Who owns what AI writes? Copyright for businesses, explained

If a logo, an article or a design comes straight out of an AI tool from a prompt, you probably cannot enforce copyright in it. Nobody else owns it either, which is thin comfort when a competitor starts using something very like it. And not owning it is not the same as it being safe to use, because AI output can still reproduce material someone else owns.

The fix is not to stop using AI. It is to know which of three layers actually protects each thing your firm makes: the law, your tool's terms, and the registrations you choose to hold.

This now matters at the scale of everyday work. The Office for National Statistics reported on 20 July 2026 that AI use among UK businesses with ten or more employees has risen from around 12% to around 35% since late 2023. The two most common uses are exactly the ones this question hangs over: generating text, and creating visual content. Over half of employees, 55%, report using AI for work or education.

So the ownership question no longer applies only to the odd experiment. It applies to a growing share of what your firm produces every day.

This is general information, not legal advice.

In the US, prompts do not make you an author

The first layer is the law, and in the US it is settled and strict. The US Copyright Office's report on copyrightability, published in January 2025, says plainly: "Copyright does not extend to purely AI-generated material, or material where there is insufficient human control over the expressive elements."

Prompting is not control. In the Office's words, "prompts alone do not provide sufficient human control to make users of an AI system the authors of the output. Prompts essentially function as instructions that convey unprotectible ideas." Writing more prompts does not change that: "No matter how many times a prompt is revised and resubmitted, the final output reflects the user's acceptance of the AI system's interpretation, rather than authorship of the expression it contains."

The courts agree. In Thaler v Perlmutter, the appeal court in Washington held in March 2025 that US copyright "requires all eligible work to be authored in the first instance by a human being", and in March 2026 the US Supreme Court declined to review that ruling. The question is closed.

Two real cases show where the line falls. The author of Zarya of the Dawn, a comic made with AI images, kept copyright in her words and in her arrangement of the pages, but lost it in the individual pictures, because the tool, not she, created the images. And the artist behind an award-winning AI image says he wrote more than six hundred prompts to get it; the Office still refused registration, and his court challenge is unresolved.

Human contribution, by contrast, is protected, and it is being registered every week. Where a person selects, arranges or genuinely rewrites AI material, that contribution carries copyright, and the Register of Copyrights told the US Senate in May 2026 that the Office "has to date registered more than 7,000 claims that include AI-generated materials" where the AI content was disclosed.

The UK has its own rule, and the government has proposed removing it

The UK is the odd one out. Since 1988 it has had a specific rule for exactly this situation. Section 9(3) of the Copyright, Designs and Patents Act says that for a computer-generated work, "the author shall be taken to be the person by whom the arrangements necessary for the creation of the work are undertaken", and protection runs for 50 years.

That rule is in force today. We checked the statute on the day this was published, and it stands unamended.

Before you build anything on it, though, there are two warnings. Nobody knows whether the rule works. Copyright also requires a work to be original, and originality is measured by human creative choices, which is why "some question whether the provision could ever apply in practice", as the government's own report records.

And the government has proposed to remove it. Its report on copyright and AI, published on 18 March 2026, says: "in the absence of evidence of its ongoing value, we propose that this specific type of protection should be removed, while copyright should continue to protect works created with AI assistance."

A proposal is not a change in the law. Nothing has been introduced in Parliament, and nothing has moved since March. But the direction is clear.

The practical conclusion is the same in both countries. Do not build your firm's ownership position on the UK's unusual rule. Build it on human authorship, which both countries protect now and the UK proposal would keep.

Your tool's terms decide what you may do with the output

The second layer is the contract. Whatever the law gives you, your tool's terms govern the output, and all four major vendors have chosen their words with care.

OpenAI and Anthropic assign you their rights in the output. Both use the same phrase: OpenAI's terms assign "all our right, title, and interest, if any, in and to Output", and Anthropic's terms use it too. A vendor can hand you whatever rights exist; it cannot make copyright exist where the law says there is none.

