Are AI Generated Images Copyrighted?

The Copyright Problem Every AI Image User Faces

You generate an image in Midjourney for your next marketing campaign, and it performs beautifully. Then a competitor copies it pixel-for-pixel and uses it in their own ads — and there may be nothing you can legally do about it. This is the copyright problem that every business and creator using DALL-E, Midjourney, or Stable Diffusion now faces, and most of them don't know it exists.

Under current U.S. law, purely AI-generated images cannot be copyrighted. Copyright protection requires human authorship, and the courts have upheld that requirement at the highest level: the Supreme Court declined to review Thaler v. Perlmutter in March 2026, leaving the human-authorship rule firmly in place. A raw AI output, in other words, may be free for anyone to copy.

It gets more uncomfortable. The Copyright Office's January 2025 report concluded that prompts alone do not provide sufficient human control to make a user the author of the output. Typing a clever prompt is not the same as creating the image — the Office compares an AI user to a client who merely gives an artist general directions. That means your most carefully engineered prompt likely secures you no ownership rights at all.

The stakes are real for anyone using AI images in branding, packaging, or products. Consider what has happened to early test cases:

• In the Zarya of the Dawn case, images in Kris Kashtanova's graphic novel were denied copyright because they were "not the product of human authorship," though the human-written text and arrangement remained protected.
• Théâtre D'opéra Spatial, the award-winning AI artwork, was refused registration for failing to disclaim its AI-generated portions.
• The SURYAST application was denied because the AI system itself determined how to combine the base and style images.

There is a workable path — but it's narrower than most users assume. The Copyright Office has registered hundreds of hybrid works incorporating AI-generated material, with protection covering only the human contribution: sufficiently creative arrangements, modifications, or combinations. The dividing line is whether AI acts as a creative tool assisting a human, or as a stand-in for human creativity.

Meanwhile, the legal ground is still shifting. Andersen v. Stability AI, with a trial scheduled for April 2027, could reshape how image-generation tools and their outputs are treated — and several dozen lawsuits over AI training are working through the courts. For businesses, the practical takeaway is uncomfortable but clear: assume no one owns a raw AI image, invest real human creativity into anything you want to protect, and document that contribution carefully.

The Legal Answer: No Copyright Without Human Authorship

If you type a prompt into Midjourney and hit generate, the resulting image is not yours in the legal sense — and the highest court in the United States has now effectively signed off on that. Under current U.S. law, purely AI-generated images cannot be copyrighted, because copyright requires human authorship.

The question reached the courts in Thaler v. Perlmutter, where Stephen Thaler sought to register an image created autonomously by his AI system. In March 2025, the D.C. Circuit affirmed that the Copyright Act "requires all eligible work to be authored in the first instance by a human being," and the full court denied rehearing on May 12, 2025, according to a Congressional Research Service analysis. The Supreme Court declined to hear the case on March 2, 2026, leaving the human-authorship requirement firmly in place.

That outcome mirrors the Copyright Office's long-standing position. The Office refuses registration for works "produced by a machine or mere mechanical process that operates randomly or automatically without any creative input or intervention from a human author." Its January 2025 report went further on prompts specifically: given current technology, prompts alone do not provide sufficient human control to make a user the author of the output. The Office compares an AI user to a client who hires an artist and gives only general directions — not the person holding the camera.

Several registration battles illustrate how strictly this line is drawn:

• Zarya of the Dawn (Feb. 2023): Midjourney images in Kris Kashtanova's graphic novel lost copyright protection because they were "not the product of human authorship," though the human-written text and arrangement remained protected.
• Théâtre D'opéra Spatial (Sept. 2023): the award-winning AI artwork was denied registration for failure to disclaim its AI-generated portions, prompting the applicant to sue the Office.
• SURYAST (Dec. 2023): an AI-stylized image was denied because the AI system itself determined how to combine the base and style images.

Importantly, this does not make AI tools useless for creators. The Office has registered "hundreds" of hybrid works that incorporate AI-generated material, with protection covering only the human contribution — creative arrangement, modification, or combination of the AI output. The dividing line is whether a human exercises "ultimate creative control" or merely lets the AI stand in for human creativity. The Office has told Congress that new legislation on this question is "currently not needed."

For businesses using AI images in marketing, product pages, or documentation, the practical takeaway is sobering: a raw AI output sits closer to public domain than to protected property, meaning competitors can copy it freely. Law firm guidance recommends documenting human creative involvement, retaining prompts, and naming a human author in any registration. And if your teams are building AI workflows around image generation, agent skills can automate much of the surrounding process — SkillzBench is building a free directory and benchmarking platform to help you find and verify skills worth trusting.

One caveat: the copyrightability question is settled, but the training-data wars are not. Several dozen lawsuits over AI training — including Andersen v. Stability AI, headed to trial on April 5, 2027 — could still reshape how image-generation tools and their outputs are used commercially.

Hybrid Works: How to Actually Get Copyright Protection

Here's the good news buried in the legal headlines: AI-assisted images can absolutely be copyrighted — just not the AI parts. The U.S. Copyright Office has already registered hundreds of works that incorporate AI-generated material, and understanding how they got there tells you exactly what to do with your own.

The framework comes from the Copyright Office's March 2023 registration guidance, which allows registration of works containing AI-generated material when a human contributes "sufficiently creative" arrangements, modifications, or combinations. The catch: applicants must disclaim the AI-generated portions, and protection covers only the human contribution. According to a Congressional Research Service analysis, the Office distinguishes between AI as a creative tool assisting a human (protectable) and AI as a stand-in for human creativity (not protectable).

