For agencies, consultancies and other businesses in the US and Europe that create work with AI, this guide explains who owns the output, when it is protected by copyright, when it can infringe, and what to promise clients.
What you'll learn
- The three different questions behind "who owns AI-generated content"
- When AI-assisted work is protected by copyright in the US, the EU and the UK
- What OpenAI, Anthropic and Microsoft business terms say about output
- When AI output can infringe someone else's copyright
- How to make AI-assisted work protectable, and what to put in client contracts
Who owns AI-generated content depends on which of three questions you ask. OpenAI's and Anthropic's business terms say the customer owns the output, as far as the law allows. Copyright law then decides what that is worth: the US Copyright Office concluded in January 2025 that AI output is protected only where a human determined sufficient expressive elements, and prompts alone are not enough. In March 2026 the US Supreme Court declined to review a ruling that an AI cannot be an author. This guide covers the US, the EU and the UK, when AI output can infringe someone else's work, and what to put in client contracts.
Agencies and consultancies now deliver work that AI helped write, design or code. Sooner or later a client asks who owns it, and the honest answer is less simple than "you do". It depends on whether you mean the contract with your AI provider, copyright law, or someone else's rights in what the AI produced.
This guide takes those three questions one at a time, for the US, the EU and the UK. It ends with what to put in client contracts so that you promise only what you can deliver.
Who owns AI-generated content?
You usually own it under your AI provider's terms, but copyright law may not protect it. OpenAI's business agreement, effective January 1, 2026, says the customer "owns all Output", to the extent permitted by law.[10] The US Copyright Office concluded in January 2025 that AI output is protected only where a human determined sufficient expressive elements.[1] Germany's justice ministry says the same for the EU: purely AI-based content has no copyright protection.[7]
Those statements do not contradict each other. A provider can promise not to claim your output and can hand you whatever rights it has. It cannot create a copyright that the law does not grant. Owning a text that nobody holds copyright in means you can use it freely, and so can anyone else who gets a copy.
| Question | Who decides | Short answer |
|---|---|---|
| Does the provider claim the output? | The provider's terms | No, under OpenAI's and Anthropic's business terms; the customer owns it[10][11] |
| Is the output protected by copyright? | Copyright law where you sell | Only the parts a human created, in the US and the EU[1][6][7] |
| Could the output infringe someone else's rights? | Copyright law, the courts | Yes, if it reproduces protected work[7][13] |
Most of the confusion about AI and ownership comes from mixing up these three rows. Each needs its own answer before you sign anything with a client.
Can AI-generated content be copyrighted in the US?
Only the parts a human created. The US Copyright Office concluded on January 29, 2025 that AI output can be protected "only where a human author has determined sufficient expressive elements", for example through creative arrangement or modification, "but not the mere provision of prompts".[1] Using AI as a tool, or including AI material in a larger human work, does not prevent protection.[1]
The courts agree on the principle. In March 2025, the US Court of Appeals for the D.C. Circuit held in Thaler v. Perlmutter that the Copyright Act "requires all eligible work to be authored in the first instance by a human being".[2] The Supreme Court declined to hear the case on March 2, 2026, which leaves that ruling in place.[3] The same court stressed that the rule does not bar protection for work "made by or with the assistance of artificial intelligence".[2]
The Copyright Office's decision on the comic book Zarya of the Dawn shows where the line runs in practice. In February 2023, the Office registered the author's text and her selection and arrangement of text and images, but not the images Midjourney generated.[4] The human parts were protected; the machine parts were not.
Can AI-generated content be copyrighted in the EU?
Not when it is purely AI-generated. EU copyright protects a work that is its author's own intellectual creation, "expressing his free and creative choices", as the EU Court of Justice put it in the Painer case in 2011.[6] Germany's Federal Ministry of Justice concluded in March 2024 that "purely AI-based content" has no copyright protection, because only human creations count.[7]
The ministry adds the case that matters for most businesses. A person can be the author of AI-assisted work when the software is only a tool and its role in creating the work is of secondary importance.[7] Whether enough human control over the form remains has to be decided case by case.[7]
The European Parliament took the same line in March 2026. In a resolution adopted by 460 votes to 71, MEPs said that "content fully generated by AI should not be protected by copyright".[8] A resolution is a recommendation to the European Commission, not law, but it shows the direction EU law-makers expect.
What about the UK?
The UK is the exception in Europe. Section 9(3) of the Copyright, Designs and Patents Act 1988 treats a computer-generated work as authored by "the person by whom the arrangements necessary for the creation of the work are undertaken".[9] Protection lasts 50 years from the end of the year the work was made, shorter than for human works.[9]
As of October 2026, the provision remains in force with no recorded pending changes.[9] How it applies to modern AI tools has not been settled in court. For a business selling into the UK, it means AI-generated output may carry a thinner, shorter copyright there than in the US or the EU, where it carries none.
