Can AI Art Be Copyrighted? Where the Law Actually Stands (2026)

The current state of AI-generated art copyright in the United States can be summarized in one sentence: works generated entirely by AI without meaningful human creative input are not eligible for copyright protection, but works that combine AI generation with substantial human creative contribution can be.

That sentence has been the source of years of legal arguments, several court cases, and substantial confusion among working artists. Here’s what’s actually true in 2026, what’s still being decided, and what working creators using AI tools should actually do.

The Thaler decision and what it means

The foundational case in this area is Thaler v. Perlmutter, decided initially in 2023 and affirmed on appeal in 2024. Stephen Thaler had attempted to register a copyright on an image generated entirely by his AI system, “Creativity Machine,” listing the AI as the author with himself as the owner.

The US Copyright Office refused registration. Thaler sued. The court agreed with the Copyright Office: human authorship is a fundamental requirement of US copyright law, and an AI system cannot be the author of a copyrighted work.

The decision was narrow but clear: works with no human author are not copyrightable. The court declined to address whether works with substantial human input alongside AI generation could be protected, since that wasn’t the case before it.

The implication: pure AI generation, with the user only providing minimal prompting, doesn’t produce copyrightable output. AI-assisted creation with substantial human creative work mixed in is a different question.

What the US Copyright Office has actually said

The US Copyright Office issued formal guidance in March 2023 (updated several times since) on AI-generated works. The current position:

Pure AI generation is not copyrightable. Output from a text prompt alone, with the AI doing all the creative work, is in the public domain.

Substantial human creative contribution can be copyrighted. When a human makes substantial creative choices that meaningfully shape the output, the human’s contributions can be copyrighted.

Prompts alone don’t constitute creative authorship. Writing a clever prompt doesn’t make you the author of the AI’s output. The Copyright Office has been explicit on this point.

Selection, arrangement, and modification can be copyrightable. If you generate 50 images and select one, the selection is a creative act. If you generate an image and then substantially modify it through human effort, the modifications are copyrightable. If you arrange AI-generated elements into a larger composed work, the arrangement is copyrightable.

What this means in practice: a Midjourney image from a single prompt is not copyrightable. A graphic novel that uses Midjourney images as starting points but adds substantial human drawing, layout, story, and dialogue is copyrightable in those human contributions.

The Zarya of the Dawn case

A specific case worth knowing: in 2023, Kris Kashtanova received a copyright registration for a graphic novel (“Zarya of the Dawn”) that contained Midjourney-generated images. The US Copyright Office initially issued the registration, then partially rescinded it.

The final outcome:

The arrangement of the work (the comic’s text, layout, and creative arrangement) was protected
The individual Midjourney images themselves were not protected
The combined work has copyright in the human-authored elements only

This case illustrates the practical line: text, arrangement, story, and human-drawn additions are copyrightable. Pure AI-generated visuals within the same work are not.

Where other jurisdictions stand

The US position is the most restrictive among major copyright jurisdictions. Other countries are working through similar questions:

United Kingdom: UK copyright law has an explicit provision (Section 9(3) of the Copyright, Designs and Patents Act 1988) for “computer-generated works” where there is no human author. These are protected for 50 years, with the author considered to be the person who made “the arrangements necessary for the creation of the work.” This is more permissive than the US position, though its application to modern AI is still being clarified.

European Union: The EU has been working on AI regulation including IP implications. As of 2026, no harmonized EU rule explicitly addresses AI-generated work copyright. Member states vary. The general direction has been toward requiring human authorship.

Japan: Japan has been relatively permissive about AI training on copyrighted works for non-commercial research purposes. The output question is less clearly addressed.

China: Chinese courts have issued decisions both granting and denying copyright protection to AI-generated works in specific cases. The law is still evolving.

For global creators, the safest approach is to assume the US rule (pure AI generation = no protection) and design workflows that include meaningful human creative contributions.

What working creators should actually do

The practical playbook for creators using AI tools:

Step 1: Document your human creative process

For any work that involves AI generation, document the human creative contributions:

Iterative prompting and creative direction
Selection from multiple generated options
Editing, modification, and refinement of generated output
Composition and arrangement of multiple elements
Original work mixed with AI-generated elements

This documentation supports your copyright claim on the human-authored portions.

Step 2: Register the human-authored portions

For substantial works that combine AI and human creation, register the work with appropriate disclosure. The US Copyright Office requires disclosure of AI use when registering. Failing to disclose can invalidate the registration.

The disclosure typically describes:

Which portions are AI-generated
Which portions are human-created
The nature of the human creative contribution

Registration grants copyright in the human contributions. The AI portions remain unprotected.

Step 3: Be honest in your registration filings

A growing problem: creators registering AI-assisted works without disclosing the AI use. The US Copyright Office has explicit penalties for non-disclosure. Discovered later, your registration can be invalidated, exposing you to liability for false claims.

When in doubt, disclose. The registered work still has copyright protection for the human contributions. The AI disclosure protects the registration’s validity.

