Who owns the output?

Lesson 4 of 5 in AI and Intellectual Property: Training Data, Outputs, and the Litigation Wave.

Flip the question: forget what went into the model — who owns what comes out?

US law has answered with unusual clarity. In Thaler v. Perlmutter, Stephen Thaler sought registration for an image he said his "Creativity Machine" generated autonomously, listing the machine as author. The Copyright Office refused; the D.C. district court agreed; and in March 2025 the D.C. Circuit affirmed: the Copyright Act requires a human author. An autonomously generated work enters the public domain at birth. Thaler ran the same experiment in patent law with his DABUS system, and lost identically — Thaler v. Vidal (Fed. Cir. 2022) holds inventors must be natural persons, and the UK Supreme Court reached the same result in 2023.

The live questions sit in the middle ground of AI-assisted work, and the US Copyright Office has drawn the lines across a guidance arc: the Zarya of the Dawn decision (2023) registered a comic’s human-written text and human-made selection and arrangement, but refused the individual Midjourney images; the March 2023 registration guidance requires applicants to disclose and disclaim AI-generated content; and the January 2025 Copyrightability report consolidated the doctrine — prompts alone, however many and however refined, do not make you an author, because the expressive choices happen inside the model. Protection attaches only to identifiable human contribution: the human-authored elements, creative selection and arrangement of AI material, or human modifications that themselves show authorship.

Not every jurisdiction agrees. The Beijing Internet Court (Li v. Liu, November 2023) found a Stable Diffusion image copyrightable, reasoning that the user’s iterative prompting and parameter choices constituted intellectual investment — and Chinese courts have followed with further pro-protection rulings. The UK, unusually, has a statutory category for computer-generated works (CDPA s9(3), author = the person who made the arrangements), though its survival keeps being questioned in consultations. For a multinational, the same output can be protected in Beijing, unprotected in Washington, and ambiguous in London.

Is this AI-touched output protectable in the US?

  1. A work created with AI involvement
  2. Generated autonomously, human contribution limited to prompts?

    Per the USCO’s 2025 report, prompts — even long, iterated ones — are instructions, not authorship: the expressive choices are made by the model.

  3. No copyright — public domain on arrival (Thaler)
  4. Identifiable human expression? Written text, human edits, creative selection/arrangement of AI material
  5. Protection for the human-authored elements only — disclose and disclaim AI portions when registering

    The Zarya of the Dawn result: text and arrangement registered, Midjourney images excluded.

  6. AI used as an assistive tool within human-controlled expression → full protection

    De-noising, in-painting a human photo, code autocomplete inside a human-designed program — the USCO treats these as ordinary tool use.

Interactive checkpoint quiz (2 questions) — open this page in a browser to take it.