Who Owns AI-Generated Content? Copyright Basics in the US

Your marketing team generated a campaign image with AI. Your product team shipped onboarding copy that an assistant drafted. A competitor copies both. Can you stop them? The answer depends on how much of the work came from a person, and US law on this is now clearer than it was 2 years ago.

This guide summarizes what the US Copyright Office and the courts have said about AI-generated content copyright, and turns it into practical rules for marketing, content and product teams. This is not legal advice. For a specific work or dispute, talk to an intellectual property attorney.

The short answer: copyright needs a human author

US copyright protects original works of authorship, and the Copyright Office and federal courts read “author” to mean a human being. In Thaler v. Perlmutter, decided March 18, 2025, the US Court of Appeals for the D.C. Circuit upheld the Copyright Office’s refusal to register an artwork that its creator said was made autonomously by an AI system, holding that the Copyright Act requires human authorship. The Supreme Court declined to hear the case on March 2, 2026, according to its docket, so that ruling stands.

That does not mean anything touched by AI is unprotected. It means the protection covers only what a person contributed.

What the US Copyright Office says

The Copyright Office has published a series of reports and guidance on AI, collected on its Copyright and Artificial Intelligence page. The 2 most important documents for businesses are:

Registration guidance (March 2023)

The Office’s registration guidance for works containing AI-generated material, published March 16, 2023, says applicants must disclose AI-generated content that is more than de minimis and exclude it from the claim, while claiming the human-authored parts. The Office has applied this in decisions such as Zarya of the Dawn (February 2023), where it registered a comic book’s human-written text and the human selection and arrangement of images, but not the individual AI-generated images.

Part 2 report on copyrightability (January 2025)

The Part 2 report, released January 29, 2025, reached several conclusions:

• Existing law is enough. Copyrightability questions about AI can be resolved under current law, and the Office did not recommend new legislation.
• Prompts alone are not enough. Based on current technology, prompts work as instructions that convey ideas, and even detailed prompts do not give the user enough control over how the output is expressed.
• AI as a tool is fine. Using AI to assist human creativity, for example to brainstorm, outline or edit, does not by itself prevent protection.
• Human selection and arrangement can count. A person who creatively selects, coordinates and arranges AI material, or makes meaningful modifications, can claim copyright in those contributions.
• Case by case. Whether a human contribution is enough has to be judged work by work.

A separate Part 3 report on using copyrighted works to train AI was released as a pre-publication version on May 9, 2025. Training data and fair use remain the subject of ongoing lawsuits.

Key takeaway: You own the copyright in what you create, not in what the machine creates. The more meaningful human authorship you add, and the better you document it, the stronger your position.

How much human input is enough? Common scenarios

These are general patterns based on Copyright Office guidance, not guarantees.

Copyright vs ownership under your AI vendor’s terms

Copyright is not the only kind of “ownership.” Many AI vendors’ terms of service say that, as between you and the vendor, you own or may freely use the output. That is a contract right. It does not make the output copyrightable, so it cannot stop a third party from copying it.

Some enterprise plans also offer intellectual property protection. Microsoft, for example, announced its Customer Copyright Commitment in 2023, promising to defend eligible commercial customers against copyright claims over certain Copilot output when they use the required guardrails. Commitments like this typically exclude free and consumer products and come with conditions, so read the current terms for any tool you rely on.

What this means for marketing teams

• Logos and brand assets. Have a designer create or substantially rework anything you need to own and enforce, such as logos, mascots and signature visuals. Trademark law can still protect a logo used in commerce, but copyright in purely AI-generated art is unavailable.
• Campaign content. AI-generated social images and ad variations are fine to use, but assume competitors can copy them. Your protectable value is in the human-written copy, the creative concept and the overall arrangement.
• Infringement risk runs both ways. AI output can resemble existing works. Avoid prompts that name living artists or copyrighted characters, and run reverse image searches on important visuals.
• Disclosure. Some platforms and state laws require labels on AI-generated or altered content, especially political ads and realistic depictions of people. Our guide on how to spot a deepfake explains why.

What this means for product and engineering teams

• AI-written code. Protection for code an assistant produces with little human input is uncertain. Your competitive advantage is more likely in architecture, data and trade secrets than in copyright on individual functions. See how teams use these tools in our guide to AI coding assistants.
• Open-source licenses. Generated code can resemble licensed code. Use tools with duplicate-detection or filtering features where available, and follow your license review process.
• Product copy and docs. Keep a human editor in the loop for content that you want to protect and to catch errors.
• Contracts. Update agency, freelancer and customer contracts to address AI use, who owns what and what happens if output cannot be copyrighted.

Where to start: 6 steps for your team

1. Decide which assets must be ownable. Logos, flagship campaigns, books, courses and core product content usually need strong human authorship.
2. Set an AI content policy. Add rules on AI use, review and disclosure to your AI acceptable use policy.
3. Document human contributions. Keep drafts, edit history and notes showing what people wrote, changed, selected and arranged.
4. Disclose properly when registering. When filing for copyright, describe the human authorship and exclude AI material that is more than de minimis.
5. Check vendor terms. Confirm output rights, data use and any IP indemnity in the plan you actually pay for.
6. Review contracts. Make sure agencies and freelancers tell you when they use AI and guarantee the rights you need.

For a wider view of how AI is used in small companies, see our generative AI for small business guide.

Conclusion

US law draws a clear line: copyright protects human creativity, not machine output. You can use AI freely, but if you need to own and enforce a piece of content, make sure a person contributes real authorship and keep a record of it. Pair that with good vendor terms and clear contracts, and your team can move fast with AI without giving away the assets that matter.

Sources

• US Copyright Office, Copyright and Artificial Intelligence — index of AI reports, guidance and decisions
• US Copyright Office, Part 2: Copyrightability report (January 2025) — prompts, human authorship and selection and arrangement
• Federal Register, Copyright Registration Guidance: Works Containing Material Generated by AI (March 16, 2023) — disclosure and exclusion of AI material
• D.C. Circuit decision in Thaler v. Perlmutter — human authorship requirement, March 18, 2025
• Supreme Court docket, Thaler v. Perlmutter, No. 25-449 — certiorari denied March 2, 2026
• Microsoft, Copilot Copyright Commitment announcement — example of a vendor IP commitment and its conditions