Step 4: Consider what’s safe to claim

If your work is substantially AI-generated:

You don’t have copyright in the AI portions
You can still use the work commercially (no one else has copyright either)
Other people can also use the AI-generated portions
Your human contributions retain copyright

If your work is substantially human-created with AI assistance:

Your human contributions have copyright
The AI assistance is part of your process, not the substance of the work
Standard copyright protection applies to the work as a whole

The AI training question (separate from output)

A different but related question: is the training of AI models on copyrighted works itself fair use?

This is being actively litigated in 2026. Major cases include:

NYT v. OpenAI (filed 2023, ongoing) - claims OpenAI’s training on NYT content is infringement
Getty Images v. Stability AI (filed 2023, ongoing) - similar claims for visual training
Andersen v. Stability AI (filed 2023, partially dismissed, partially proceeding) - class action by visual artists

The outcomes will shape AI copyright law for the next decade. As of 2026:

AI companies argue training is fair use
Rights holders argue it’s mass infringement
Courts have generally let cases proceed past initial motions
No final appellate decisions yet on the core fair use question

What this means for working creators:

If your work has been used in AI training, you may have potential claims (especially if class actions develop)
Opt-out registries exist for some platforms (haveibeentrained.com, etc.)
Documentation of your published work is valuable regardless

For the broader analysis of where blockchain timestamping fits in, see our piece. The same registration discipline that helps for traditional copyright disputes also positions you for potential AI training claims.

The publicity rights angle

Sometimes the more useful claim for AI-generated content isn’t copyright but rights of publicity (in the US) or image rights (in other jurisdictions).

When AI generates content that includes your likeness, voice, or distinctive style:

Copyright covers the underlying work
Rights of publicity cover commercial use of your likeness
These are separate protections, sometimes available simultaneously

For the related discussion of voice cloning and AI, see our audio guide.

The compute power and process arguments

A specific argument that’s emerged in some cases: that the human creator’s choices about which AI model to use, what hardware to use, how to direct the generation process, and which iterations to keep constitute sufficient creative contribution.

This argument has had mixed reception. The US Copyright Office’s position is that these technical choices don’t substitute for substantive creative input into the output itself. UK courts have been more receptive to similar arguments.

In practice, working creators are best served by ensuring that meaningful human creative work is present in the actual output, not just in the process choices.

Specific scenarios

A few common situations and what they mean for copyright:

“I used Midjourney to generate a piece I’m selling as a print”

Pure AI generation from a prompt = no US copyright. You can still sell prints (you have the right to use the work; no one else has exclusive rights either). But anyone else can also reproduce and sell the same image.

”I used Stable Diffusion to generate elements that I then composited and painted over in Photoshop”

Substantial human additions = copyright in the human additions. The Stable Diffusion elements remain unprotected; your composite work is protected for your specific creative additions.

”I used ChatGPT to draft a blog post, then heavily edited it”

The original ChatGPT output is not copyrightable. Your edits, additions, and creative refinements are copyrightable. The final work has copyright in the human contributions.

”I used an AI tool to enhance/upscale my own photograph”

Your original photograph remains your copyrighted work. The AI enhancement is generally treated as a tool, similar to other editing software. The enhanced version retains your copyright with the AI portions as derivative work additions.

”I trained an AI on my own original artwork to generate new variations”

You retain copyright in your training data. The generated variations are a more complex question. If they substantially copy your training works, copyright applies through that route. If they’re novel outputs with no direct copying, the standard AI generation rule applies.

What’s actually changing

Three things to watch as 2026 progresses:

Pending court decisions. Major lawsuits will produce decisions over the next 12-18 months that may significantly clarify both training and output questions.

Possible legislative action. Congress has discussed AI copyright legislation. Specific bills exist but have not advanced. The next session may produce action.

Industry standards. AI companies are introducing more sophisticated attribution and ownership systems. Standards for tracking AI involvement in works are emerging.

For working creators, the realistic strategy is to operate as if the current US position is the rule (substantial human creative contribution required for copyright), document your process clearly, and stay engaged with the evolving legal landscape.

For the related question of who owns AI-written content, see our companion piece.

The honest summary

AI-generated art is in a complicated legal moment. The current rule in the US is clear in principle (human authorship required) but messy in application (where does the human contribution begin?). The rule in other jurisdictions varies.

What working creators should do:

Use AI as a tool, not as the author. Substantive human creative contributions should be present in any work you intend to claim copyright on.
Document your creative process. Make your human contributions visible and defensible.
Register honestly. Disclose AI use when registering. Don’t try to claim copyright on AI-generated work without human contributions.
Stay informed. The law is evolving rapidly. What’s true in mid-2026 may not be true in 2027.
Keep traditional creative skills. AI tools augment human creativity. They don’t replace it for copyright purposes.

The creators who handle this well are those who use AI as one tool among many in a creative process that has substantial human authorship throughout. The creators who get into trouble are those who try to register pure AI generation as if it were their own work.

This is one of those moments where the law is catching up with technology. The current rules are clear enough to work with. The future rules may be different. The discipline of attribution and human-led creativity is what makes you robust to either outcome.

Can AI-Generated Art Be Copyrighted? Where the Law Stands in 2026