AI Content Copyright: Can You Own AI-Generated Writing?

The Case That Changed Everything: Thaler v. Perlmutter

To understand where AI copyright law stands in 2026, you need to understand one case that has been winding through the federal courts since 2022. Thaler v. Perlmutter was brought by Dr. Stephen Thaler, who sought copyright registration for an image created entirely by his AI system, the Creativity Machine — with Thaler explicitly and repeatedly disclaimed any personal creative input. The US Copyright Office rejected the application in August 2022. Thaler appealed.

In August 2023, the DC District Court upheld the Copyright Office's refusal, citing the bedrock principle that copyright requires human authorship — a principle rooted in the Constitution's "Authors and Inventors" clause and codified in copyright statute. In 2025, the DC Circuit Court of Appeals affirmed the lower court's ruling. And on March 2, 2026, per reporting by CNBC, the US Supreme Court declined to hear the case — leaving the DC Circuit ruling as the definitive federal statement on the issue.

The Supreme Court's denial of certiorari does not technically create binding precedent, but in practice it is the closest available signal that the Court agrees with the existing ruling — or at minimum, does not find the question urgent enough to take up. For all practical purposes, the question of whether purely AI-generated works can be copyrighted in the United States has been answered: they cannot.

Per analysis by Mayer Brown and Holland & Knight, both firms published detailed commentary on the Supreme Court denial in March 2026, describing it as confirmation that "human authorship is a foundational requirement of US copyright law, even in this era of rapid AI advancement."

The Copyright Office's January 2025 Report: Defining "Sufficient Human Control"

The Copyright Office published Part 2 of its Copyright and Artificial Intelligence series in January 2025. Unlike the case law — which addressed the extreme of a work created with zero human input — the report addressed the far more common and commercially important question: when does human involvement in AI-assisted creation create protectable copyright?

The report's central test is "creative control." Quoting directly from the Copyright Office's January 2025 document: "What matters is the extent to which the human had creative control over the work's expression." This is a deliberately flexible standard, intended to accommodate the enormous variety of ways humans and AI tools now interact in creative work.

The report also addressed the question of prompts specifically: "Prompts essentially function as instructions that convey unprotectable ideas." This is the Office's direct answer to the intuition that a highly crafted, creative prompt deserves copyright protection. It does not. Prompts are analogous to a commissioning editor telling a freelancer what to write — the instructions are not the creative expression, and the copyright belongs to the writer, not the editor. In AI's case, the "writer" is not a human, so no one holds the copyright.

The Business Consequence No One Is Talking About

The copyright question has a practical implication for businesses that has received surprisingly little coverage: purely AI-generated content is effectively in the public domain. This means that any competitor, aggregator, or platform can legally copy, republish, reuse, or build on your AI-generated content without your permission and without compensation. The content you paid a team to generate using ChatGPT is not your proprietary asset — it is freely available for anyone to use.

For content-heavy businesses — media companies, marketing agencies, publishing operations — this is a strategic threat that is easy to underestimate. A content moat built on AI-generated articles provides zero competitive protection from a copyright perspective. A competitor can scrape your AI-generated content, repost it, and face no legal consequences beyond possible terms-of-service claims between you and the scraper — which are a far weaker protection than copyright.

The practical response is to ensure that AI-generated content is substantively enhanced with human editorial contribution sufficient to create copyright protection. This means more than light editing — it requires genuine creative decisions, original analysis, and author-specific perspective that distinguishes the final work from the AI's raw output. Knowing whether your published content is substantially AI-generated is therefore not just an editorial question but a legal and competitive strategy question.

Tools that assess AI content proportion — like those covered in our AI detector comparison — serve a dual purpose here: they tell you both whether your content will be flagged by external AI detectors and, implicitly, whether it likely meets the human creative contribution threshold for copyright protection.

The Training Data Question: Fair Use or Infringement?

A separate but related legal question is whether AI companies violated copyright when they trained their models on copyrighted text. This question is being litigated in multiple cases simultaneously, and the outcomes so far have been mixed.

The most significant 2025 ruling was in June, when a court granted partial dismissal in Meta's favor, finding that LLM training constitutes fair use because the training process is transformative — it produces a statistical model, not reproductions of specific copyrighted works. This finding, if it holds on appeal, would substantially limit AI company liability for training data.

However, the $1.5 billion settlement in Bartz v. Anthropic — reported by the Authors Guild in 2025 — tells a different story. Anthropic faced potential statutory damages penalties for downloading millions of pirated copies of books for training purposes, even if the fair use argument might ultimately have prevailed. The settlement amount suggests AI companies are not fully confident in fair use defenses when the source material was obtained from illegitimate sources rather than the open web.

As of April 2026, per Norton Rose Fulbright's litigation tracker, multiple cases remain pending — including cases against OpenAI where a March 2026 court order compelled production of server logs containing millions of training data records. The litigation landscape will continue evolving through at least 2027, and any business decision predicated on "AI training is clearly fair use" is operating on uncertain legal ground.