What do OpenAI, Anthropic and Microsoft terms say about output?
All three leave the output to the business customer. OpenAI's Services Agreement, effective January 1, 2026, says the customer "owns all Output" and assigns OpenAI's rights in it "if any".[10] Anthropic's Commercial Terms say the customer "owns its Outputs" and assign Anthropic's rights.[11] Microsoft has said since September 2023 that it does not claim intellectual property in the output of its commercial Copilot services.[12]
| Provider (business terms) | Who owns the output | Defense if someone claims infringement |
|---|---|---|
| OpenAI Services Agreement | The customer, to the extent the law allows; OpenAI assigns its rights "if any"[10] | IP indemnity for the services, with exclusions, plus service-specific terms[10] |
| Anthropic Commercial Terms | The customer; Anthropic assigns its rights "if any"[11] | Defends paid customers against IP claims over outputs from authorized use, with exclusions[11] |
| Microsoft commercial Copilots | Microsoft claims no rights in the output[12] | Defends and pays judgments if a commercial customer is sued over Copilot output, if the built-in guardrails were used[12] |
The two words "if any" carry the point. The provider transfers its rights without promising that there are any. Consumer plans can have different terms, so check the plan your team actually uses. Whether a provider may train on what you send it is a separate question, covered in our guide to which AI providers train on your data.
Can AI output infringe someone else's copyright?
Yes. Output that reproduces a protected work infringes when you copy, publish or sell it without permission. In November 2025, the Munich Regional Court held that ChatGPT's reproduction of the lyrics of nine German songs infringed the songwriters' rights, and that the model's operators were liable for those outputs.[13] Germany's justice ministry states the general rule: where existing works are recognizable in AI output, using it needs the rightsholder's permission.[7]
The Munich case concerned a model reproducing lyrics almost word for word after simple prompts.[13] The court found that the songs were memorized in the model and that simple prompts did not shift responsibility to the user.[13] It was a first-instance decision about the provider's liability, not about a business that used the output. A business that publishes such output still makes its own copy, and the rightsholder can act against that.
You can own an AI output and still have no way to stop a competitor copying it.
The risk is highest where AI output is most likely to echo a known work: song lyrics, famous quotes, characters, logos, images "in the style of" a named artist, and code. Where nothing recognizable from an existing work remains in the output, the ministry notes that it can be used freely.[7] A quick search for distinctive phrases, and an image search for visuals, catches most close copies before they reach a client.
Can you trademark an AI-generated logo?
Often yes, because trademark law asks a different question from copyright. US federal law defines a trademark as a word, name, symbol or device used "to identify and distinguish" goods and indicate their source.[14] The EU trade mark regulation of 2017 asks whether a sign is capable of "distinguishing the goods or services of one undertaking from those of other undertakings".[15] Neither definition asks who or what designed the sign.
That makes trademark registration the main protection for an AI-generated logo, which as an image may carry no copyright in the US or the EU.[1][7] The usual tests still apply: the logo must be distinctive, and it must not be confusingly similar to a mark someone already uses. AI image tools tend to produce familiar-looking designs, so a clearance search before adopting one matters more, not less.
For agencies, the contract point follows. A client can register an AI-generated logo you designed as a trademark. A clause that assigns "all copyright" in it may add little, so say plainly which protection the client is getting.
How do you make AI-assisted work protectable?
Make the creative choices yourself and keep a record of them. The US Copyright Office protects human selection, arrangement and modification of AI material, and in February 2023 it registered the text and arrangement of Zarya of the Dawn while excluding the AI images.[1][4] In the EU, the question is whether the AI was a tool and the human kept control over the form.[7]
In practice, that means using AI for drafts and options, and having a person decide what stays, rewrite the important parts and arrange the whole. The more of the final form a person shaped, the stronger the claim. Keep the prompts, drafts and edits that show who decided what; they are your evidence if anyone disputes authorship.
Registering a copyright in the US adds one step. Since March 2023, applicants must disclose AI-generated content that is more than de minimis and exclude it from the claim, describing the human contribution instead.[5] The Office warns that applicants who fail to correct the record for AI-generated material "risk losing the benefits of the registration".[5]
The same human step also protects you elsewhere. Google's search rules treat mass-produced pages without human value as spam, as our guide on AI-written content and SEO explains.
What should go into client contracts for AI-assisted work?
