Can you copyright AI-generated text or images?

Not if the content is purely AI-generated with no human creative input. The US Copyright Office has consistently ruled since 2023 that copyright requires human authorship, and purely AI-generated content fails that test. However, if a human made creative choices in the prompt, selected from multiple AI outputs, arranged AI-generated elements alongside human-authored material, or substantially edited the AI output, the human-authored portions and selections may be protectable. The key is documenting what the human actually decided.

What did the Thaler v. Perlmutter case decide?

The US District Court for the District of Columbia held in August 2023 that an artwork generated entirely by an AI system without human creative input could not be registered with the US Copyright Office. The court reaffirmed that copyright protection under US law requires human authorship. The ruling aligned with the Copyright Office's prior policy guidance and has not been reversed on appeal. It does not address AI-assisted work where humans made substantial creative choices; it applies to purely autonomous AI generation.

What copyright does OpenAI or Anthropic claim over AI outputs?

OpenAI's terms of service assign ownership of outputs to the user, subject to the terms of use. Anthropic takes a similar position. Google's Gemini API terms also generally assign output ownership to the user. However, vendor terms do not create copyright protection where none exists under law. The vendor's assignment of rights only matters if those rights exist in the first place. For purely AI-generated content, there may be no copyright to assign regardless of what the vendor's terms say.

Does the EU have a different position on AI copyright?

The EU has not enacted a specific AI copyright statute, but the prevailing interpretation under EU copyright law (InfoSoc Directive and national implementing laws) also requires human authorship. The European approach treats AI as a tool; the human operating that tool may claim copyright if they made genuine creative choices. The EU AI Act's requirements for GPAI model transparency about training data are separate from copyright ownership questions. Several EU member state courts and the EU Commission have confirmed the human authorship requirement.

How do I document human authorship for AI-assisted content?

Keep records of the creative choices you made. Save your prompt iterations and note what you were trying to achieve with each. Record which of multiple AI outputs you selected and why. Document edits you made to AI text, including the original output and your revised version. For images, note selection criteria and any post-processing. This documentation does not need to be elaborate, but it should be able to demonstrate that a human exercised genuine creative judgment, not just typed a one-line prompt and accepted the output unchanged.

What are the implications for employees who create AI-assisted content at work?

Work-for-hire doctrine means that copyrightable content created by an employee within the scope of employment is owned by the employer. For AI-assisted content where the human contribution is sufficient for copyright protection, the same rule applies: the employer owns it. For purely AI-generated content, there is nothing to own regardless of who created it. Contractors are different: copyright defaults to the contractor unless there is a written work-for-hire agreement. Review contractor agreements to confirm they assign AI-assisted output ownership clearly.

Generative AI copyright ownership 2026: who…

Person writing at a desk with a notebook and pen

TL;DR: Purely AI-generated content is not copyrightable in the US or EU. What matters is human authorship: the choices a person makes in prompting, selecting, arranging, and editing AI outputs. Document those choices as you work. Check your vendor contracts for output ownership clauses. For commercial content, plan your human contribution before you start generating.

The question of who owns AI-generated content matters more now than it did two years ago. Organizations are producing marketing copy, product documentation, code, images, and even books using generative AI tools. The legal framework for protecting that output is not what most people assume.

The short version: under both US and EU copyright law, content generated by AI without meaningful human creative input is in the public domain. Anyone can copy it, sell it, or use it without permission. But the line between "AI-generated" and "human-authored with AI assistance" is real and navigable. This guide explains where that line sits, what the major cases say, what your vendor contracts cover, and how to document the human contribution that makes protection possible.

The baseline ruling: what the US Copyright Office decided

The US Copyright Office began issuing guidance on AI-generated content in 2023 after receiving registration applications for AI-produced works. Its position has been consistent across several decisions and its formal guidance documents.

Copyright in the United States requires human authorship. The Copyright Act's protection extends to "original works of authorship," and "authorship" has been interpreted by courts for over a century to require human creativity. The Copyright Office applied this principle directly to AI-generated content.

In its February 2023 guidance and subsequent decisions, the Copyright Office held that content generated by AI without human creative input cannot be registered or protected. This is not a close question under current law. The Copyright Office's position is that when a user types a prompt and an AI system produces content autonomously, there is no human author in the legal sense, because the expressive choices were made by the machine, not the person.

