Why UC’s Academic Senate Library Committee, Creative Commons, and International Experts Stand Behind Open Access with CC BY in the Age of AI

By Günter Waibel / November 3, 2026

Amid growing concerns over AI systems mining scholarly work without attribution, many researchers are asking whether there is a way to definitively withhold their publications from AI training. Some are considering the use of restrictive Creative Commons reuse licenses, like CC BY-NC and CC BY-NC-ND, while others question whether to publish open access at all.

In a recent letter, UC’s Academic Senate Committee on Library and Scholarly Communication (UCOLASC) considered these concerns within the context of US copyright law, publisher contract practices, and author rights. The committee concluded that retreating from open access or utilizing restricted licensing models is counterproductive, asserting that “open access with a CC-BY license continues to be the publishing option most aligned with author and university self-interest and the academic research mission.”

Central to UCOLASC’s conclusion is the recognition that there is “no known technological or licensing pathway to withhold a published article from AI training once such work enters the public or commercial stream.” The letter explains why both the current state of the law and the actions of publishers render publishing behind a paywall or with a restrictive license ineffective strategies for keeping publications out of AI models – and harmful to the university’s own nonprofit research enterprise.

Factor 1: AI training, Copyright and Fair Use

Commercial AI training has been found to be “fair use” in at least two prominent court cases (Bartz v. Anthropic and Kadrey v. Meta; the links point to detailed summaries of each case by the Authors Alliance). Over 100 AI copyright lawsuits have been filed in the US, and as of this writing none of them have found training generative AI to be an infringement of authors’ or publishers’ rights. Unless subsequent cases come to the opposite conclusion, AI companies do not need permission to conduct training; all they need is access to the content.

For paywalled content, AI companies can buy that access from publishers (see Factor 2 below). For open access content already publicly available with a Creative Commons license, that access pre-exists by definition. As UCOLASC’s letter underscores, “all Creative Commons licenses (including the non-commercial licenses) operate within a framework of copyright that includes fair use. That means that if fair use is applicable, the license itself is irrelevant to begin with.” Even a Creative Commons license with a non-commercial restriction does not prevent AI training by a commercial entity if that training falls under fair use.

Factor 2: Commercialization of academic content

As AI companies seek more and more authoritative content to train their models, publishers see new opportunities to create additional revenue, in particular from the academic content they exclusively control. As a result, publishers’ incentives to establish that exclusive control have further increased.

Under subscription models, publishers gain exclusive commercial control through copyright transfers. This enables, in UCOLASC’s analysis, “a lucrative marketplace where publishers package and resell faculty content to private industry for AI training. In this scenario, the commercial publisher becomes the financial beneficiary and sole gatekeeper for the use of a critical new technology and research methodology.”

In summary: Authors cannot, within this framework, employ paywalls or reuse licenses to restrict AI companies from training their LLMs on their scholarly publications.

The academic collateral damage of the publisher quest for commercial control

This exclusive commercial control doesn’t just fuel publisher deals with private AI companies; it also means that universities themselves have to be vigilant about retaining Text and Data Mining and Artificial Intelligence (TDM/AI) rights in their subscription agreements with publishers. Libraries in the US, including at UC, have increasingly struggled to protect academic fair use for a broad range of computational research methodologies from being contractually blocked in their publisher negotiations. The cascade of letters within the Academic Senate culminating in an endorsement of TDM/AI by the UC President and Provost testify to this ongoing conflict between the academy and some publishers.

And the quest for commercial control does not stop at the subscription model: in open access publishing with restrictive licenses (such as CC BY-NC or CC BY-NC-ND), for-profit publishers deftly effect the author’s surrender of commercial rights. While authors choosing this type of license typically intend to reserve those rights for themselves, the publisher’s standard contract (which authors are required to sign) forces the surrender of the reserved commercial rights to the publisher. (Here is a presentation with more details.) This allows the publisher to sell these articles for commercial uses: for instance, to pharmaceutical companies wanting to utilize scholarly articles to promote their drugs. Perhaps publishers may even be able to sell them to AI companies seeking a risk mitigation strategy against future court cases adjudicating the reach and limit of fair use in commercial settings.

UCOLASC continues to recommend Open Access and CC BY

The subscription system does not prevent AI training on academic content. And because of copyright law (assuming courts continue to find AI training to be fair use) and publisher contract practices, restrictive licenses like CC BY-NC and CC BY-NC-ND fail to block AI training as well.

Beyond its analysis of the limitations of these efforts to restrict AI access to published content, though, UCOLASC’s letter also underscores the far reaching and fundamental self-interests protected for the University of California and the general public by choosing to publish open access, including:

Academic sovereignty: Authors retain ownership in their intellectual property rather than transferring exclusive rights, with open access serving as “the only known pathway for researchers to retain autonomous decision-making and agency […] to perform computational research on the scholarly literature, which is quickly emerging as a vital research methodology.”
Counterweight to AI misinformation: Open access ensures university research “remains a public good, contributing high-quality, factual information to the public sphere as a vital counterweight to AI misinformation.”
Return on public investment: Rather than locking publicly funded knowledge behind publisher paywalls, open access “provides taxpayers, who directly support faculty through FTEs and invest in research through extramural grants, with free access to scholarly work.”
Financial stewardship: Open access guarantees that “UC will never have to pay commercial vendors more than once to access the scholarly literature or to allow faculty to use such literature for research,” while allowing researchers globally to access UC scholarship without paywalls and fees.

While UCOLASC outlines these broad benefits of open access, the specific license choice determines how those benefits enable day-to-day scholarly practice. Because fair use applies to AI training regardless of license terms, attempting to stop AI through restrictive licenses fails. Instead, it erects barriers for desirable academic reuse. Shifting the focus from AI mitigation back to human scholarly collaboration makes the case for CC BY clear. As such, UCOLASC’s recommendation strongly tracks the assessment of Creative Commons itself, which continues to recommend CC BY (articles) and CC0 (datasets) for open science. As CC clarified in its August 2026 whitepaper: “Given that copyright exceptions [fair use in the US context] often mean CC license restrictions don’t apply to AI use in the first place, using a more restrictive license isn’t effective in preventing it. In fact, a more restrictive license may prevent the exact types of human sharing and reuse licensors are hoping to allow, while at the same time putting no restrictions against AI use.”

In an academic context, some of the desired academic human reuses enabled by a CC BY license are:

Sharing on course, conference, lab, news, and other websites;
Reproduction in print coursepacks, even when they are sold by commercial copy shops;
Translation into other languages, and sharing by their translators;
And use and adaptation by peers and students to build on knowledge in ways that their original creators never imagined.

UCOLASC’s call for UC action

In light of this constellation of issues, UCOLASC has renewed its call on UC’s negotiation team (of which I am a member) “to enter into agreements with publishers whereby open access publishing and Creative Commons Attribution licenses (CC BY) are default for all UC authors, and subscription publishing is deprecated as an option-of-last-resort.”

And UC is not alone in taking this position. In the final statement from the 18th Berlin Open Access conference, negotiators from forty countries reinforce the same message: “CC BY must be the default license offered to authors, and the license-to-publish agreements they sign for CC-licensed articles must not require them to assign exclusive rights to the publisher. Publishers must state plainly, at the point of license choice, the combined effect of the license and the license-to-publish agreement, including any rights retained to sell or license content to third parties, such as for AI training.”