Who Owns the Logo Your AI Just Designed?

A founder needs a logo. They open an AI image tool, describe the business, and after a few prompts a clean, distinctive mark appears. It looks the part. The platform's terms say the user owns the output. So the founder puts it on the website, the invoices, the van and the signage, and starts building a brand around it.

The instinct is understandable: the tool made it, the terms hand it over, therefore it is yours. But “Who owns this?” is not one question with one answer. It is at least five separate questions that businesses tend to collapse into one — and a permissive answer to the first does not settle the rest.

Five questions hiding inside one

The reason ownership feels confusing is that people ask a single question — “Is it mine?” — while the answer lives in five different places. Each has its own logic, and they do not agree with one another automatically.

Work through them in order and the picture becomes clearer — and more honest — than “the platform says it's mine.”

Question one: what does the platform actually grant?

Most well-known AI tools address output rights in their terms of use. Many assign or license the output to the user, sometimes with conditions, and often with different rights on free versus paid tiers. That is worth checking carefully: a grant on a paid business plan may look nothing like the grant on a free consumer account.

But even the most generous grant is a promise from the platform to you. The platform cannot give you rights it does not hold, and it cannot promise that the output is free of everyone else's rights. Read the terms for what they exclude as much as what they grant — warranties, indemnities and liability caps tell you how much risk the provider is prepared to carry, which is usually far less than a business assumes.

• Check the terms for the specific tool and the specific plan you are on, not a general impression of “what AI tools do”.
• Note any conditions on commercial use, attribution or resale.
• Look for what the provider does not warrant — originality and non-infringement are commonly excluded.
• Save a dated copy of the terms you relied on, because they change.

Question two: does copyright even arise?

Here is where many businesses are surprised. Owning something and there being something to own are different things. If little or no enforceable copyright arises in a largely machine-generated image, then “the platform assigned it to me” assigns a thinner right than expected.

UK law contains an unusual provision. Section 9(3) of the Copyright, Designs and Patents Act 1988 addresses “computer-generated” works where there is no human author, stating that the author is taken to be the person by whom the arrangements necessary for the creation of the work are undertaken. On its face, that looks like a ready-made answer for AI output.

The provision predates generative AI by decades, and commentators disagree about how it applies to today's tools — including what “the arrangements necessary” means when a model, a prompt and a user all contribute, and whether such works meet originality requirements at all. The UK government has been examining exactly these questions as part of its work on copyright and AI. In short: there may be copyright, it may be thinner than you hoped, and reasonable experts currently disagree about where the lines fall.

Question three: could it infringe someone else's rights?

This is the risk that platform terms simply cannot switch off. Generative models are trained on large volumes of existing material, and outputs can sometimes resemble, echo or reproduce protected works, existing logos, trademarks or a recognisable house style. That risk exists whatever your platform terms say about ownership.

A logo that unknowingly resembles an existing registered mark in your sector is a problem you inherit at the point you start using it commercially — regardless of the fact that an AI produced it and the terms handed it over. The tool will not have run a trademark search for you.

• Avoid prompting for a named living artist's, designer's or competitor's style for commercial work — style imitation invites disputes even where the law is unsettled.
• Run image, trademark and general web similarity checks on any mark before you adopt it.
• Be especially careful with anything that echoes a well-known brand's colours, shapes or typography.
• Remember that third-party risk sits with the business using the mark, not the tool that generated it.

Question four: how much human contribution can you prove?

Across many legal systems, human creative contribution matters to whether and how strongly a work is protected. A raw, one-prompt generation with no further human input sits on the weakest ground. A design that a person has selected, arranged, edited, refined and directed sits on much firmer ground — and, crucially, you can evidence that work.

For a business, the practical takeaway is not a legal theory but a habit: treat AI as a starting point that a human then shapes, and keep the trail. That trail is your evidence of human authorship if anyone ever asks.

• Keep the prompts you used and how they evolved.
• Keep the intermediate versions and the final human edits — the cropping, recolouring, redrawing and refinement.
• Note who did the creative direction and the decision-making.
• Record the date and the tool and plan used to generate the material.

Question five: can the logo be registered and defended?

A logo you rely on is usually protected in practice as a trademark, not just through copyright. Trademark protection asks different questions again: is the mark distinctive, is it available in your classes, and can you register and defend it? AI-generated marks can raise added questions about originality and about similarity to other outputs or existing marks.

For a logo the business intends to depend on — the one going on the building and the contracts — this is the point to bring in a trademark professional before committing. A clearance search and a registration strategy are far cheaper than a forced rebrand after the brand has taken hold.

Training-data disputes and policy: watch, don't assume

Two developments are worth watching, precisely because it is tempting to over-read them in either direction.

The first is litigation over the use of copyrighted works to train AI models. In the UK, Getty Images brought proceedings against Stability AI in the English High Court. In a judgment handed down in November 2025, the court dealt with a narrowed set of claims; Getty did not succeed on the copyright issues that remained before the court, and the widely reported training and text-and-data claims had already fallen away or were not pursued to a UK ruling. Legal commentators have stressed that the decision turned on specific issues and does not settle the broader questions about AI training and copyright.

The second is UK government policy. The government ran a consultation on copyright and artificial intelligence, which closed in early 2025, and has since published further material as it works through the responses. The direction of any reform — for example around text and data mining, transparency and how the law treats AI-created works — was not finally settled at the time of writing.

Do not treat ongoing litigation or an open policy process as though it had already decided anything. What a business can do today is practical rather than speculative.

Provenance records: cheap insurance

You cannot resolve unsettled law on your own, but you can put yourself in the strongest evidential position for very little cost. Provenance — a simple, honest record of how something was made — is the cheapest insurance available for AI-generated brand assets.

1. 1Record the tool, the plan and the date used to generate each asset.
2. 2Keep the prompts and how they were refined.
3. 3Keep the drafts and the human edits, so the creative contribution is visible.
4. 4Save the platform terms you relied on at the time.
5. 5Log the checks you ran — similarity searches, and any trademark clearance.
6. 6Note who approved the asset for use and when.

None of this is onerous, and it turns a vague “the AI made it” into a documented account you can stand behind. For assets that matter, add a review step before adoption and a professional check before registration.

So — who owns the logo?

The most accurate answer is: it depends on five questions, and the platform's permission only answers the first. AI-generated content is usable, but unevenly protected. For a throwaway social graphic that hardly matters. For the logo you will build a business on, it matters a great deal.

Treat AI output as a strong starting point that a human then shapes, evidence and check the work, and take advice before betting a brand on a design. That is not caution for its own sake — it is the difference between a mark you can defend and one you merely hope no one challenges.

AI Governance hub — practical governance for small businesses →

Guide: Who Owns AI-Generated Content? (AI, copyright and ownership) →

Download: AI Governance Checklist for small businesses (PDF) →

Sources and further reading

The following are official and reputable sources for the points above. Legislation, live litigation and government policy change — check the current position before relying on any of them. This article is general information, not legal advice.

Copyright, Designs and Patents Act 1988, section 9 (legislation.gov.uk) →

GOV.UK — Copyright and Artificial Intelligence (consultation) →

GOV.UK — Report on Copyright and Artificial Intelligence →

Courts and Tribunals Judiciary — Getty Images v Stability AI →

Intellectual Property Office — trade marks guidance (GOV.UK) →