Copyright and AI law

UK copyright and AI: where the law stands after the 2026 report

A group of people shelter beneath colourful umbrellas on open grassland.
Umbrellas in an open field.Photograph by Robert Harper. Robert Harper Photograph Library, Archive.London. From the archive, shown for illustration.

The government spent fifteen months proposing to let AI developers train on copyright works unless owners opted out. In March 2026 it withdrew that preference and put nothing in its place.

On 18 March 2026 the UK government published the report that Parliament had ordered it to write on copyright and artificial intelligence. Its central sentence is a retreat: a broad copyright exception with an opt-out "is no longer the government's preferred way forward". No other option was preferred in its place, and no bill was proposed. Eight months later that is still the position.

This article is general information, not legal advice. Checked on 18 November 2026.

What is the law on AI training in the UK right now?

The starting point is unchanged since 2014. Copying a photograph, or any other copyright work, without permission infringes unless an exception applies. The only exception written for computational analysis is section 29A of the Copyright, Designs and Patents Act 1988, which came into force on 1 June 2014. It allows a person who already has lawful access to a work to copy it for text and data analysis, but only for the sole purpose of non-commercial research. Contract terms that try to prevent that are unenforceable.

Commercial model training falls outside section 29A. The government's own impact assessment, published alongside the March 2026 report, said permission would usually be needed to copy protected works during AI training, as quoted by Reed Smith. That is a description of the existing statute, not a new rule.

Two things complicate the plain reading. UK copyright is territorial, so it bites only on copying that happens in the UK. And no court has yet ruled on whether training itself infringes, for reasons set out below. A longer comparison with Europe and America is in our guide to text and data mining exceptions in the UK, EU and US.

What did the March 2026 report say?

The report exists because of the Data (Use and Access) Act 2025. After a long dispute between the Lords and the Commons over transparency amendments, the Act received Royal Assent on 19 June 2025 without them, but with three reporting duties: an economic impact assessment, a report on the use of copyright works in AI development, and a progress statement. The progress statement appeared in December 2025. The report and impact assessment were published on 18 March 2026, within the nine months the Act allowed.

Its conclusions, taken from the report itself and from law-firm summaries:

Topic What the report says
Opt-out exception No longer the preferred way forward
Preferred option None stated
Legislation None proposed; reforms will not be introduced "until we are confident" they will work
Transparency No statutory duty; government will "work with industry and experts to develop best practice on input transparency"
Technical tools and standards Need for regulation kept "under review"; no standard mandated
Computer-generated works Protection for works with no human author should be removed
Digital replicas Options to be explored, possibly including a new right
Next steps Gather evidence; monitor litigation, the licensing market and other countries

The report also records why rights-reservation tools are hard to rely on. Metadata can be stripped from digital files, robots.txt works at the level of a website or domain and not an individual work, and standards that would let owners set different preferences for search crawlers and AI training crawlers are still being developed. It identifies those gaps without legislating for them.

Why was the opt-out plan dropped?

The opt-out idea came from the "Copyright and AI" consultation run by the Department for Science, Innovation and Technology, the Intellectual Property Office and the Department for Culture, Media and Sport between 17 December 2024 and 25 February 2025. It set out four options. The stated preference was the third: a commercial text and data mining exception that rights holders could opt out of, backed by transparency duties.

About 11,500 responses came in. According to Mishcon de Reya's summary of the December 2025 progress statement, 88% of respondents to the online survey backed a requirement for licences in all cases, and 3% backed the government's preferred option. Reed Smith reports 81% support for the licensing option across all 11,520 submissions. The figures use different bases, so quote the base with the number.

This was the second failed attempt to widen the exception. In June 2022 the Intellectual Property Office proposed a text and data mining exception for any purpose with no opt-out at all; ministers abandoned it in February 2023. Neither proposal was ever law, though both are still sometimes described as if they were.

Did Getty v Stability AI settle the question?

No. Mrs Justice Joanna Smith's judgment of 4 November 2025 is often summarised as a win for AI developers, but the court never ruled on training. Getty dropped its primary infringement claim during the trial because there was no evidence the copying had taken place in the UK. What the judge did decide was narrower: the Stable Diffusion model was not an "infringing copy" for the purposes of secondary infringement, because the model weights do not store or reproduce the photographs.

That point went to the Court of Appeal on 11 and 12 November 2026, with the BPI intervening in support of Getty. Judgment has not been handed down. Until it is, the High Court ruling stands. On its reasoning, a model trained abroad on UK photographs that does not store or reproduce them is not an infringing copy, so offering it in the UK does not attract secondary copyright liability on that ground. Whether that survives is the single most important open question in UK law for anyone who owns or buys image data. The full sequence is in our plain-English timeline of Getty v Stability AI.

