, but commercial use including AI model training requires contacting the initiative.

The release holds 983,004 volumes of the 1,075,899 scanned, with 242B tokens counted by the o200k_base tokenizer (arXiv, 2025-06-10), and its Hugging Face dataset card applies noncommercial early-access terms (checked November 9, 2026).

The initiative publishes no price.

Universities Enforcing Dataset Terms and Defending Dataset Claims

A university that publishes a dataset under its own terms can enforce them, and one that builds a dataset from others' works can be sued over it. On September 28, 2026, the University of Denver filed a DMCA notice over the redistribution of the AffectNet dataset in 3 repositories on Hugging Face.

On the other side, EVOX Productions, LLC sued The Board of Trustees of the Leland Stanford Junior University in the Northern District of California, San Jose Division (5:24-cv-09521-NW, filed December 31, 2024) over copyright in vehicle images in a Stanford research dataset.

A stipulation on Stanford's first amended answer was filed October 13, 2026 (Dkt. 90), so the case is past the pleadings stage as of that date. The docket and its status are tracked under EVOX v. Stanford research dataset claims.

Who Owns Research Data Inside a University

Copyright in works employees create in the course of employment generally belongs to the employer unless an agreement says otherwise, so a university's own policy and its contracts decide who may sign. Three rules apply:

  • employee-created works: under 17 U.S.C. 201(b), the employer is the author and owns all rights unless a signed written instrument provides otherwise; in the UK, section 11(2) of the Copyright, Designs and Patents Act 1988 makes the employer first owner unless agreed otherwise
  • multi-institution datasets: data comes from several contributing institutions, as on the Open X-Embodiment page, so rights can sit with each of them
  • institutional policies: each institution's own policy and its agreements with staff, students and collaborators set who may sign, and no common rule is on public record

A rights audit before licensing starts with the employment and collaboration contracts behind the data.

Grant, Sponsor and Collaboration Terms That Can Block a License

Data produced under a sponsored project or collaboration may belong to the sponsor or partner, and a buyer listing excludes such data outright. Mercor's research-lab listing names "Data owned by a sponsor or partner" among its exclusions (checked November 16, 2026).

For publicly funded work, Directive (EU) 2019/1024, the Open Data Directive (in force July 16, 2019), governs whether public bodies, libraries, museums, archives and publicly funded research can charge for or exclusively license data, and its recital 55 excludes documents in which third parties hold intellectual property rights.

No funder's own data condition is stated here, so the grant agreement and the funder's policy are read first. Sponsor clauses and database rights in research records are covered under scientific data and papers as AI training data, and this page does not repeat them.

Personal data in university records is licensable only within the consent, the ethics approval and the records law it was collected under.

Buyers ask about consent and ethics review before they take research data with people in it. Mercor's lab listing asks "Is any of it patient-level or genomic data, and under what consent?" and lists "Genomic data collected with research consent" as nice to have (checked November 16, 2026).

Labelbox Prism states for medical data "IRB approval or qualified de-identification required" (checked November 16, 2026), and Gradient Health states "Our Ethics Board reviews research projects" (checked November 16, 2026).

Whether reuse fits the original research purpose is a question of purpose compatibility when reusing data for AI training. In the UK, the Data (Use and Access) Act 2025 section 67 counts research as scientific "whether publicly or privately funded and whether carried out as a commercial or non-commercial activity" (in force February 5, 2026).

Student Records at Universities

A US university can license education records without consent only after removing all personally identifiable information and making a reasonable determination that students cannot be identified. Under 34 CFR 99.31(b)(1), the determination covers identification "whether through single or multiple releases".

Patient Data at Academic Medical Centers

An academic medical center that is a HIPAA covered entity needs patient authorization to sell protected health information, and the research carve-out covers only a reasonable cost-based fee. Under 45 CFR 164.502(a)(5)(ii) the authorization must state any remuneration (164.508(a)(4)), and de-identified data falls outside the Privacy Rule (164.502(d)).

A limited data set under 164.514(e) is for research, public health or health care operations only, so licensing one for a fee is a sale of protected health information.

