On March 2, 2026, the U.S. Supreme Court denied certiorari in Thaler v. Perlmutter (No. 25-449), declining to review the D.C. Circuit’s March 2025 holding that human authorship is a bedrock requirement of U.S. copyright law. For research administrators and journal offices, the case itself was never really the interesting part — the interesting part is that it lands on a question scholarly-publishing bodies had already answered on their own, years earlier, through authorship criteria and disclosure policy rather than litigation.
What the Supreme Court actually decided
The underlying case began when Dr. Stephen Thaler sought to register a visual work, “A Recent Entrance to Paradise,” generated entirely by an AI system he called the Creativity Machine, listing the machine itself as sole author. The U.S. Copyright Office refused registration. The D.C. Circuit affirmed that refusal on March 18, 2025 (No. 23-5233), holding that the Copyright Act of 1976 requires a human being to be the author of a work in the first instance, and that a work produced autonomously by a machine, with no traditional authorial contribution from a person, is not eligible for copyright protection.
By denying certiorari, the Supreme Court did not issue an opinion, did not endorse the D.C. Circuit’s reasoning in its own voice, and did not resolve any question the lower court didn’t already reach. It simply left the D.C. Circuit’s ruling as the last word for now. That distinction matters for how much weight the outcome should actually carry in institutional policy — see the next section.
A cert denial is not an opinion — what it does and doesn’t settle
A denial of certiorari carries no precedential value of its own and signals nothing about the Court’s view of the merits; it simply means the case as briefed didn’t meet the Court’s criteria for review (commonly a circuit split, a question of exceptional national importance, or a lower-court error the Court wants to correct). What the denial does do practically is remove, for the foreseeable future, any near-term prospect that a higher court will disturb the “no purely machine-authored work is copyrightable” rule the Copyright Office, the district court, and the D.C. Circuit all converged on. Institutions that were holding off on AI-authorship policy pending a possible Supreme Court reversal no longer have a reason to wait on that specific question.
What the denial does not resolve is the harder, more common case: works where a person used AI assistance but also exercised real creative judgment — selecting, arranging, substantially editing, or combining AI output with original human-authored material. The Copyright Office’s own February 2025 report on AI and copyright authorship concluded that “prompting an AI system, by itself, does not supply the kind of human creative control the law requires” for the AI-generated portion, but left open that sufficiently substantial human modification or selection of AI output can itself be copyrightable. Thaler was always the easy case — zero human creative input, claimed machine authorship on the record. The mixed-authorship cases research offices actually encounter day to day (an AI-drafted figure caption revised by a co-author, an AI-assisted literature summary rewritten into original prose) sit in the territory this ruling leaves untouched.
Two tracks, one answer: courts and scholarly-publishing bodies already agreed
What makes this genuinely relevant to research administration isn’t the copyright holding in isolation — it’s that scholarly-publishing governance reached the same conclusion through a completely separate mechanism, years before any court weighed in:
- The ICMJE updated its recommendations in May 2023 to explicitly address AI tools alongside its four authorship criteria, and its position — developed further in guidance summarized in CASRAI’s AI co-authorship rejection (ICMJE 2023) entry — is that AI tools cannot satisfy authorship because they cannot take responsibility for a work, cannot agree to be accountable for its accuracy, and cannot be listed as a corresponding author.
- COPE’s position statement “Authorship and AI tools,” published February 13, 2023, reaches the identical conclusion by a different route: AI tools cannot be authors because, as non-legal entities, they cannot take responsibility for a work’s integrity, cannot declare conflicts of interest, and cannot hold or manage copyright or licensing agreements — they can only be disclosed as tools used, typically in the methods section.
Copyright law asks who legally originated a work; journal-authorship policy asks who can be held accountable for one. Thaler answers the first question. ICMJE and COPE had already answered the second, and arrived at the same practical outcome: an AI system is a tool to disclose, never a contributor to credit.