Google and Microsoft take the other route and simply disclaim ownership: Google "won't claim ownership" of generated content, and Microsoft's business terms say "Output Content is Customer Data. Microsoft does not own Customer's Output Content."

Microsoft's consumer terms add: "You'll need to make your own determination regarding the intellectual property rights you have in output content." That is a vendor telling you, in its own contract, that the ownership question is yours to solve.

Three more things sit in those terms that most firms have never read. The first is that the output may not be exclusive. OpenAI's terms state that "Output may not be unique and other users may receive similar output", and Google's Workspace terms acknowledge the services "may, in some scenarios, produce the same or similar Generated Output for multiple Customers".

The second is that the plan tier changes your legal position. Anthropic's UK consumer terms apply, in their own words, to "a consumer who is resident in the United Kingdom", meaning someone "acting wholly or mainly outside your trade, business, craft or profession", and they add: "You agree not to use our Services for any commercial or business purposes." A sole trader doing client work on a personal Claude plan is not the customer those terms describe.

The third is that the protections firms actually want, the copyright indemnities that vendors offer when output attracts an infringement claim, all sit on the paid business tiers. Microsoft calls its scheme the Customer Copyright Commitment; Google covers paid Workspace; OpenAI covers its API and Enterprise customers; Anthropic covers paid commercial use. Every one has conditions, and the two common ones are that you did not know the output was infringing and did not switch off the safety filters.

Coverage is not uniform even among paid plans either. What protects you is the specific product you buy, not the word "business" on the invoice.

Across all of it the pattern is the same: the contract decides what you may do with the output, and only the law decides what you own. Together they point to one setup: a business-tier account, with a person genuinely shaping the work.

Who owns what: the two countries side by side

Four steps keep your rights enforceable:

1. Make the human contribution real, and keep the evidence. Prompts are not authorship in the US, but selection, arrangement and genuine rewriting are. Keep the drafts and edit history that show a person shaped the finished work.
2. Use the right commercial product for client work. Vendors assign you their rights in output on consumer plans too, but the permission to use it commercially and the indemnity if a claim lands come only with the paid product, and they vary between products. Check the terms of the one you actually buy.
3. Put one line in your client contracts. Say what AI may be used for in deliverables and what the client is getting. Our piece on the contract clauses AI breaks has the wording.
4. Register the marks that matter. This is the third layer: for anything that identifies your business, trade mark registration does not depend on who authored the work (see below).

Client work and logos: the two hard cases

Deliverables. If your firm sells creative or professional work, the ownership gap passes straight through to your clients: a deliverable that is partly AI-generated is a deliverable you may not fully own, assigned under a contract that assumes you do. The advertising industry's joint principles, from the IPA and ISBA, put the professional standard plainly: "Neither should include AI-generated content in materials provided to the other without the other's agreement." Tell clients, agree it, and get it in writing.

Logos and brand assets. A logo is a poor thing to leave resting on AI-generated copyright, and it does not have to. Trade mark registration is decided on the IPO's criteria of distinctiveness and conflicts with marks and rights others already hold; authorship is not among them. The one thing registration cannot cure is a logo that copies someone else's work, because their copyright can block or invalidate the mark. Registration costs from £205 and lasts ten years, renewable. Be clear about what it gives you: it protects use of the sign for the goods and services you register it for, which is what stops a competitor trading under your brand. It does not give you copyright in the artwork, and it does not prevent every reuse of the image. A design business generating brand assets with AI, as many of the studios in our design piece now do, protects its clients by registering the mark rather than waiting to see how the copyright question settles.

What to do this week

Pick the three assets it would hurt most to see a competitor copy freely. For each, ask which layer protects it: a person's documented creative work, the tool's paid-tier terms, or a registration. Anything resting on none of the three is exposed, and now you know it.

Then make the two cheap moves: shift client work onto business accounts, and add the AI line to your engagement terms.

If you would rather work through it with us, book a short call and we will map which of your firm's assets are enforceable, which are not, and what to change.

This is general information, not legal advice. For decisions that turn on ownership of specific work, in specific contracts, take advice from an intellectual property specialist.