Three high-profile cases show where the line sits:

• Zarya of the Dawn (Feb. 2023): Kris Kashtanova's Midjourney-generated images lost copyright because they weren't the product of human authorship — but the human-written text and the selection and arrangement of images stayed protected.
• Théâtre D'opéra Spatial (Sept. 2023): This award-winning AI artwork was denied registration for failure to disclaim the AI portions — a reminder that honesty in the application itself matters. The applicant has since sued the Copyright Office.
• SURYAST (Dec. 2023): An AI-stylized image was denied because the AI system itself determined how to combine the base and style images, leaving no human creative control over the result.

Notice what these cases reward: creative human control and substantial modification. The Office's January 2025 report was blunt about what doesn't qualify — prompts alone "do not provide sufficient human control" to make you the author of the output, because they function as instructions conveying unprotectible ideas. You need to shape, edit, arrange, or combine the output in ways a human author would.

Practical steps follow directly from the precedents. Law firm guidance after the Supreme Court left the human-authorship rule intact recommends a clear playbook: ensure a human contributes creatively and is named as author, document that involvement, retain your prompts and edit history, and disclose AI-generated portions in your application. The Zarya registration was reconsidered precisely because the application didn't disclose Midjourney's role — omission can cost you everything.

For businesses building visual content workflows with AI, this documentation discipline is the difference between an asset you can enforce and one a competitor can freely copy. And if you're orchestrating AI agents to produce that content at scale, agent skills can help standardize the human-review and documentation steps — you can browse and test free ones on SkillzBench as the platform comes online.

The Office has told Congress new legislation on AI copyrightability is "currently not needed," so this hybrid framework is the system you'll be working within for the foreseeable future.

Practical Steps to Protect Your AI-Assisted Images

Knowing the rules is one thing; operationalizing them across a marketing or design team is another. The good news is that the Copyright Office has already registered hundreds of works incorporating AI-generated material where a human exercised genuine creative control, so a clear path to protection exists for businesses that follow it deliberately (per Congressional Research Service analysis).

Start by adding meaningful human creative input — and documenting it as you go. The Office's January 2025 report concluded that prompts alone do not make you the author of an AI output, so substantial human modification, arrangement, or editing is what earns protection. Keep records of every prompt, edit, and revision so you can prove the human contribution later. Law firm guidance recommends exactly this: name the human author, document their involvement, and set internal policies on ownership and attribution.

Your implementation checklist:

• Add and document meaningful human creative input — editing, composing, or arranging beyond the raw output.
• Retain prompts, edit histories, and drafts as evidence of the human contribution.
• Name the human author explicitly in registration applications.
• Disclaim AI-generated portions in applications — omissions have cost applicants their registrations.
• Treat raw, unmodified AI outputs as copyable by anyone, since no one holds copyright in them.

That last point deserves emphasis. Disclosure failures have real consequences: the Théâtre D'opéra Spatial registration was denied for failing to disclaim AI-generated portions, and Zarya of the Dawn's Midjourney images lost protection after the applicant's nondisclosure prompted reconsideration (Reuters reported). Honest applications protect what you can protect.

Finally, keep one eye on the courtroom. Training-data litigation remains unsettled commercial risk: Andersen v. Stability AI — where Judge Orrick allowed claims that image-generation systems illegally store artists' works — goes to trial April 5, 2027, and Getty Images v. Stability AI adds trademark claims over outputs replicating watermarks (case trackers note both). A verdict there could reshape which tools and outputs are safe to use commercially.

If your team manages this workflow across agents and tools, SkillzBench is building a free directory of vetted agent skills — security-scanned and benchmarked — to help automate documentation and record-keeping steps like these.

Where This Is Heading — and How to Stay Ahead

The dust hasn't settled — it's just starting to kick up. While the human-authorship rule now stands firm after the Supreme Court's March 2026 cert denial in Thaler v. Perlmutter, the fair-use question around AI training data remains wide open, and it could reshape how every image-generation tool operates.

Two June 2025 rulings trended favorably toward AI companies: Bartz v. Anthropic called training "quintessentially transformative" (while finding pirated-book downloads were not fair use), and Kadrey v. Meta ruled training on pirated books fair use — while explicitly noting that market-harm evidence might tip the balance in future cases, per a Congressional Research Service analysis. But these are district-level decisions, not the final word.

Meanwhile, the image-specific cases grind forward. Andersen v. Stability AI is scheduled for trial on April 5, 2027, and Getty Images v. Stability AI adds trademark claims over outputs replicating Getty watermarks, according to a case tracker from BakerHostetler. Judge Orrick has already found plausible allegations that Stable Diffusion "creates copyright infringement ... by design."

Don't expect Congress to rescue anyone soon. The Copyright Office has told lawmakers that new legislation on AI copyrightability is "currently not needed," so the courts will keep defining the landscape case by case.

What businesses should watch:

• Appellate rulings on fair use — the *Bartz* and *Kadrey* decisions will almost certainly be appealed, and higher courts may draw the line differently.
• The *Andersen* trial in April 2027, which directly addresses image-generation systems.
• New Copyright Office guidance at its AI resource hub — its August 2023 inquiry drew over 10,000 comments, and follow-up reports are still rolling out.
• Hybrid-work registrations — the Office has already registered "hundreds" of works with AI material, so the protectable middle ground keeps growing.

The practical takeaway: keep documenting human creative input, retain prompts and edit histories, and disclaim AI portions in any registration. That paper trail is what separates a protectable work from a public-domain asset competitors can freely copy.

Staying current is itself an operational challenge — the law, the tools, and the workflows around AI image use all shift monthly. That's where automation helps: agent skills can handle the documentation and workflow steps around AI image use, and SkillzBench is building a free directory where every skill is security-scanned, license-checked, and benchmarked so you know which ones actually work. Browse and test free skills on SkillzBench to see how verified agents can keep your AI image pipeline — and its paper trail — running clean.