In short: UK copyright law has not changed for AI. The government's report of 18 March 2026 dropped its favoured plan for a training exception with an opt-out, proposed no intervention in licensing and said it will not legislate until it is confident. Until then, existing law applies, so treat AI output like any supplier's work and put the risk in writing.
Nothing has changed in UK copyright law because of AI, and the government has said it will not change it until it is confident the change works. What changed in March 2026 is the direction of travel. This page sets out where things stand on 5 November 2026.
It is general information for people who commission content, not legal advice. If a decision turns on it, ask an IP solicitor.
The short version
- The plan the government preferred in 2024, a broad exception for AI training with an opt-out for rights holders, is no longer its preferred way forward.
- There is no new law yet, and the report gives no date for one.
- Copyright still works by country. The law of the place where the copying happens applies.
- The government proposes to leave licensing to the market for now.
- Ownership of AI output, labelling and protection for voices and faces are all still open questions.
How we got here
| Date | What happened |
|---|---|
| 17 December 2024 to 25 February 2025 | The government consulted on copyright and AI and received 11,520 responses. |
| 15 December 2025 | A progress statement said all options were still under consideration and that the report and impact assessment would follow by 18 March 2026. |
| 18 March 2026 | The report was published under the Data (Use and Access) Act 2025. |
In the consultation the government had named an exception with a rights reservation as its preferred option. Most respondents rejected it. The progress statement records that creative-industry respondents largely favoured licensing and that most technology respondents preferred a broader exception. The March report is the government's answer.
What the March 2026 report proposes
| Topic | Proposal | What it means for you |
|---|---|---|
| Training exception | A broad exception with opt-out is no longer preferred. More evidence will be gathered. | No new permission to train on your content has been created. |
| Licensing | No intervention in the licensing market at this stage. | Licensing deals stay a commercial matter between rights holders and AI developers. |
| Transparency | Work with industry on best practice for telling rights holders what was used in training. | There is no legal duty on developers to disclose training data yet. |
| Labelling | Work with industry on best practice for labelling AI-generated content. | The report says there are currently no UK obligations to label. |
| Computer-generated works | Remove the specific protection for works with no human author. | Fully automated output may lose copyright protection if the proposal becomes law. |
| Digital replicas | Explore options, including a possible new digital replica or personality right. | Nothing new protects a voice or face yet. Contracts do the work. |
| Overseas systems | Keep monitoring global developments and court outcomes for AI systems developed outside the UK. | The position on models trained abroad stays under review. |
What the law says today
Training. The one existing text and data mining exception, section 29A of the Copyright, Designs and Patents Act 1988, applies to non-commercial research only, so it does not cover commercial training. Where the training happens also matters, as the next point explains.
Territory. The report states that copyright is territorial, so the law of the country where a relevant act takes place applies. It also notes that most training is likely to happen in the most permissive jurisdictions. That is why the government says it will watch the United States and the European Union closely.
Outputs. If protected work appears in a model or its output, the report says existing law on communication and distribution to the public, and on secondary infringement, applies. It refers to the Getty Images v Stability AI judgment, in which the High Court held that a trained model could in principle be an "article" and so an infringing copy, but found that the particular model did not contain copies of the works. The report notes that the judgment is under appeal. For a business the practical point is simple: a tool that can reproduce someone else's work can put you in the chain of infringement if you publish what it produces.
Ownership. Section 9(3) of the 1988 Act has protected computer-generated works without a human author since 1988. The government now proposes to remove it and keep protecting AI-assisted works. If you want to own what you commission, a person has to make real creative decisions, and you need a record that they did.
A worked example shows the difference. A designer who types one prompt and accepts the first image has made very few creative choices. A designer who plans the composition, generates several versions, combines parts, retouches by hand and records each step has made many. The second file is much easier to defend as AI-assisted work. Neither is a guarantee, because the report does not define where the line sits, and a court would decide on the facts.
What to do if you commission content
- Ask which tools were used. A supplier who uses AI for stills, voice or edits should say so before the work starts, and say which tool.
- Put the risk in the contract. Decide who warrants that the output does not infringe, who pays if a claim arrives, and whether the supplier carries insurance for it. Our note on who is responsible for an AI-made ad covers the usual positions.
- Keep a creative record. Save prompts, selections and the human edits. If the protection for fully automated work goes, this record is how you show a person shaped the result.
- Read the tool's terms. They set what the vendor claims over output and whether your uploads can be used for training. Check both before you feed in unpublished material.
- Do not prompt with someone else's name or character. Asking for "in the style of" a named living artist, or for a protected character, raises the chance of output that copies.
- Get written consent for any voice or face. Until the government decides on a personality right, a signed agreement is the only protection a performer has. Use the voice and likeness consent checklist, and see what performers and producers can do about voice cloning.
- Label by choice. There is no UK duty to label yet, but audiences and platforms are moving that way. The guide to labelling AI content explains the options.
Choosing between AI and a shoot
Copyright risk is one input to the decision. If a brand needs a face, a product or a place that must look exactly right, a camera avoids the questions above. The page on when to use AI and when to shoot it sets out the other inputs, and AI video for small businesses covers what the tools can do today.
How we handle it
At Granary the Phoenix Formula course is partly voiced with a consented cloned voice and labelled as synthetic. A person approves every post in our weekly content engine. Governance you can show a board is one of the twelve capabilities in The AI Studio.
What would change this page
Watch for four things: a bill or consultation on training, an appeal decision in the Getty case, a government move on a digital replica right, and any labelling code. The report says the government expects greater clarity on the law in the United States and the European Union within a year, so expect another update around then.
Sources
- GOV.UK (DSIT and DCMS), Report on Copyright and Artificial Intelligence, read 5 November 2026
- GOV.UK (DSIT and DCMS), Copyright and artificial intelligence: statement of progress under Section 137 Data (Use and Access) Act, read 5 November 2026
Related guides
- AI video for small businesses: what's real, what's risky
- When to use AI and when to shoot it for real
- AI in the production workflow, stage by stage
- Who is responsible for an AI-made ad?
- Voice and likeness: a consent and contract checklist
- All guides and tools
The picture at the top of this page is an illustration, not a photograph of a client or a real job.
Questions we get asked
Is it legal to train AI on copyrighted work in the UK?
The only text and data mining exception in current UK law covers non-commercial research, so it does not cover commercial AI training. The government's March 2026 report says a broad exception with an opt-out is no longer its preferred way forward, and it has not yet proposed a replacement. Anything specific needs a solicitor.
Who owns the copyright in AI-generated content in the UK?
Today the Copyright, Designs and Patents Act protects computer-generated works that have no human author, and the government now proposes removing that specific protection. It says AI-assisted works, where a person makes the creative choices, would stay protected. Check the terms of the tool you use as well, because they set what the supplier claims.
Do I have to label AI-generated content in the UK?
The government's March 2026 report says there are currently no UK obligations to label AI-generated content. It proposes to work with industry on best practice for labelling. Advertising, platform and broadcaster rules can still apply, so check those before you publish.
granary.digital · 01243 884993 · studio@granary.digital · The Old Grainstore, Adsdean, Chichester, West Sussex, PO18 9DN · Prices exclude VAT unless marked inc VAT