Who owns AI-generated code or content?
By SuLe · Updated 2 June 2026
There is no clean statutory rule that hands you copyright in AI-generated code or content — ownership works on two layers: what copyright law will protect, and what the AI provider's terms give you by contract. In practice the major API providers assign output rights to you, but the copyright itself can be thin where little human work went in.
Key facts
- UK law protects "computer-generated works" with no human author under CDPA 1988 s.9(3); protection lasts 50 years and belongs to the person who made the arrangements for creation.
- How s.9(3) applies to modern generative AI is untested, and the provision has been under review in the UKIPO's AI/copyright consultation.
- Ownership of AI output is governed by the provider's terms — major API providers, including OpenAI and Anthropic, assign their rights in outputs to the customer.
- AI output receives weaker copyright protection where human authorship is thin; your own selection, editing and arrangement strengthen your position.
- An AI output can resemble third-party copyrighted material or embed licensed code patterns, so the ownership question sits alongside an infringement risk.
Does anyone actually own AI-generated code or content?
The honest answer is that it is unsettled, and you should treat it that way rather than assuming a tidy yes. Ownership of AI output depends on two separate questions that people often blur together.
The first is a copyright question: does the law recognise a protected work at all, and who is its author? The second is a contract question: whatever rights exist in the output, has the AI provider given them to you? You can win on the contract layer and still hold a weak copyright.
For a startup, that means your strongest ownership claim comes from combining the two: a provider that assigns output rights to you, plus enough human input that a real copyright can arise.
What does UK copyright law say about AI outputs?
UK law is unusual: under the Copyright, Designs and Patents Act 1988, section 9(3) recognises "computer-generated works" that have no human author. The author is deemed to be the person who made the arrangements necessary for creation, and protection lasts 50 years.
That provision predates generative AI by decades, and how it maps onto a modern model prompted by a user is untested in the courts. It has been under review in the UKIPO's AI and copyright consultation, so the position may change.
Where a human meaningfully selects, prompts, edits and arranges the output, you can instead argue ordinary copyright in the human-authored result — a far more comfortable footing than relying on s.9(3) alone.
Who owns the output under OpenAI's or Anthropic's terms?
In day-to-day practice, ownership of AI output is governed by the provider's terms of service, not by a statutory grant. This is the contract layer, and it usually does the heavy lifting.
The major API providers, including OpenAI and Anthropic, assign their rights in the outputs to you, the customer. That transfers whatever rights the provider holds — but it cannot create copyright that never existed, and it does not warrant that the output is free of third-party rights.
Because these are contractual terms, they can differ by product and by tier, and they change. Check the current terms of the specific service and plan you use before you rely on owning anything you generate on it.
What are the risks of relying on AI output in your product?
Two risks matter most. An AI output may be similar to third-party copyrighted material, and generated code may embed licensed code patterns drawn from training data — either of which can expose you to an infringement claim when you ship.
The second risk is thin protection. Where human authorship is minimal, the copyright you can assert against a copycat is correspondingly weak, which undercuts the very ownership you were trying to secure.
The practical fix is human involvement: real selection, editing and arrangement both strengthens your copyright and gives you a chance to catch outputs that look too close to someone else's work. This is closely related to who owns copyright in software in the UK.
| Human-authored work | AI output with substantial human editing | Pure one-shot AI output | |
|---|---|---|---|
| Likely author in law | The human creator | The human, as editor/arranger | Possibly s.9(3) "arrangements" maker — untested |
| Strength of copyright | Strong | Reasonable | Weak / uncertain |
| Ownership by contract | N/A | Provider terms assign output rights | Provider terms assign output rights |
| Protection term | Life + 70 years | Life + 70 years (human parts) | 50 years under s.9(3) if it applies |
| Infringement risk | Normal | Review before shipping | Highest — review before shipping |
How do I strengthen my ownership of AI-assisted work?
Do human work on top of the machine's, and keep a record of it. Selection, editing and arrangement are what turn a raw output into something a court can more readily treat as your protected work.
Get your contracts right too. If a contractor or agency generated code or content with AI, you still need a written IP assignment from them — the tool does not change who must sign it. See who owns the IP my employees and contractors create.
Finally, run a light review before shipping generated code, especially anything that looks like a well-known implementation, so an embedded licensed pattern does not travel into your product unnoticed.
Worked example
Maya runs a Bristol healthtech startup and uses an AI coding assistant to draft a scheduling module and its onboarding copy. Under the provider's business-tier terms, the provider assigns its rights in the outputs to her company, so the contract layer is covered.
Her engineers then restructure the generated code, write the test suite and rework the copy — real human selection and arrangement — which strengthens the copyright she can assert. She pays a freelance developer £4,000 to finish the module and has them sign an IP assignment, because the provider's terms passed rights to the freelancer, not to Maya, until that assignment is in place. Before release, the team reviews one suspiciously familiar function and rewrites it to avoid shipping an embedded licensed pattern.
Where founders go wrong
Assuming the tool gives you copyright
— the provider's terms transfer whatever rights exist; they do not manufacture a strong copyright out of thin human input.Skipping the contractor IP assignment
— AI-assisted code from a freelancer still needs a signed assignment, or the rights are not yours.Shipping generated code unreviewed
— an output can embed licensed patterns or resemble a third party's work, and you carry that risk downstream.Treating s.9(3) as settled
— the computer-generated-works rule predates modern AI and its application is untested and under review.
Related questions
Does copyright even exist in AI-generated work?
It is unsettled. UK law has a provision for "computer-generated works" with no human author, but how it applies to modern generative AI is untested. Where a human meaningfully selects, edits and arranges the output, ordinary copyright is much easier to argue. [More: Who owns copyright in software in the UK?]
Who owns the output under OpenAI or Anthropic terms?
The major API providers assign their rights in the outputs to you, the customer, under their terms of service. That is a contractual transfer, not a statutory grant, so always check the current terms and the specific tier you use before relying on it.
Can I claim my AI-assisted code is my own IP?
Your position is strongest where a human did substantial selection, editing and arrangement. Pure one-shot output with no human input attracts the weakest protection, because thin human authorship means thin copyright.
Can AI output infringe someone else's copyright?
Yes. An output can resemble third-party copyrighted material or embed licensed code patterns from training data. You carry that risk when you ship it, so review generated code and content before it goes into your product. [More: Can I legally use open-source code in my commercial product?]
Who owns code my contractor generated with AI?
Not automatically you. You need a written IP assignment from the contractor, exactly as you would for hand-written code, plus comfort that the provider's terms passed the output rights to them in the first place. [More: Who owns the code if an agency built my MVP?]
Owning your product's core IP is what investors diligence and acquirers pay for — and AI-assisted work adds a provider-terms layer and an infringement risk that a template cannot see. A SuLe solicitor can map who actually owns your code and content and close the gaps before a raise. Book a free consultation about your product's legals and get a regulated startup lawyer to check the chain of title.
Keep reading: Can I train an AI model on customer or user data? · Can I legally use open-source code in my commercial product? · What should an AI product's terms of service cover? · Who owns copyright in software in the UK? · Who owns the IP my employees and contractors create? · Can I patent software in the UK?
Primary sources: Copyright, Designs and Patents Act 1988 · Open Source Initiative — licences
