Academic Publisher AI-Training Data Deals Face Library Consortium Pushback

The deals that broke the silence

In August 2024, trade press reports revealed that Taylor & Francis, the academic publishing arm of Informa, had signed a three-year agreement to make its books and journals available to Microsoft as training data for artificial intelligence models. The deal was worth about $10 million. Researchers learned about the arrangement from book-trade coverage, not from a direct publisher notice at the time, and the absence of an author opt-out became the immediate complaint.

Wiley has been clearer about the money. The publisher disclosed $23 million in AI licensing revenue for fiscal year 2024 and told investors it expected roughly $44 million in fiscal 2025. The two figures do not define an entire market, but they establish the base rate: leading scholarly publishers are treating licensing for AI training as a recurring revenue line rather than a one-off experiment.

For a research lab, these are not distant contracts. The work an author submitted to a journal, often after years of public funding and peer review, is the same content now being bundled into a training corpus. The copyright forms many researchers sign at acceptance were written for print and electronic distribution, not for model training, and that gap is now the center of the library fight. The Taylor & Francis agreement was visible in investor reporting long before most faculty saw an opt-out question, and that sequencing is part of the objection.

Why libraries object before they see the invoice

The library objection starts with the invoice. A university can pay hundreds of thousands of dollars a year in subscription and read-and-publish fees. If the publisher then earns separate revenue from training an AI system on that same content, the library asks whether its subscription is effectively subsidising a transaction the university cannot see.

Libraries raise a second problem about authorship. In most cases the researcher created the intellectual content. The publisher may hold a license or a transferred copyright, but the author's control over new derivative uses was not negotiated in an AI market. Consortium heads say the absence of consent is a structural defect, not a paperwork oversight.

There is also a collection-development issue. If a publisher is paid twice for the same corpus, the library has reason to ask what value the subscription itself is still providing. That question moves faster in public universities, where procurement rules require a demonstrable institutional benefit for each large contract. If the answer is access plus an undisclosed AI corpus sale, that answer is difficult to defend in a budget hearing.

Photo by Robert Bye on Unsplash

The mechanics of a licensing fight

Follow the paper trail. When an author signs a copyright transfer or exclusive license, the publisher typically receives the right to reproduce, distribute, put into databases, and sometimes sublicense the work. Older contracts rarely mention machine learning, text and data mining, or model training. Publishers have argued that broad rights clauses cover these new uses. Library consortia argue the opposite: the clauses were never disclosed as AI permissions and should not be read that way now.

The Bookseller report on the Taylor & Francis Microsoft agreement showed the practical gap for faculty. Authors had no clear mechanism to check whether a specific article was included, and no visible royalty or notice. That gap is what university librarians mean when they say the deals lack transparency.

Library consortia have started drafting language to stop AI rights from being folded into routine renewals. The International Coalition of Library Consortia has pressed publishers to treat AI training rights separately, with authors informed and consent obtained before inclusion. US research libraries and UK consortia, along with Australian library groups in the CAUL network, have raised the same points with their members.

This is not a single-publisher problem. The same institutions that spent years negotiating open-access terms have been forced onto a second front, one that runs alongside the budget pressures tracked in AcademicJobs' earlier reporting on open-access mandates and library budgets.

What this means for your lab

Base rate first: most researchers will not be asked. An author may learn about a publisher's AI agreement months after it closes, or never. The exception is a publisher that sends an opt-out link or direct notice, and that exception usually arrives with a short deadline. A colleague at a midwestern research university, call him Dr. O, received an email from his publisher in early 2025 with a link to an opt-out portal. He had four days. His lab manager found the message on a Friday afternoon, and the university's scholarly communications office had no procedure for checking the license terms quickly.

What this means for your lab: treat AI training rights as part of every new submission, not a future policy question. Before signing a copyright form, check whether the publisher separates AI uses from standard publication rights. If it does not, ask the journal editor for the clause and keep the exchange where your institution can find it. The fix is not a policy document, it is a habit: one shared folder for publisher AI clauses, one README per journal, enforced from submission day.

Practical steps for a lab in the next submission cycle:

• Keep the publisher's AI clause with the submitted manuscript and the final accepted version.
• Record whether the journal editor offered a separate opt-out and when.
• Send a copy to your library's scholarly communications office before the deadline.
• Raise the AI-use question in a lab meeting now, not when a publisher email arrives.

For principal investigators, the concrete step is to ask the library what the university's consortium says about AI clauses. Libraries do not want individual researchers to negotiate alone, but they do want the researcher's contract history to strengthen their negotiating position.

Photo by Priscilla Du Preez 🇨🇦 on Unsplash

The negotiation season has already started

The consortium response is moving from statements to contract language. In recent renewal cycles, libraries have asked publishers to confirm which portions of a subscription are used for AI training, whether authors can exclude their work by a visible deadline, how AI revenue is treated in pricing discussions, and what audit rights the library retains. Publishers have not answered uniformly. Some have offered opt-out mechanisms; others have treated the questions as outside the subscription negotiation.

The compute side matters too. A university that pays separately for AI research infrastructure is now being asked to account for the rights to train models on campus, while publishers monetise the corpus that faculty produced. That tension is laid out in AcademicJobs' earlier piece on AI compute access and research infrastructure.

If publisher deals continue without author-level tracking, the next generation of academic contracts will carry a new obligation: researchers will have to read AI clauses the way they now read embargo periods and author rights. The base rate is stacked against the individual author. The exception is the author whose university has already placed the AI clause on the consortium agenda before a renewal deadline.

Librarians are not asking researchers to stop publishing. They are asking them to stop signing forms that assign AI rights without a separate look. Ask your library consortium what its next meeting date is and put publisher AI terms on the agenda. That is the one step that converts a quiet licensing decision into an institutional position.