In the EU, Regulation (EU) 2025/327 sends secondary use through health data access bodies and never through a direct sale; Chapter IV applies from March 26, 2029. The rules are set out under HIPAA de-identification, sale of PHI and limited data sets in full.

Keeping Research Organization Status While Licensing to Companies

Research organizations in the EU may mine works for scientific research without a license, and German law withdraws that status from organizations controlled by a private company with privileged access to results. Directive (EU) 2019/790 Article 3 grants the exception with no opt-out, and Article 7(1) protects it from contract override.

Section 60d of the German Copyright Act (UrhG) implements Article 3 and excludes organizations controlled by a private company with privileged access to the results of the research. A university that takes on a commercial partner with privileged access therefore checks whether it keeps the status.

In Kneschke v. LAION, the Higher Regional Court of Hamburg (5 U 104/24) dismissed the appeal on December 10, 2025 and applied section 60d in addition to section 44b.

The Federal Court of Justice (I ZR 281/25) scheduled its pronouncement for December 17, 2026 (as of November 16, 2026); the Kneschke v. LAION rulings and status page follows it.

License Terms That Matter to a Research Institution

A research institution usually needs terms a company seller does not: continued research and publication rights, limits on what the licensee may do, and clarity about access after the license ends. Three terms come first:

  • a field of use that keeps the institution's research use and publication
  • publication and access after the term, which Protege's provider terms set at 12 months for existing licensees and 3 years for academic licensees (checked November 9, 2026)
  • what the buyer must disclose, since the training content summary under Regulation (EU) 2024/1689 Article 53(1)(d), applicable from August 2, 2025 for general-purpose AI models, shows licensed datasets only as Yes or No plus modality

The same terms grant Protege a worldwide, royalty-free, non-exclusive license (section 1.1). Drafting guidance is under field of use and scope limits in AI data licenses.

Licensing or Not: Benefits, Risks, Downsides and Alternatives for Universities

Licensing can fund data that already exists, but it can conflict with consent, sponsor terms and the institution's research status.

Benefits

  • a consortium shows a pricing model that charges commercial members more for commercial rights
  • a published buyer range is a program claim to test against the institution's records, not a paid amount
  • cleared and owned data can move from archive to use

Risks

  • a dataset built from others' works can draw a claim, as the EVOX docket shows
  • consent or ethics terms may not cover AI training, and buyers ask about both
  • commercial ties can cost a research organization its status under section 60d

Downsides

  • the training content summary shows licensed data only in coarse form
  • text and data mining by research organizations stays free under Article 3
  • no university AI license value is on public record, so there is no benchmark

Alternatives

  • consortium distribution through a submissions portal
  • open release with commercial use on request
  • noncommercial-only terms
  • not licensing at all

The general decision is set out under should you sell or license your data.

How Universities Differ From Academic Publishers and Other Research Holders

Universities differ from academic publishers in what they hold: publishers license journals and books under author contracts, while universities hold raw research data, records and collections under consent, funding and records rules. University presses and journal deals are covered under academic and scientific publishers licensing to AI. Related holders are:

The rules and terms on this page come from statutes, regulations, dockets and the organizations' own pages, each dated, and this page is general information and not legal advice. Further pages:

Frequently Asked Questions

Can a university license its research data to AI companies?

Yes, where the university owns the data and no sponsor, consent or records rule stands in the way; one buyer listing excludes data owned by a sponsor or partner.

Do universities need ethics approval to license participant data?

Buyers' own terms ask for it: Labelbox Prism requires IRB approval or qualified de-identification for medical data. Gradient Health states that its Ethics Board reviews research projects (both checked November 16, 2026).

Can a university license student data for AI training?

Only de-identified records without consent, under 34 CFR 99.31(b)(1).

How much do AI companies pay universities for data?

No university AI license value is on public record; the published figures are consortium fees and buyer program ranges. The Linguistic Data Consortium fees and the Mercor range above are fees and program claims, not payments. How buyers value data is explained under what company data is worth to AI companies.