What is still pending?

Several things were promised and, as far as we could verify on 18 November 2026, not all have materialised.

  • The Court of Appeal judgment in Getty. Heard, not decided.
  • A final policy decision. The House of Lords Communications and Digital Committee asked in March 2026 for a commercial text and data mining exception to be ruled out, for large AI developers to disclose what data they use, and for a decision within twelve months. The government's response, published in May 2026, repeated the work programme it had announced to Parliament on 18 March 2026: a consultation on digital replicas, promised for the summer; a taskforce on labelling AI-generated content, with an interim report due in the autumn; a review of how creators can control their works online; a working group on whether government should help smaller creative organisations to license their content; and continued work on a marketplace called the Creative Content Exchange.
  • Delivery of those commitments. The March 2026 statement said a pilot of the Creative Content Exchange with public institutions had launched. We found no primary-source confirmation that the digital replicas consultation has opened or that the labelling taskforce has reported. Any claim that they have is best treated as unverified until it appears on a government page.
  • Legislation. None published.

One further gap is worth stating plainly because it is often misreported. There is a collective licence for generative AI in development from the Copyright Licensing Agency and its partners, but it covers text. We found no launched UK collective licence for AI training on images.

What does this mean for photographers, agencies and data buyers?

For rights holders, the practical position is the one that existed before the consultation: on the plain reading of the statute, commercial training carried out in the UK will usually need a licence, and the licence has to come from whoever actually owns the copyright. No court has yet tested that reading. For older work that is not always the photographer, as our article on who owns the copyright in a commissioned photograph explains.

For buyers, the absence of an exception does not make every unlicensed dataset unlawful in the UK, because of the territorial point. It does mean that the more dependable route to UK-origin photographs is a contract with the owner. A copyright licence is also not the whole job, and specific projects call for professional advice. Photographs of identifiable living people are personal data under UK GDPR, authors have moral rights that survive an assignment of copyright, and brands and artworks shown in a picture carry rights of their own.

For anyone publishing images online, embedding rights information in the file can still be useful as a record, even though no UK statute gives it formal effect as an opt-out. Our free AI training opt-out metadata tool writes the IPTC "Data Mining" property into a JPEG in the browser.

Where the Robert Harper archive sits in this

Archive.London is the home of the Robert Harper Photograph Library: 95,487 unique photographs by one photographer, owned by 825 Ventures Ltd, the owner and rights holder. The figure comes from the archive's inventory and has not yet been independently reconciled. More than half of the archive is still on film and has yet to be scanned, and only a small selection has been published online by Archive.London, including the studio still life in the jewellery collections.

In a legal landscape where the default is permission, a single counterparty is a practical advantage, and no more than that. Licences are negotiated directly with the rights holder and are subject to rights review, as set out on our rights and stewardship page. If you are assessing image data for a model, the AI training data licensing page explains how to start that conversation.

Frequently asked questions

No UK copyright exception is written for commercial AI training on copyright works without permission. Section 29A of the 1988 Act covers only non-commercial research by someone with lawful access. No court has ruled on training directly: in Getty Images v Stability AI the training claim was abandoned because the copying was not shown to have happened in the UK.

Does the UK have an opt-out system for AI training?

No. An exception with a rights-reservation opt-out was the government's stated preference in the consultation that ran from 17 December 2024 to 25 February 2025. It was never enacted, and the report published on 18 March 2026 says it is no longer the preferred way forward. Nothing has replaced it.

Very little in legal terms. It dropped the opt-out preference, proposed no bill and no regulator, and said the government would gather evidence, monitor litigation and overseas law, and develop best practice on input transparency with industry. It also proposed removing copyright protection for computer-generated works that have no human author.

No bill has been published and no timetable has been given. The March 2026 report says reforms will not be introduced until the government is confident they will work. Commentary on the May 2026 King's Speech reported no AI bill. That could change, so check the date of anything you read on this.

Sources

This article is general information, not legal advice. It reflects the position as we understood it on the date shown and may be out of date. Published by Archive.London. Corrections are welcome under our editorial policy.

Licensing the archive for AI and data projects

95,487 photographs from one photographer and one rights holder. Read the dataset card, search the 503-folder inventory, discuss a sample set under NDA, or arrange a first look as scanning proceeds.

See AI and data licensingWrite to the archive