Who Owns What the Machine Made: Copyright Authorship in the Age of Generative AI
By Casey Scott McKay ·
This article explains who owns the output of a generative AI system under United States copyright law, and why the answer is so often that nobody does. It traces the human authorship requirement from the 1884 Burrow-Giles decision on photographs — where the argument against copyright was the same one now aimed at a Midjourney image — through Bleistein, Feist, Naruto, and the district court and appellate rulings in the Thaler litigation. It works through the Copyright Office's March 2023 registration guidance, the Zarya of the Dawn correspondence and the selection-and-arrangement carve-out it created, the Copyright Review Board's refusals in Theatre D'opera Spatial and SURYAST, and the Office's 2025 reports on digital replicas, copyrightability, and generative AI training. It then covers what practitioners actually get wrong: the disclosure duty on the application form and its consequences under 17 U.S.C. § 411(b), the gap between owning a work and being able to stop anyone, the difference between the input question of training-data infringement and fair use and the output question of substantial similarity and memorization, and the real content of model vendor indemnities. A final part maps the genuinely unsettled ground — how much human contribution is enough, whether a fine-tuned model is a derivative work, and whether copyright management information claims under § 1202 survive standing and identicality objections — and points to the companion guide and checklist for the step-by-step workflow.
IP and Technology > Copyright | Article | Published 23 December 2025 - Updated 2 July 2026 | Casey Scott McKay - marksy.us
Summary. Ask a founder who owns the image their design tool just generated and they will say "we do — we paid for the subscription." That answer is wrong in a specific and expensive way. This article explains the human authorship requirement that decides the question, from the 1884 Supreme Court case about whether a photograph could have an author at all, through the Copyright Office's March 2023 registration guidance and its 2025 reports, to the D.C. Circuit's 2025 affirmance in Thaler v. Perlmutter. It covers what survives when the machine did most of the work — the selection-and-arrangement copyright the Office recognized in Zarya of the Dawn — why prompt engineering has so far failed to qualify as authorship, what you must disclose on the application form and what happens under 17 U.S.C. § 411(b) if you do not, and how the input question (training-data infringement) differs from the output question (substantial similarity and memorization). It closes on the genuinely unsettled parts: derivative works, model weights, copyright management information under 17 U.S.C. § 1202, and the ownership gap no contract fully fills. It explains the law; the procedure lives in the companion guide and checklist.
Keywords: ai-generated works · human authorship requirement · thaler v. perlmutter · burrow-giles v. sarony · copyright office ai guidance · zarya of the dawn · selection and arrangement · prompt engineering · 17 u.s.c. 102 · copyright registration disclosure · limitation of claim · generative ai training data · substantial similarity · memorization · 17 u.s.c. 1202 · copyright management information · model vendor indemnity · digital replicas · derivative works · thin copyright
This is premium Marksy content — the full document is available to subscribers.