The Copyright Status of AI-Generated Art: Still Unresolved

What happened

Can AI-generated images, text, music, and other creative works be copyrighted? Almost three years into the generative AI boom, the answer remains unsettled — and the uncertainty itself is reshaping who captures value from creative work. This question sits alongside the training data copyright fight (see AI Copyright Rulings Are Shaping the Training Data Economy) as the other half of a legal pincer that will define the economics of AI Creative Industries.

The US Copyright Office and federal courts have been chipping away at the question since 2022, producing a body of rulings and guidance that points in a clear direction without quite arriving. The core principle: copyright requires human authorship. What that means when a human uses an AI tool to produce creative output is where things get complicated.

The landmark case is Thaler v. Perlmutter, decided in August 2023 by Judge Beryl Howell in the District of Columbia. Stephen Thaler sought to register a visual artwork generated entirely by his AI system, DABUS, listing the AI as the author. The Copyright Office refused, and the court agreed. "Copyright has never stretched so far as to protect works generated by new forms of technology operating absent any guiding human hand," Howell wrote. The ruling was narrow — it addressed only the case of a work produced autonomously by a machine with no claimed human creative input — but it established the baseline: an AI cannot be an author under US law.

The more instructive case is Zarya of the Dawn. In February 2023, the Copyright Office partially revoked the copyright registration for a comic book by Kris Kashtanova that used Midjourney-generated images alongside human-written text and human-directed layout. The Office granted copyright for Kashtanova's text and the selection and arrangement of images — the editorial choices a human made — but denied protection for the individual AI-generated images themselves. The reasoning: Midjourney users cannot predict or control the specific images the tool produces with enough precision to claim authorship of those images.

This distinction — between the human's creative choices and the AI's output — became the foundation of the Copyright Office's broader guidance. In a series of Federal Register notices through 2023 and 2024, the Office laid out its position: works generated by AI are not copyrightable, but works that involve sufficient human creative control may be, depending on the degree and nature of human involvement. A human who uses AI as a tool — making substantive creative decisions about selection, arrangement, and modification — may have a copyrightable work. A human who types a prompt and accepts what comes out likely does not.

The prompt question sits at the center of the debate. Is writing a detailed, carefully crafted prompt an act of authorship? The Copyright Office has said, generally, no. The reasoning is about control and predictability: a prompt describes a desired outcome, but the AI's interpretation introduces an element of unpredictability that breaks the chain between human intention and creative expression. This is not a settled legal conclusion — no federal appellate court has addressed it — but it reflects the Office's current operating framework and has guided registration decisions since 2023.

Why it matters

The copyright status of AI-generated works is not an abstract legal question. It determines who owns the output of a rapidly growing segment of the creative economy — and the current answer creates an asymmetry that favors AI companies over everyone else.

Start with the practical implications. If AI-generated content is not copyrightable, then any business using AI to produce marketing materials, stock images, product descriptions, design assets, or code has no legal protection for that output. Competitors can copy it freely. A company that generates its entire visual identity with Midjourney or DALL-E may find that nothing in its brand toolkit is protectable. A publisher that uses AI to generate illustrations cannot prevent others from reproducing them. A developer whose codebase includes substantial AI-generated code may discover that those portions are in the public domain.

This matters for contracts and licensing as well. When a client commissions work from a freelancer, they typically expect to receive copyrightable work product — something they can own, license, and enforce against infringers. If the freelancer uses AI tools and the output lacks copyright protection, the client has paid for something that cannot be exclusively owned. Employment agreements, work-for-hire contracts, and licensing deals all assume the existence of copyrightable output. AI-generated content may fall through the legal floor.

The redistribution dynamics here are stark — and they mirror the Pricing Power dynamics across the AI industry. AI companies have built proprietary, trade-secret-protected models — their core asset is legally shielded. But the output those models produce for users may carry no legal protection at all. The value concentrates at the tool layer. OpenAI, Stability AI, Midjourney, and others capture revenue through subscriptions, API fees, and enterprise contracts. Their customers generate output that competes in the marketplace but cannot be defended as intellectual property. The tool maker extracts rent; the tool user gets convenience without ownership.

Now layer in the training data problem. The models that generate uncopyrightable output were themselves trained on copyrighted human-made works — the very works they now compete with. As detailed in AI Copyright Rulings Are Shaping the Training Data Economy, the legal question of whether training on copyrighted material constitutes fair use remains unresolved. But the structural effect is already visible: human artists' copyrighted works were used, without permission or compensation in most cases, to build tools that produce competing works that need no copyright because they need no legal defense — they are cheap enough to be disposable.

For freelance creators and independent artists, the asymmetry is corrosive. Your hand-made illustration is copyrightable. The AI-generated image competing with you for the same commission is not. In theory, this should advantage human creators — their work has legal protections that AI output lacks. In practice, it does not help when the client chooses the cheaper, faster option and does not care about copyright ownership. The legal distinction between protected and unprotected work matters only when someone wants to enforce that protection. For most commercial applications — social media graphics, blog illustrations, internal presentations — enforcement is irrelevant. The buyer wants the image, not the right to sue over it.

