Custodial Authorship and Orphaned Knowledge in the Age of Generative Artificial Intelligence

Policies on generative artificial intelligence in scholarly publishing and intellectual property law now share a settled judgment: a machine cannot be an author because it cannot assume responsibility, hold rights, disclose conflicts, or answer for the consequences of a claim. This article accepts that judgment, but argues that its usual justification remains too narrow. Much of the debate treats authorship as a matter of origination, attribution, and reward. Drawing on Foucault, Merton, Shapin, Hardwig, and scholarship on scientific trust, I argue that modern authorship has long joined three functions that are easier to see now that generative AI has pulled them apart: origination, reward, and custodianship. AI can produce text and, in some cases, results. What it cannot supply is the continuing human commitment to defend, develop, teach, correct, and preserve those results. I call the resulting condition orphaned knowledge: claims that may be true or useful, yet have no identifiable custodian responsible for their future life in the scholarly record. The article makes two claims. First, human knowers organize both the credit economy of journals and the property economy of copyright, leaving machine-originated knowledge in a double vacuum. Second, authorship should be understood less as a retrospective claim to credit than as a prospective undertaking to steward a finding. A machine may originate an output. Only a human being, group, or institution can assume custody of knowledge. The question raised by machine discovery is therefore not only who made the claim, or whether it is true, but who will keep it answerable over time.