Copyright law develops in response to technological changes and has done ever since Gutenberg invented moveable type. Artificial intelligence is just the latest challenge to established copyright laws, and on past form we can be sure that the rules will prove to be flexible enough to cope up to a point. But in the fullness of time, we can also be sure that the law will have to be amended to deal with new challenges that we can’t foresee at the moment.
The UK government has recently completed a consultation on the topic of AI, which it recognises as important to the country’s future prosperity, and intellectual property. In this article, our copyright lawyers discuss copyright law and AI. We also explain what you should do if you have been accused of copyright infringement as a result of using AI.
If you happen to be reading this article because you are involved in a copyright dispute involving AI and you need immediate advice, please contact our copyright lawyers who can advise you on your options and the steps you need to take to resolve your dispute.
Contents:
- Are works created by AI protected under UK copyright law?
- If copyright exists in AI works, who owns it?
- Will copyright protect everything made by AI?
- Can AI own copyright?
- How does AI use copyright works?
- Is using copyright works to train AI an infringement?
- Where does the government stand at the moment?
- How to reduce the risk of copyright infringement when using AI
- What to do if you have been accused of copyright infringement
- What steps should I take before I implement AI content generation tools in my business?
- Summary
Are works created by AI protected under UK copyright law?
Copyright protects original works of authorship, literary, dramatic, musical and artistic works and certain other matters, including films and sound recordings. Being able to identify the author of a work is the first step in working out whether copyright protects it.
When the Copyright, Designs and Patents Bill was going through Parliament in the late 1980s, the Government was pleased with the way it treated computer-generated works. It applauded itself for creating the first copyright law in the world to deal with this emerging issue, which at the time was thought to involve programming a computer to print out a graphic pattern, and not much else. Section 9(3) says that the author of (and the owner of copyright in) a computer-generated work would be the person who makes the arrangements necessary for the creation of the work were undertaken. It applies only to literary, dramatic, musical and artistic works, and does not cover film, sound recordings or broadcasts.
It was scarcely earth-shattering, and for thirty-odd years it worked well. Whether it still does is now openly doubted. In its March 2026 report, the government recorded that most of those who answered the question wanted protection for wholly computer-generated works removed. It proposes to do exactly that, while keeping protection for work made with the assistance of AI where a person makes the creative choices. That said, removing the provision would need parliamentary intervention, and none has been announced, so section 9(3) remains in force for the time being. Even so, it would be unwise to build a business model around it.
If copyright exists in AI works, who owns it?
The existing rules about copyright ownership are clear, although applying them is sometimes tricky. Generally, the author of a work is the first owner of copyright in it, and as the law deals with the authorship of computer-generated works it is going to be clear in many cases who is the owner of the copyright. If the author is an employee, and creates the work in the course of their employment, it’s the employer who will be the first owner of copyright. Commissioning a work does not of itself transfer the copyright, however much you paid for it. So, as usual, anyone who commissions someone else to make something needs to be alert to the need for an assignment. This will apply just as much when the creator uses AI to make the commissioned work.
Will copyright protect everything made by AI?
What is going to be interesting to see is how the judges approach the question of originality in an AI-created work. When a creator uses a computer as a tool to realise their creative idea, sometimes called the 'clever pencil' approach, it is usually fairly clear that the original content of the work came from the person using the computer. AI, however, involves the machine doing at least some of the original work. The user sets the parameters, and the AI machine does the rest. So, will setting the parameters amount to that person’s own intellectual creation?
Whether writing a prompt can qualify as a user making free and creative choices is not a question we can answer at this stage, and it's never going to be a question to which there is one single answer. It's always going to be highly fact-specific, and there are sure to be some cases in which the individual puts enough into setting those parameters to get over the originality threshold. Equally, there will be cases in which all the original work has been done by the AI machines.
Can AI own copyright?
The courts have not yet been asked to consider whether AI can be the author of a copyright work. They have been asked whether it can be the inventor of a patent, and their approach to that question gives us an idea of the approach judges may take to the copyright one.
In Thaler v Comptroller-General of Patents, Designs and Trade Marks [2023] UKSC 49, the DABUS case, the Supreme Court held unanimously in December 2023 that an inventor under the Patents Act 1977 must be a natural person. Dr Thaler had declined to name himself as inventor, insisting that his machine had done the work and that his own entitlement came from owning the machine. That proved fatal to his applications. As far as the court was concerned, a machine has no legal personality, so it had no rights capable of being transferred to him.
Most of the judgment in the DABUS case is about the wording of the Patents Act. The reason the machine could not hold or pass on rights was that the law does not treat it as a person, and that is as true of copyright as it is of patents. Applied to copyright, that reasoning makes machine authorship a difficult argument to run, and section 9(3) makes it harder still. Parliament has already said that the author of a computer-generated work is the person who made the arrangements necessary for its creation. An author must be a person recognised in law, which a machine is not. The chances of Parliament finding time to overhaul copyright law any time soon are slim.
How does AI use copyright works?
AI doesn’t just create new stuff; it uses older stuff in doing so. There is nothing new in that, and creative people have always built on the work of earlier ones. What is different is the scale, and the fact that it all happens in digital form. If a composer hears another composer's music and something about it lodges in the memory, copyright is not engaged. Training a model means making copies, and copying all or a substantial part of a protected work without permission infringes copyright unless an exception applies. That is so whether or not the finished output resembles anything in the training data.
Which works are used makes a difference. Ask AI for an image in the style of Leonardo da Vinci and it will draw on works long out of copyright, so no issue arises, leaving aside any rights in digital photographs of the originals. Andy Warhol or David Hockney is another matter.
