Why current AI cannot be an author under copyright, according to one philosopher
A peer-reviewed paper argues that copyright treats an author as someone who can answer for a work. On that reading, current AI systems fall short, but people who steer AI closely may still qualify.

AI illustrationWhat the study shows
This is a paper in legal philosophy by Caterina Moruzzi of the University of Edinburgh. It was published in Philosophy & Technology after peer review. It has no experiments or data. It is an argument built from court cases, statutes and earlier scholarship, with a focus on US law.
The author's starting point is that being an author is a status, not just a fact about who pressed the button. That status carries rights, such as credit and control over copies. It also carries duties, such as not plagiarising. So an author must be someone who can answer for the work.
She names three conditions that together make someone a fitting author. She calls them the ACE cluster. Intentional agency means the work comes from a purposeful aim. Guided control means the person steers how the work takes shape. Relational expression means the work states a position that others can attribute to the author and challenge.
The author argues that the familiar rule that authors must be human is a shortcut for these three capacities. It is not, on her reading, a bias about species. Current AI systems are excluded because they have not been shown to meet all three together, not because they are machines.
She tests this against known cases. In Thaler v. Perlmutter, a US appeals court refused copyright for an image listed as made solely by an AI system. In her reading, nobody met the three conditions, so the image had no author. In the Zarya of the Dawn comic, the US Copyright Office protected the human text and layout but not the raw images made with the Midjourney image generator.
She then weighs three legal options. Naming AI as an author is hard to justify. Treating human users as authors works when their control is substantial. Leaving fully automated output unowned, in the public domain, fits cases where no one meets the bar.
Our reading
In our reading, the most useful move here is reframing the human-only rule. Critics call it prejudice against machines. The author recasts it as a test of responsibility: can someone be named, praised, blamed and asked why. That gives a firmer reason for current law than species alone, and it leaves the door open if AI systems ever meet the test.
The practical message for most organisations is about the middle ground. The author sets the bar for human authorship above one-click generation and below control of every detail. A designer who writes many prompts, picks among outputs, edits them and arranges them into a finished piece may well be the author. Someone who types one generic prompt and keeps the first result probably is not.
A concrete example: a marketing team uses an image generator for a campaign. If the team only selects a stock-style output, a competitor may be free to reuse it. If the team has detailed records of how it shaped, edited and combined the images, it has a much stronger claim to own the result.
Read the rest of this story
Membership is free. You get every article in full and the Issue every Sunday.
We send a sign-in link, no password needed. Already a member? Sign inReadiness assessment
| Level | Status | Evidence |
|---|---|---|
| I Lab only | Met | The paper is a theoretical framework with no implementation, system or tool. The ACE cluster is a conceptual device, not something that could be deployed. |
| II Benchmarked | Not yet | There is no benchmark or empirical test. The paper itself says the threshold for authorship cannot yet be measured precisely. |
| III Open source | Not yet | No code, weights or data are released. The data availability statement says 'Not applicable'. |
| IV In production | Not yet | No named organisation uses the ACE framework. Courts and the Copyright Office are cited as consistent with it, but the paper says they do not state it in these terms. |
What would move it up
The readiness scale fits this paper poorly because it is not a technical method. The nearest steps would be turning ACE into a testable checklist, then checking it against a public set of registration decisions or court rulings to see if it predicts outcomes (level 2). Release of that checklist or case set would reach level 3. A copyright office, court or company documenting use of it would reach level 4.
The study
Caterina Moruzzi. AI and the Responsibility-Bearing Status of Authorship. Philosophy & Technology, 14 November 2026. doi:10.1007/s13347-026-01235-w
Notes
- Moruzzi, C. (2026). AI and the Responsibility-Bearing Status of Authorship. Philosophy & Technology 39:225. Peer reviewed, open access under CC BY 4.0. Source
- No code, data or tools are released. The paper's data availability statement reads 'Not applicable'.
- Readiness: level I, lab only. Conceptual framework with no benchmark, no released artefact and no documented use by a named organisation.
- US Copyright Office letter on Zarya of the Dawn (21 February 2023), as cited in the paper. Source
- The work was funded by the Bridging Responsible AI Divides (BRAID) programme through the Arts and Humanities Research Council.
Edited and signed by
Stefan PerdalEditor and publisher86 stories editedStefan read the study, set the readiness level and approved this story. An AI system found the paper and wrote the first draft.
How this was made
An AI model read the study and wrote this article in the structure Science Underway uses. An editor reviewed it and approved it for publication. Check the original study before you rely on the details.
How we workNo corrections

