ACM to open Digital Library to LLMs, invites public feedback

For years the story with big scholarly publishers and generative AI has been the same one, cautious silence, quiet lawsuits, no clear public stance on training. ACM has now written its stance down. In an opinion piece in Communications of the ACM, Scott Delman, ACM's Director of Publications, argues that the time has come to give large language models access to the ACM Digital Library.

The framing is deliberate. Delman writes that the strategic question is no longer simply whether to permit LLMs to train on ACM Publications, but how to ensure that ACM authors, their scholarship, and the version of record remain visible, attributable, and influential within an increasingly AI-mediated knowledge ecosystem. In other words, ACM has decided that being outside the training corpus is worse for its authors than being inside it under negotiated terms.

This lands on top of a licensing change already in motion. All ACM articles published after January 1, 2026 will be published open access, and authors are required to select either a CC-BY or CC-BY-NC-ND license. CC-BY lets third parties, including LLM trainers, reuse the work commercially; CC-BY-NC-ND blocks commercial reuse and prevents derivatives without explicit written permission. That license choice, which used to be a technicality, is effectively the author's opt-in or opt-out for AI training.

The honest caveat is what the piece does not settle. Delman flags attribution, hallucinations, governance, commercial concentration, and intellectual property as real concerns, and ACM is seeking broad community feedback via a public form before entering into any formal licensing agreements. There is no dollar figure in the piece, no named counterparty, no mechanism spelled out for how attribution will actually be enforced inside a model's output. Not everyone in the community is convinced by the framing either; on LinkedIn, Anthony Bucci pushed back that describing AI as a primary interface through which knowledge is discovered and used treats a choice ACM itself is making as if it were an inevitability.

For working researchers, the practical question over the next few months is small and specific, which of the two Creative Commons licenses to pick on the next paper. For labs and companies training on the open scientific record, the more interesting shift is that one of computing's largest archives is publicly signalling it is open to being licensed, rather than only defended.