What Publishers Reserve for Humans: Reading Generative AI Policies as a Map of Delegable Work

Publisher policies on generative AI are commonly compared in terms of permission, disclosure, and the prohibition of AI authorship. This article reverses that perspective and asks what publishers implicitly consider delegable to machines and what they reserve for humans. It analyzes the official policies of Elsevier, Springer Nature, Wiley, Taylor & Francis, Sage, and IEEE available on 14 July 2026 through a three-layer model: the Disclosure-Exempt Acceleration Envelope, the Disclosed Acceleration Envelope, and the Human-Reserved Core. Although publishers differ substantially in disclosure exemptions and permitted workflows, they converge on reserving verification, evidential judgment, interpretation, final approval, responsibility for correction, and legal-ethical answerability for human authors. Authorship is therefore being relocated from manual textual production toward epistemic warranty and answerability. Yet policies rarely operationalize how to distinguish a nominally responsible person from a substantively competent guarantor. The article identifies this missing criterion as a central governance gap in AI-assisted scholarly publishing.