AICode ownershipContracts
Who owns AI-generated code?
9 min read
Code that an AI writes on its own most likely has no copyright owner. In the United States, the Copyright Office and the courts require a human author, and in the European Union software is protected only when it reflects its author's own intellectual creation. What you actually own is the human part (what someone designed, selected, arranged or reworked) plus whatever your contracts transfer to you. AI providers' terms assign you their rights "if any", which does not settle the question.
This article summarises official sources as of 3 November 2026. It is not legal advice: the law is moving quickly here, and a specific project should be reviewed by a lawyer in your jurisdiction.
United States: the clearest position
On 29 January 2025 the US Copyright Office published Part 2 of its report on copyright and AI, which deals with the copyrightability of AI outputs. Its press release sums up three conclusions:
- using AI as a tool in the creative process does not prevent protection;
- merely providing prompts does not, with current technology, make the user the author of the output;
- protection is available where a human-authored work is perceptible in the output, and for creative arrangements or modifications a human makes to it.
The courts agree. In Thaler v. Perlmutter, the US Court of Appeals for the D.C. Circuit upheld the human authorship requirement in March 2025, and the Supreme Court declined to hear the case on 2 March 2026. A denial is not a ruling on the merits, but the appeals court decision stands.
For code, that means a file generated wholesale from a prompt is probably not protected, while the architecture you designed, the functions you rewrote or the way you selected and assembled the pieces may well be.
United Kingdom: an exception on its way out
UK law has long been the odd one out. Section 9(3) of the Copyright, Designs and Patents Act 1988 protects computer-generated works with no human author, and treats as the author the person by whom the arrangements necessary for the work's creation are undertaken. However, in a report published on 18 March 2026, the UK government proposes removing that protection for wholly computer-generated works, while keeping copyright for AI-assisted works where a human makes creative choices. As of this writing, no legislation has changed section 9(3).
European Union: the originality test
EU law has no rule written specifically for AI output, so you start from general principles. The Software Directive 2009/24/EC says, in Article 1, that a computer program is protected if it is original "in the sense that it is the author's own intellectual creation". The Court of Justice of the EU links originality to the author's free and creative choices (Painer, C-145/10, 2011). National laws such as the French Intellectual Property Code apply the same test.
Output produced by a machine, with no identifiable human choice, is therefore unlikely to be considered original. To our knowledge, no EU or national court has yet ruled on AI-generated code specifically, so caution applies in both directions.
| United States | United Kingdom | European Union | |
|---|---|---|---|
| What the AI produces alone | Not protected (Copyright Office, Thaler) | Protected under s.9(3), removal proposed | Protection very doubtful |
| Creative human contribution | Protectable | Protectable | Protectable if original |
| Prompts only | Not enough | Open question | Not enough in principle |
What AI providers' terms say
Major providers do not claim the code produced for you. Anthropic's commercial terms, for example, state that the customer owns its outputs and assign the customer the provider's right, title and interest in them "(if any)". That caveat is honest and important: a provider can only assign what it holds. If copyright never arises in output generated without human authorship, the assignment creates nothing.
Read the terms of the tool you use, in their dated version, and look for three things: who owns the outputs, whether your data may be used to train models, and what protection, if any, the provider offers if a third party makes a claim.
The second risk: other people's code
Ownership hides a second question: generated code may resemble existing code. A model trained on public code can reproduce a snippet close to an open source project, along with that project's licence. Some licences, such as the GPL, impose obligations when you distribute the software.
Some tools help. GitHub Copilot offers code referencing: depending on the policy, a suggestion that matches public code is either blocked or shown with the matching repositories and their licence, where found. That is useful but partial: it covers certain suggestions and the public code it has indexed, not everything.
The remedy is the same as for any software: an inventory of components and their licences, and a review of long or unusual snippets. Our guide to who owns the source code of custom software explains the three layers of any application: your specific code, the developer's reusable building blocks and open source.
Three common situations
You generated the app yourself. You own at least the parts you genuinely designed or reworked. For the rest, your best protection is not copyright but confidentiality: a private repository, limited access, and non-disclosure agreements with anyone who touches it.
Your employee uses an AI assistant. Under US law, work created by an employee within the scope of employment is a work made for hire owned by the employer, and the UK and EU have comparable rules for software. That covers the protectable part of their work, AI-assisted or not. Set an internal policy on approved tools and on what data may be shared with them.
Your development partner uses AI. That is now the norm, and it is not a problem in itself. What matters is that the contract says so and covers the points below.
How to document the human contribution
If ownership rests on the human part, keep evidence of it as you go. You do not need a special tool; ordinary engineering habits already leave a trail:
- Design documents and decisions written before the code: data model, architecture, business rules.
- Version history showing who changed what, with commit messages that say why, and code reviews that record what a person rejected or rewrote.
- A note of the AI tools used, per project or per period, with the version of their terms.
None of this guarantees protection, but it turns a vague claim into something you can show.
What a development contract should cover
- Disclosure of AI use, naming the tools and their terms of use.
- Confidentiality: no sensitive data, secrets or confidential code sent to a tool without written consent, and tools whose terms exclude training on your data.
- A licence warranty: the developer warrants that included components are used in line with their licences and provides the list.
- An assignment or licence covering everything protectable in the deliverable, however it was written.
- Human review: delivered code has been read, tested and understood by someone who can explain and maintain it.
These come on top of the clauses covered in software development contract: the clauses to check. If you are weighing an AI app builder against a studio in the first place, AI app builders vs a software studio sets out where the line sits.
Where LPR Studio stands
We use AI to write code, with one simple rule: every change is reviewed, tested and approved by a person before it reaches production. That review is first about quality, but it also documents the human part of the work.
On ownership, our answer stays deliberately cautious: the demonstration prototype remains LPR Studio's property, and beyond that, rights are defined case by case in the contract, depending on the model you choose. We do not promise legal protection that the law itself does not yet guarantee.
Where to start
If your product already relies on generated code, do three things: list the tools used and their terms, get the open source components inventoried, and identify the parts you want to be able to defend (algorithms, business rules, architecture) to make sure they carry a real human contribution. For a new project, raise the question of rights in the first conversation: at LPR Studio, we explain what each model transfers to you when we present your free prototype, in a video call within 5 business days.