What this confirms for CRediT and contributor attribution
The CRediT taxonomy assigns each of its 14 contributor roles to a person who can be identified, credited, and held accountable for that specific contribution — the same accountability logic COPE and ICMJE apply to authorship generally. Nothing in Thaler changes CRediT’s structure or role definitions, and the case doesn’t purport to. What it does is remove a piece of legal uncertainty that occasionally got raised as a hypothetical objection to that structure: if AI-generated content could someday be independently copyrightable to the AI system itself, some institutions worried that might eventually complicate how contributor statements characterize AI-assisted work. With the human-authorship requirement now settled at the appellate level and cert denied, that specific hypothetical is off the table. The operative distinction for a contributor-role statement remains exactly what it was before the ruling: disclose the AI tool and how it was used (see CASRAI’s generative-AI disclosure statement entry), and credit the human contributors who exercised the judgment, verification, and revision that CRediT roles like Writing – Original Draft or Formal Analysis actually describe.
What research offices, journals, and PIs should do now
- Don’t wait on courts to finalize AI-disclosure policy. The Supreme Court route to a different outcome is closed for now; institutional policy shouldn’t have been contingent on it in the first place, but any that was should proceed on the ICMJE/COPE baseline directly.
- Keep authorship credit and AI disclosure as separate mechanisms, not a spectrum. An AI tool is never listed as an author or given a CRediT role, regardless of how substantial its contribution was — see CASRAI’s guide on whether AI can be listed as an author for the full ICMJE/COPE/publisher-by-publisher breakdown.
- Revisit copyright-ownership assumptions for AI-assisted outputs — manuscript text, figures, code, and data visualizations included. CASRAI’s guide on AI-assisted writing and copyright ownership covers the practical revision-and-documentation steps that keep a work’s human-authored expression clearly established.
- Watch the disclosure-standard track, not the litigation track, for what actually changes next. COPE, the International Science Council, STM, the World Conferences on Research Integrity Foundation, and the Global Young Academy are jointly developing a Global Reporting Standard for AI Disclosure in Research (informally the “Vancouver Standard”), with a second public-consultation round running July–September 2026 and a finalized standard targeted for later in 2026. That process, not any pending litigation, is where the next real change to disclosure requirements is likely to come from.
What’s still genuinely open
Three things Thaler leaves unresolved are worth tracking specifically: how much human modification of AI-generated content is “substantial” enough to be independently copyrightable (the Copyright Office’s February 2025 report sets a direction, not a bright line); how mixed human-AI authorship should be described in a contributor statement when the human role was primarily selection and editing rather than drafting; and how international jurisdictions with different authorship doctrines will treat the same fact patterns — the D.C. Circuit’s reasoning binds U.S. registration practice, not journals or funders operating under other countries’ copyright regimes. None of these are questions a Supreme Court cert denial was ever going to answer, and none are new as of March 2026 — they’re the same open questions CASRAI’s existing authorship and AI-disclosure guidance already flags, just without any lingering possibility that a reversed copyright ruling might reopen them from a different direction.
Frequently asked questions
Does the Supreme Court’s cert denial mean AI can now get some form of credit on a paper?
No. The denial leaves in place a copyright ruling about who can legally be an “author” of a registrable work; it says nothing about journal contributor policy. ICMJE, COPE, and the CRediT taxonomy already excluded AI tools from authorship and contributor-role credit on independent grounds, and that position is unaffected either way.
Does this change CRediT taxonomy or ICMJE’s four authorship criteria?
No changes to either are prompted by this ruling. See CASRAI’s comparison of ICMJE vs. COPE and the ICMJE authorship criteria guide for how those standards are actually structured, independent of this case.
Does this affect copyright in AI-assisted research data, code, or figures produced for a grant deliverable?
The same human-authorship principle applies to any copyrightable output, not just text or images — but the harder, more common case for research offices is mixed human-AI work, which this ruling doesn’t directly resolve. See the Copyright Office’s February 2025 guidance on substantial human modification, referenced above, for the current federal registration position.
Is the D.C. Circuit’s ruling binding outside the United States?
No. It governs U.S. copyright registration practice specifically. Institutions publishing or funding research internationally should confirm the authorship and AI-disclosure rules that apply under the relevant journal’s or funder’s own jurisdiction, which may differ.