The Office acknowledged that AI-assisted works are different. When a human makes genuine creative choices, including through the process of prompting, selecting, and editing, the human-authored and human-selected portions may qualify for protection. The challenge is that the AI-generated portions remain unprotectable, and the human must be able to identify what they contributed.

The Thaler v. Perlmutter decision

The August 2023 decision in Thaler v. Perlmutter directly addressed the human authorship requirement. Stephen Thaler sought to register a copyright in an image generated entirely by his DABUS AI system, listing the AI as the author and himself as the owner. The Copyright Office refused, and Thaler challenged that refusal in federal court.

The US District Court for the District of Columbia affirmed the Copyright Office's position. The court held that human authorship is a bedrock requirement of copyright law in the United States, pointing to the Constitutional and statutory basis for copyright protection as incentivizing human creative expression. A machine cannot be an author. An image generated by a machine without human creative input cannot be copyrighted.

The practical implication for businesses is clear: if you generate an image, a piece of text, or code using an AI tool and you do not make substantial creative choices in the process, you have no copyright in the output. Your competitors can copy it freely.

What you CAN protect

The harder and more practically important question is what qualifies for protection when humans and AI tools work together.

Prompting with genuine creative direction can contribute to protectable human authorship if the prompts reflect original creative choices about expressive elements. A prompt specifying "a melancholy winter landscape in the style of Andrew Wyeth, with a solitary farmhouse and a woman in a red coat viewed from behind at dusk" involves specific creative decisions about composition, mood, and reference. The Copyright Office has indicated that extensive, specific creative prompting can be evidence of human authorship, though the output remains AI-generated.

Selection among multiple outputs is potentially protectable. If you generate 50 AI images and select one because of particular compositional choices it makes, the selection and arrangement may reflect human authorship in the same way that a photographer selects from many shots. Document what you chose and why.

Arrangement and compilation is a well-established copyright category. A collection of AI-generated and human-authored elements arranged with human creative judgment can create a protectable compilation, even if the individual AI-generated pieces are not independently protectable.

Substantial human editing creates protectable expression. If you take AI-generated text and substantially rewrite it, your rewriting is protectable even if the underlying AI output is not. The practical challenge is that you must be able to separate your contribution from the AI's.

Works where AI assistance is incidental to substantial human authorship retain full protection. A human author who writes 90% of an article and uses AI to suggest a few sentence revisions has a fully protectable work.

What you CANNOT rely on

Several common assumptions about AI copyright are wrong.

You cannot assume that AI-assisted content is automatically protectable just because you were involved in creating it. Minimal involvement, such as typing a brief prompt and accepting the output unchanged, does not create copyright protection. The human contribution must be genuinely creative, not merely mechanical.

You cannot rely on your AI vendor's terms of service to create copyright protection. OpenAI's terms assign output ownership to users. Anthropic and Google do the same. But a contractual assignment of rights only functions if the rights exist. If there is no copyright in AI-generated output under law, there is nothing for the vendor to assign. The vendor's terms govern the contractual relationship between you and the vendor; they do not create intellectual property rights under copyright law.

You cannot use registration as a substitute for actual protectable authorship. The Copyright Office will reject applications for purely AI-generated content. Attempting to register by concealing AI involvement would be fraudulent and could invalidate any registration obtained.

The EU position

The EU has not enacted specific AI copyright legislation, but its approach is broadly consistent with the US human authorship requirement, with some differences in how human contribution is assessed.

Under EU copyright law, works are protected if they reflect the author's own intellectual creation. This formulation, established by the Court of Justice of the EU in cases like Infopaq and Painer, requires that the work involves free and creative choices by the human author. AI-generated content without human creative choices fails this test.

The EU AI Act, which is in phased application as of 2026, requires providers of general-purpose AI models to make publicly available a summary of training data that may be subject to copyright. This requirement addresses the input side (what was used to train the model) rather than the output side (who owns what the model produces). These are separate legal questions.

The EU approach treats AI as a tool. The human operating the tool may claim copyright if they made free and creative choices in using it. This is substantively similar to the US approach, though EU courts have been somewhat more flexible in recognizing human authorship in AI-assisted work where the human's creative intent is evident.

IP clauses in AI vendor contracts

Understanding what your AI vendor's terms say about output ownership is important for commercial work, even given the limitations discussed above.

OpenAI currently assigns ownership of outputs to users, subject to compliance with usage policies. Users can claim and enforce rights in outputs to the extent permitted by applicable law. OpenAI does not claim ownership of outputs.

