The ACM digital library fight reveals a deeper rift over AI training data

The Association for Computing Machinery, the world’s largest computing society, is in the middle of a fight over whether to license its Digital Library to large language model companies. The proposal, published as an opinion piece titled “[Now is the time to give LLMs access to the ACM digital library](https://cacm.acm.org/opinion/now-is-the-time-to-give-llms-access-to-the-acm-digital-library/),” argues that the benefits of opening the archive — which holds decades of peer-reviewed computer science research — outweigh the risks. But the piece was not presented as a settled position. It was presented as “ACM Believes,” a framing that has triggered a revolt among the society’s own members.

Jeanna Matthews, a long-time ACM volunteer leader, posted on LinkedIn that “ACM Council has not even had a chance to vote on the question.” She said she was “deeply saddened to see this presented as ‘ACM Believes’ it is time to do this. Many people within ACM do not agree and no one should be saying ‘ACM Believes’ without even a Council discussion or vote.” The backlash exposes a deeper tension: who owns the research that trains the most powerful AI systems, and who gets to decide how it is used.

The opinion piece was authored by several senior volunteer and staff leaders, including the co-chairs of the ACM Publications Board and the ACM Digital Library Board. Scott Delman, ACM’s Director of Publications, clarified in a LinkedIn response that “ACM has not made any decisions regarding the possible licensing of ACM content to the LLMs (either for training or inference).” He said the intention was to “share the opinions of a number of Volunteer and Staff leaders” and to “request feedback from the community to inform future decision making.” Delman reported receiving over 330 comments, ranging from very positive to very negative.

The backlash is not about the technical merits of giving LLMs access to research. It is about process and consent. The ACM Digital Library contains papers submitted by researchers under specific expectations. Those expectations did not include having their work licensed to AI companies for training or inference without explicit permission. James Davis, a researcher who has reviewed hundreds of papers for ACM without compensation, wrote in the comment form: “I strongly oppose any proposal that would allow ACM to license authors’ scholarly work to AI companies without explicit, paper-specific consent and appropriate compensation.” He proposed that any such policy apply only prospectively, to papers published after 2026, and that ACM disclose the arrangement to authors before submission.

The conflict mirrors a broader pattern in AI training data. Companies like OpenAI, Google, and Meta have trained models on large web crawls that include copyrighted material, academic papers, and other content scraped without permission. The legal framework for this practice is unsettled. Several lawsuits are pending, including cases brought by The New York Times, Getty Images, and individual authors. The ACM debate is different because it involves a membership organization that explicitly controls access to its archive. The question is not whether the content is available on the open web — much of it is behind a paywall or requires institutional access. The question is whether ACM, as a gatekeeper, should actively license that content to AI companies.

Matthews pointed out that a Presidential Taskforce on Responsible Use of AI in ACM has been working since January on a governance framework. That draft framework, she said, “calls for ACM Council to make a decision regarding LLM Access to the Digital Library after community consultation.” She noted that the taskforce has discussed the draft twice with ACM Council and is waiting for more internal discussion before wider public consultation. “I wish this article had followed the same path,” she said.

The community comments reveal a range of concerns. Ian Soboroff, another researcher, wrote: “Conferences and journals are being hammered with generated papers. A dozen or more from the same author at the same venue, and actual scientists need to spend effort to review that crap. ACM: you have a mission. Step up and serve your members, the ones who make your DL possible, valuable, meaningful. Not the AI companies.” Patrick Madden expressed frustration with “AI summaries” already attached to papers in the Digital Library, calling for a more aggressive stance.

The push for LLM access is not without merit. Opening the archive could accelerate research, enable better literature reviews, and help models generate more accurate scientific summaries. The opinion piece argues that the benefits outweigh the risks. But the framing as a settled belief, without a vote or broad consultation, has damaged trust. Matthews observed: “I have seen people saying ‘it probably doesn’t matter what we say.’ It will be hard for people to trust that volunteer leadership has carefully considered their comments when it was framed in a one-sided way to start.”

The ACM is not the first scholarly publisher to face this question. Elsevier, Springer Nature, and others have signed licensing deals with AI companies. Those deals were often negotiated quietly, and researchers only learned about them after the fact. The ACM has a chance to do something different — to build a governance process that actually consults the people who create the content. Whether it will take that chance is an open question.

Delman said all comments will be shared with volunteer and staff leadership and taken into account before any decision. The form for submitting feedback, which was temporarily inaccessible, is now open. The decision has not been made. But the process by which it is made will set a precedent for how scholarly societies handle AI training data going forward. The community is watching, and many of them are not happy with what they see so far.