What Publishers, HR Teams, and Educators Actually Need to Know

For publishers and media organizations: your editorial contracts should explicitly address AI-generated content and require human authorship sufficient to support copyright claims. A freelancer who submits AI-generated text is not delivering copyrighted work you can own or exclusively license — even if they wrote the prompt. Contracts should specify the degree of human authorship required and grant representations to that effect.

For HR professionals dealing with AI-generated job application materials: there is no copyright issue with scanning, analyzing, or sharing AI-generated resumes and cover letters. Because pure AI output cannot be copyrighted, there is no candidate IP at stake when you copy or quote from a purely AI-generated application. The concern is entirely about authenticity and candidate misrepresentation — not copyright.

For educators and academic publishers: the academic integrity question and the copyright question are independent. A student who submits AI-generated text without disclosure may violate academic integrity policies regardless of copyright status. Journal authors submitting AI-generated research face similar disclosure requirements that are independent of copyright. Many publishers — including major academic publishers — now require authors to disclose AI tool use in their manuscripts. Our guide to ethical AI use in academic writing covers disclosure norms in detail.

International Dimensions: EU, UK, and China

The US framework — human authorship required, no exceptions — is not universal. The international AI copyright landscape is fragmented, and businesses operating across jurisdictions face genuine complexity.

The European Union is developing its framework through the EU AI Act, which came into force in August 2024. The Act imposes transparency obligations on AI-generated content but does not directly resolve copyright ownership questions. EU member states retain divergent national approaches to AI authorship — some EU jurisdictions allow "computer-generated works" to receive shorter copyright terms with a designated human author, a concept closer to what the UK has historically allowed.

The United Kingdom has a specific provision under the Copyright, Designs and Patents Act 1988 (Section 9(3)) that grants copyright in computer-generated works to "the person who makes the necessary arrangements for the creation of the work." This would potentially support copyright in AI-generated works for the human user who commissioned them — a materially different outcome than US law. However, the UK Intellectual Property Office launched a consultation in 2022 questioning whether this provision should be maintained, given AI's evolution beyond traditional "computer-generated" work.

China took a different path: the Beijing Internet Court ruled in November 2023 that AI-generated content can be copyrighted when "human originality and creativity are embodied in the expression." This ruling predates the US Copyright Office's January 2025 report and diverges from the US position — potentially creating opportunities for businesses with operations in China to establish copyright claims they cannot make under US law.

For businesses with global content operations, the practical advice is jurisdiction-specific IP registration strategy combined with content documentation practices — maintaining records of the human editorial decisions made in AI-assisted content production, not just the AI tool prompts.

How to Establish Copyright in AI-Assisted Content: A Practical Framework

The Copyright Office's "creative control" standard is deliberately fact-specific. There is no bright-line rule. But practitioners can construct a defensible paper trail by doing the following:

• 1.

  Document selection decisions

  Keep records of which AI outputs you selected from multiple generations, and why. The act of creative selection from a range of AI-generated options is itself an authorship contribution.

• 2.

  Track editorial revisions

  Use version control or tracked changes to show the extent of human editing on AI-generated drafts. Substantial tracked changes provide evidence of human creative contribution that goes beyond minor stylistic cleanup.

• 3.

  Add original material explicitly

  Human-authored sections interspersed with AI-generated content create a mixed work where at minimum the human sections are protected. Original analysis, personal experience, and exclusive research are the strongest contributions.

• 4.

  Register works where human contributions are substantial

  For high-value content, copyright registration provides legal presumption of validity and entitlement to statutory damages — making enforcement more practical. The Copyright Office's March 2023 guidance requires disclosure of AI contributions in registration applications.

• 5.

  Assess AI proportion before making copyright claims

  Before asserting copyright in content, understand its actual AI proportion. Using an AI detection tool on your own content is a reasonable due diligence step — if a tool scores your content as 85% AI-generated, your copyright position is correspondingly weak.

What Remains Genuinely Unsettled

The foundational question — can purely AI-generated content be copyrighted in the US — is settled. But several important adjacent questions are not.

Where is the threshold for "sufficient" human creative control? The Copyright Office refused to draw a bright line, and courts have not yet provided one. A work where a human substantially rewrote 40% of AI output is clearly on one side; a work where a human only corrected punctuation and adjusted tone is clearly on the other. The vast middle ground — light structural editing, topic-guided prompting, iterative refinement — remains legally ambiguous.

How does copyright work in multi-human, multi-AI collaborative pipelines? Increasingly, content workflows involve multiple humans and multiple AI tools at different stages. Determining which human contributions are "creative control" contributions versus ministerial contributions — and how to allocate authorship accordingly — is an unsolved problem that will likely require litigation to clarify.

What are the remedies when someone copies purely AI-generated content without permission? Even though there is no copyright, other legal theories may apply: trade secret law (if the AI output reflects proprietary prompting methods), unfair competition, or contractual restrictions on use. These are weaker protections than copyright, but they are not zero.

The speed of legal development in this area — three major milestone rulings or reports in 15 months (CO January 2025, Meta training fair use June 2025, Supreme Court cert denial March 2026) — suggests the framework will continue to evolve. For businesses and creators, the practical approach is to document human contributions carefully, avoid relying on copyright protection for purely AI-generated content, and monitor litigation developments in your specific industry vertical.

Frequently Asked Questions