Promise what you can deliver: the work, the rights you hold, and honesty about how it was made. A standard clause that assigns "all copyright" in deliverables may promise something that does not exist for the AI-generated parts.[1][7] These six points belong in contracts for AI-assisted work; a lawyer turns them into clauses for your jurisdiction.
Disclose AI use
Assign the rights you hold
Warrant only what you control
Match your provider's terms
Agree who carries infringement risk
Keep the record
Clients increasingly ask about AI in procurement questionnaires. A short, accurate answer about which parts are human-made, and how you check the rest, does more for trust than a broad assignment clause.
When does this matter for an AI agent?
Ownership of AI output matters as soon as an agent produces work that leaves your business: proposals, reports, articles, designs or code for clients. In the US and the EU, the agent's output is protected only through the human choices made around it.[1][7] An agent set up so that a person reviews and shapes every deliverable produces work you can stand behind and, where the human contribution is enough, protect.
That review step is also where quality is decided. Our guide on what human-in-the-loop means in practice covers how to keep it real when volume grows. For the choice between general AI tools and an agent built around your own material and approval steps, see what a custom agent is and when to build one.
Frequently asked questions
Who owns AI-generated content? Under the business terms of OpenAI and Anthropic, the customer owns the output, to the extent the law allows, and the provider assigns any rights it has. Whether that ownership includes copyright is a separate question: in the US and the EU, purely AI-generated material is not protected, because copyright requires a human author who made the creative choices. You own it, but you may not be able to stop others from copying it.
Can AI-generated content be copyrighted in the US or the EU? Only the parts a human created. The US Copyright Office concluded in January 2025 that AI output is protected only where a human determined sufficient expressive elements, for example through creative selection, arrangement or modification, and that prompts alone are not enough. In the EU, a work must be its author's own intellectual creation, expressing free and creative choices; Germany's justice ministry states that purely AI-based content has no copyright protection.
Can I sell AI-generated work to a client? Yes, under most business terms, but be careful what you promise. You can hand over the work and assign whatever rights you hold, but you cannot assign a copyright that does not exist. Tell the client which parts were AI-generated, warrant originality only for the human-created parts, and check that nothing in the output reproduces someone else's protected work.
Can I be sued for using AI-generated content? Yes, if the output reproduces someone else's protected work and you publish or sell it. In November 2025, a Munich court held that ChatGPT's reproduction of song lyrics infringed copyright and that the model's operator was liable for those outputs. Some providers, including Anthropic and Microsoft, promise to defend paying business customers against such claims, subject to conditions.
Notes
- U.S. Copyright Office, Copyright Office Releases Part 2 of Artificial Intelligence Report, NewsNet Issue 1060, January 29, 2025, accessed October 16, 2026.
- Thaler v. Perlmutter, No. 23-5233, U.S. Court of Appeals for the District of Columbia Circuit, decided March 18, 2025, accessed October 16, 2026.
- Supreme Court of the United States, Docket No. 25-449, Thaler v. Perlmutter, petition denied March 2, 2026, accessed October 16, 2026.
- U.S. Copyright Office, letter re Zarya of the Dawn (Registration # VAu001480196), February 21, 2023, accessed October 16, 2026.
- U.S. Copyright Office, Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence, 88 FR 16190, effective March 16, 2023, accessed October 16, 2026.
- Court of Justice of the European Union, Case C-145/10, Painer, judgment of December 1, 2011, paragraphs 89 to 94, accessed October 16, 2026.
- German Federal Ministry of Justice, Künstliche Intelligenz und Urheberrecht: Fragen und Antworten, March 2024, in German, accessed October 16, 2026.
- European Parliament, Protecting copyrighted work and the EU's creative sector in the age of AI, press release, March 10, 2026, accessed October 16, 2026.
- Copyright, Designs and Patents Act 1988, section 9, and section 12, legislation.gov.uk, accessed October 16, 2026.
- OpenAI Services Agreement, updated December 1, 2025, effective January 1, 2026, sections 4.1 and 13, accessed October 16, 2026.
- Anthropic, Commercial Terms of Service, sections on customer content and indemnification (K), accessed October 16, 2026.
- Microsoft, Microsoft announces new Copilot Copyright Commitment for customers, September 7, 2023, expanded as the Customer Copyright Commitment on November 15, 2023, accessed October 16, 2026.
- Landgericht München I, judgment of November 11, 2025, 42 O 14139/24 (GEMA v OpenAI), in German, accessed October 16, 2026.
- 15 U.S. Code § 1127, Construction and definitions, definition of "trademark", Legal Information Institute, accessed October 16, 2026.
- Regulation (EU) 2017/1001 on the European Union trade mark, Article 4, accessed October 16, 2026.
