AI-generated work and copyright: who owns what an agent makes

When an AI agent writes the report, designs the image or drafts the code, a practical question follows: who owns the result, and can anyone own it at all? The answer matters for anyone whose work product is now partly made by AI. This page summarises the clearest official position available, from the US Copyright Office, and notes where the EU AI Act touches the subject. It is a plain-language overview, not legal advice, and copyright law differs between countries.

The short answer: in the United States, copyright protects human authorship. Purely AI-generated material is not protected, and according to the US Copyright Office, prompts alone do not currently give enough control to make someone the author. What a person contributes, such as their own expression, their creative selection and arrangement of material, or their creative changes to the output, can be protected. The more of the expressive work a person actually does, the stronger their claim.

The US Copyright Office's conclusions

In January 2025 the US Copyright Office published Part 2 of its report on copyright and artificial intelligence, on copyrightability (US Copyright Office, Copyright and AI). Its main conclusions include (Copyright and AI, Part 2: Copyrightability, 2025):

• Questions of copyrightability and AI can be resolved under existing law, without legislative change.
• Using AI tools to assist rather than stand in for human creativity does not affect the availability of copyright protection for the output.
• Copyright protects the original expression in a work created by a human author, even if the work also includes AI-generated material.
• Copyright does not extend to purely AI-generated material, or material where there is insufficient human control over the expressive elements.
• Whether human contributions are sufficient for authorship must be analyzed case by case.
• Based on how current generally available technology works, prompts do not alone provide sufficient control.
• Human authors are entitled to copyright in their own works that are perceptible in AI outputs, and in the creative selection, coordination or arrangement of material in the outputs, or creative modifications of them.

The Office says it will keep monitoring technology and the law to decide whether any of these conclusions should be revisited.

Prompts are not authorship, yet

The point most relevant to digital collar work is the one about prompts. Writing a good brief is skilled work, but the Office's view is that, with current generally available tools, a prompt alone does not give the person enough control over the expressive elements of the result to be its author. In its words, prompts alone are, at this stage, unlikely to meet the requirements; outputs can be copyrightable in whole or in part where AI is used as a tool and a human has been able to determine the expressive elements they contain (same report, Conclusion).

Where human contribution counts

Under the Office's conclusions, protection can attach to:

• Your own material in the output: text, images or code you wrote that remain perceptible in the result.
• Your selection and arrangement: the creative choice and ordering of AI-generated pieces into a larger work.
• Your modifications: creative changes you make to what the AI produced.

In practice, the work a digital collar worker does after the agent produces a draft, cutting, rewriting, combining and shaping it, is also the work most likely to carry protection.

Registering work that includes AI material

The report recalls the Office's 2023 registration guidance: if a work contains more than a minimal amount of AI-generated material, the applicant should disclose it and briefly describe the human author's contribution. Since that guidance, the Office has registered hundreds of works that include AI-generated material, with the registration covering the human contribution (Copyright and AI, Part 2).

The EU angle

The EU AI Act does not decide who owns AI output, but it sets copyright-related duties for providers of general-purpose AI models. Since 2 August 2025 they must put in place a policy to comply with EU copyright law, including identifying and respecting rights holders' reservations of rights against text and data mining, and publish a sufficiently detailed summary of the content used for training, using a template from the AI Office (EU AI Act, Article 53(1)(c) and (d)). Separate transparency rules require certain AI-generated content to be marked or disclosed (Article 50). Whether a particular AI-assisted work is protected in an EU country is a question for that country's copyright law and courts.

Risks on the other side: using others' work

Ownership is half the picture. NIST lists intellectual property among the risks of generative AI: if a model's training data included copyrighted material, outputs that reproduce memorised training data could infringe copyright, and the status of content that is similar to protected work without strictly copying it is still being debated (NIST AI 600-1). Before publishing or shipping AI-assisted work, check that it does not reproduce someone else's text, images, code or trademarks.

Practical steps for teams

• Keep records of who did what: the brief, the AI output, and the human edits, selection and additions.
• Do real creative work on outputs that matter, especially anything you want to protect or license.
• Read your AI tool's terms on output ownership, use of your inputs and any indemnities.
• Check contracts with clients and employers for clauses on AI use and ownership of deliverables.
• Screen outputs for copied or closely matching third-party material before release.
• Disclose AI material where registration rules, contracts or transparency rules require it.