Is using copyright works to train AI an infringement?
Cases on issues like this usually come to court in the USA first, and several have. A group of visual artists has started a class action against Stability AI Ltd, Midjourney Inc, and DeviantArt Inc in San Francisco. The artists allege that Stability AI's Stable Diffusion software copies billions of images, enabling Midjourney and DeviantArt to create images in the styles of your choice of artist, and does so without permission. An earlier class action has been filed against GitHub Inc and its business partner OpenAI Inc for allegedly scraping copyrighted source code, to train AI systems.
The key issue for the American courts is whether any of this amounts to fair use within the meaning of the US copyright legislation. In Bartz v Anthropic, Alsup J held that training a large language model on lawfully bought books was transformative and fair, but that building the training library out of pirated copies was not. That case has since settled for US$1.5 billion, with final approval given in July 2026. In Kadrey v Meta, Chhabria J also found fair use, though on much narrower ground. His decision was confined to the thirteen authors before him, and turned on their failure to produce evidence of harm to the market for their books rather than on any general approval of the practice. Neither decision offers AI developers general guidance on the use of copyright protected material for AI training.
The first UK judgment came on 4 November 2025, in a claim brought by Getty Images against Stability AI over the use of millions of Getty photographs to train Stable Diffusion. It was widely expected to settle whether training a model on protected works infringes UK copyright. It did not.
Most of the training took place abroad, and UK copyright only covers acts done in the UK, so the central claim never got off the ground. Getty argued instead that supplying the finished model in this country infringed, because the model itself contained copies of its photographs. The judge found that it did not: the images are not stored inside the model in any form, so there was nothing to infringe. She did accept that a model trained abroad could infringe if it did contain copies, which may matter in a future case, but it did not help Getty here. Getty has permission to appeal the copyright ruling, which had not been heard when this article was reviewed. So, as things stand, the position remains unclear for rights holders and businesses using AI alike.
Where does the government stand at the moment?
In the UK, the government has consulted on how the law should deal with the issue. Its proposals, which entail allowing AI companies to make free use of copyright material to train their programs while giving copyright owners a right to opt out, have been very badly received by composers, authors and others. Creators not only face the prospect of their rights being disregarded, but they will also face competition from AI-created versions of their own work. In March 2026, the government confirmed that it had dropped the opt-out model, that it has no preferred option for the time being, and that it will not reform copyright law until it is confident any change will work.
How to reduce the risk of copyright infringement when using AI
Exposure to copyright infringement actions is a problem for both the AI companies and commercial end users. The process of making an image in the style of Andy Warhol entails a lot of copying and processing of existing works by the artist, to produce an end result that looks different from any of them. The finished product may well not reproduce a substantial part of any Warhol image, and therefore not amount to an infringement of his copyright. AI end users would do well to minimise the risk by not asking the machine to make something in the style of a particular artist. We cover practical tips for reducing your exposure in our copyright infringement guide.
What to do if you have been accused of copyright infringement
If you are accused of infringement, the most important thing to do is to consult a copyright lawyer with experience in dealing with copyright claims, not just novel AI-related claims. Often, they can be made to go away completely, or at least settled for much less than what is being demanded by the other party.
What steps should I take before I implement AI content generation tools in my business?
Like many innovations, AI has attractions for people who would like to use it in their working lives. That could be just to write minutes of meetings, or to automate some other routine tasks. But there could be a temptation to use it for less mundane purposes, which could involve creating imitations of other peoples work, and every business is going to have to consider putting in place an AI policy to ensure that employees do not misuse AI tools.
One recent example we have seen made clear that employees could use AI to take minutes of meetings, but was very specific about which AI products (Microsoft Copilot) could be used. The policy makes clear that employees cannot use AI tools for anything else, unless specific permission is obtained: the employer has to be sure about what AI is being used, and for what, in its business. No company would tolerate employees installing whatever software they liked on the office computer system, and AI is not materially different in this respect. Of course, how far you will go depends on the nature of the business.
Other matters to consider are whether you need to restrict what employees might do outside the office if they are working at home in the evening or over the weekend, and indeed if they are working from home and using their own laptops. You’ll want to be able to monitor what AI is used for, and might require a formal record to be kept of AI tools used in the business. Do you need to list AI tools that may not be used under any circumstances? Do employees need to be told that they cannot buy AI tools and expect the company to reimburse them?
It is important to keep your policies and procedures under review, because the law is unsettled and things will likely change. The government has said its next phase of work will cover digital replicas, meaning synthetic recreations of a person's voice, face or likeness, the labelling of AI-generated content, and giving creators better control over how their works are used online. A consultation on digital replicas was promised for the summer, and a taskforce on labelling is due to report in the autumn.
Most businesses will probably think the last thing they need is another policy, but there’s real scope for a rogue employee to create legal and financial headaches. It’s also a good opportunity to revisit your brand and creative asset protection measures. We cover how to do that in our copyright protection guide.
Summary
Artificial intelligence is a hornets’ nest of possible legal problems, from infringing other people’s copyright to creating new copyright works, the owner of which might not be easy to identify.
The technology is no longer new, but the law about it very nearly is. The first UK judgment was handed down in November 2025 and is under appeal, and the government has decided that the most sensible thing it can do for now is to wait and see. That is an uncomfortable answer for anyone who has to make decisions in the meantime, but waiting is not an option open to a business already using these tools or planning to do so. The time to have a plan is now, before one of your employees goes off and does something that turns out to be a bad idea.
If you would like help putting that plan together, our intellectual property lawyers can advise on your AI policy, review where your valuable IP sits, and check the contracts and supplier terms that sit behind it.