Anthropic takes a similar position. Users own their outputs, and Anthropic does not claim ownership of content generated through its API or products.

Google (Gemini API) also assigns output ownership to users, though the terms differ between the API and consumer products. Review your specific agreement.

One area where vendor terms matter regardless of copyright questions: indemnification for third-party IP claims. Some vendors (including OpenAI with its Copyright Shield program for certain enterprise customers) offer indemnification if a third party claims their copyrighted training data appears in outputs. Others do not. If you are producing commercial content at scale, review the indemnification terms carefully.

Work-for-hire implications

Work-for-hire doctrine affects who holds whatever rights exist in AI-assisted content.

For employees, content created within the scope of employment is owned by the employer as a work made for hire. This applies to protectable AI-assisted content the same way it applies to traditional employee-created content. If your employee uses AI tools to create marketing materials that involve sufficient human authorship, your company owns the copyright.

For contractors, the default rule is different. Copyright vests in the contractor unless the work falls within one of the statutory categories of works that can be made for hire by contract, and there is a written agreement. Marketing copy does not fall within the statutory categories; a written work-for-hire agreement is required to transfer ownership to the commissioning party. Review all contractor agreements to confirm that AI-assisted output is addressed.

For KDP authors and content creators: if you self-publish or create content for commercial licensing, your ability to license and enforce rights depends on whether you can document human authorship. Amazon's KDP platform now requires disclosure of AI-generated content, but separate from disclosure, your ability to enforce rights against infringers depends on copyright protection. Plan your human contribution before you generate.

Practical documentation framework

If you produce AI-assisted content for commercial purposes, build documentation habits from the start.

Keep a record of your prompts, including iterations and the creative intent behind them. Note what alternatives the AI generated and which you chose and why. Track substantial edits you made to AI outputs. For image work, note post-processing choices. This does not require elaborate systems: a simple log file or a folder with versioned outputs and brief notes is often sufficient.

When registering copyright in AI-assisted works, be prepared to identify the human-authored elements specifically and to submit only those elements for registration if the Copyright Office requires it. The Copyright Office's 2024 registration guidance addresses how to handle this on applications.

For high-value commercial content, consult IP counsel before assuming protectability. The law is developing and the specific facts of how the work was created matter to the analysis.

Person writing at a desk with a notebook and pen

TL;DR: Purely AI-generated content is not copyrightable in the US or EU. What matters is human authorship: the choices a person makes in prompting, selecting, arranging, and editing AI outputs. Document those choices as you work. Check your vendor contracts for output ownership clauses. For commercial content, plan your human contribution before you start generating.

The baseline ruling: what the US Copyright Office decided

The Thaler v. Perlmutter decision

What you CAN protect

The harder and more practically important question is what qualifies for protection when humans and AI tools work together.

What you CANNOT rely on

Several common assumptions about AI copyright are wrong.

The EU position

The EU has not enacted specific AI copyright legislation, but its approach is broadly consistent with the US human authorship requirement, with some differences in how human contribution is assessed.

IP clauses in AI vendor contracts

Understanding what your AI vendor's terms say about output ownership is important for commercial work, even given the limitations discussed above.

Anthropic takes a similar position. Users own their outputs, and Anthropic does not claim ownership of content generated through its API or products.

Google (Gemini API) also assigns output ownership to users, though the terms differ between the API and consumer products. Review your specific agreement.

Work-for-hire implications

Work-for-hire doctrine affects who holds whatever rights exist in AI-assisted content.

Practical documentation framework

If you produce AI-assisted content for commercial purposes, build documentation habits from the start.

For high-value commercial content, consult IP counsel before assuming protectability. The law is developing and the specific facts of how the work was created matter to the analysis.

Generative AI copyright ownership 2026: who owns AI-generated content

The baseline ruling: what the US Copyright Office decided

The Thaler v. Perlmutter decision

What you CAN protect

What you CANNOT rely on

The EU position

IP clauses in AI vendor contracts

Work-for-hire implications

Practical documentation framework

Generative AI copyright ownership 2026: who owns AI-generated content

The baseline ruling: what the US Copyright Office decided

The Thaler v. Perlmutter decision

What you CAN protect

What you CANNOT rely on

The EU position

IP clauses in AI vendor contracts

Work-for-hire implications

Practical documentation framework