Paper Permission: What Publisher AI Policies Allow and Practice Withholds
Publisher policies on generative AI are usually read as restrictions. Read instead as a map of delegable work, they reserve for the human author a narrow core: the integrity of primary research images, and the warranty of the claims, by which we mean the cluster of human functions those policies keep together, including verification, interpretation, evidential judgment, final approval, responsibility for correction, and legal and ethical answerability. Almost every other task becomes conditionally delegable, provided the author keeps that core and discloses help that is more than cosmetic. Yet in practice this permission is seldom used in the open. Disclosure appears in a small share of submissions while self-reported use runs far higher. We call the gap Paper Permission: a permission that is present in the text of policy and largely absentfrom open practice. We propose that one mechanism helping to explain its underuse is that disclosure can function as a confession signal, so that the authors who comply, and who are least protected, carry the cost, while the judging side uses AI under fragmented rules that are hard to observe. We propose closing the gap by making standards explicit rather than by adding conduct duties to authors or reviewers: name the core, reduce the grey zone around it, and make the standard of judgment visible.