This connects to a deeper question about the fair use intersection. The argument that training on copyrighted works is transformative fair use — that the model produces something fundamentally new — sits in tension with the argument that the model's outputs are not copyrightable because they lack human authorship. If training is fair use because the output is transformative, but the output is simultaneously not original enough (in the human authorship sense) to be copyrightable, then the legal framework creates a one-way valve: copyrighted works go in, unprotectable works come out, and the only entity that captures durable value is the one that owns the model.

What to watch

The legal landscape is moving on multiple fronts, and the resolution — if one comes — will not be uniform.

International divergence. The United States is not the only jurisdiction grappling with this question, and other countries are reaching different conclusions. The United Kingdom's Copyright, Designs and Patents Act 1988 already contains a provision — Section 9(3) — that assigns authorship of "computer-generated" works to "the person by whom the arrangements necessary for the creation of the work are undertaken." This was drafted long before generative AI, and whether it covers AI-generated outputs is contested, but it provides a legal hook that US law currently lacks. The UK Intellectual Property Office launched a consultation on AI and copyright in 2022, and the debate remains active. In the European Union, copyright generally requires a work to reflect the author's "own intellectual creation" — a human standard — but member states vary in their application, and the EU has not issued unified guidance on AI-generated works. In China, courts have begun recognizing copyright in AI-generated works where there is sufficient human input: a Beijing court ruled in late 2023 that an AI-generated image could be copyrighted when the human user made significant creative choices in prompting, selecting, and refining the output. These divergent approaches mean that the same AI-generated work may be protected in one jurisdiction and unprotectable in another — a headache for any business operating internationally.

Legislative responses. Congress has shown interest but taken no action, even as the executive branch has signaled awareness of the issue. The Copyright Office published its report on copyright and AI in mid-2025, recommending that Congress consider whether to create new forms of protection for AI-generated works — something short of full copyright but more than nothing. Proposals floated in academic and policy circles include sui generis rights (a new category of IP specifically for AI-generated works), compulsory licensing schemes (requiring AI companies to pay into a fund that compensates creators whose works were used in training), and attribution requirements (mandating disclosure when content is AI-generated). None of these has advanced to legislation. The political dynamics are complex: the tech industry opposes new obligations, the creative industries want stronger protections, and Congress has demonstrated limited capacity for nuanced technology regulation.

The ownership question. Separate from copyrightability is the question of who owns AI-generated work product when it is produced in a commercial context. If a company's employee uses an AI tool to create marketing copy, and that copy is not copyrightable, does the company have any proprietary claim to it? Current work-for-hire doctrine assumes copyrightable output. If the output is not copyrightable, the contractual framework may need rebuilding. This is a quieter issue than the headline court cases, but it affects every business incorporating AI into creative workflows.

The unresolved copyright status of AI-generated art is not a gap that will fill itself. It is a structural feature of a legal system that was built for human creators and has not yet adapted to a world where machines produce creative work at industrial scale. The longer the uncertainty persists, the more it benefits the companies that own the models — they operate in the ambiguity, profiting from tools whose outputs exist in a legal no-man's-land.

But the uncertainty also creates an opening. If AI-generated works cannot be copyrighted, then human creativity retains a legal advantage that no amount of compute can replicate. The challenge is translating that legal distinction into economic leverage — through Collective Bargaining, through platform policies that distinguish human and AI content, and through the kind of provenance and attribution systems that organizations like AI Now and others are building. For creators, the message is plain: the work you make with your own hands and mind has legal protection. The work the machine makes in your name, for now, does not. And the machine was trained on yours. The question is whether creators can organize around that distinction before it becomes economically irrelevant.

Where this fits

The copyright status of AI output is the mirror image of the training data copyright question. Together, they define the legal envelope for generative AI: what goes in (training data) and what comes out (generated content). The answers will determine whether the value created by AI tools flows to the labs that build them, to the users who prompt them, or — through licensing, regulation, or collective action — back to the creators whose work made them possible. The EU AI Act and emerging international frameworks will shape the regulatory context, while market experiments like Shutterstock's contributor licensing deals are testing economic models in real time.

Related

• AI Copyright Rulings Are Shaping the Training Data Economy
• AI Creative Industries
• Say No to Suno
• Collective Bargaining
• Who Owns Training Data
• Open (For Business)
• EU AI Act Takes Effect

Sources

• Thaler v. Perlmutter, No. 1:22-cv-01564 (D.D.C. Aug. 18, 2023). Court Opinion
• US Copyright Office. "Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence." Federal Register 88, no. 51 (March 16, 2023). Federal Register
• US Copyright Office. "Zarya of the Dawn (Registration # VAu001480196)." Letter to Van Lindberg, February 21, 2023. Copyright.gov
• UK Copyright, Designs and Patents Act 1988, Section 9(3). Legislation.gov.uk
• Guadamuz, Andres. "Artificial Intelligence and Copyright." WIPO Magazine, October 2017 (updated 2024). WIPO
• US Copyright Office. "Copyright and Artificial Intelligence, Part 1: Digital Replica Report." 2025. Copyright.gov