UK copyright and AI: strategic pause or stuck in the past?

Insight

15.05.2026

Overview of the UK position

UK copyright law is rooted in the Copyright, Designs and Patents Act 1988 (the CDPA) – legislation drafted decades before large‑scale generative AI was conceivable. Although provisions of the CDPA on text and data mining for non‑commercial research (section 29A) and computer‑generated works (section 9(3)) have been the subject of much debate in this area, the CDPA contains no express treatment of AI training or outputs, leaving key uncertainties such as:

whether AI-powered data scraping constitutes unlawful copying (an act of primary copyright infringement), especially where the act of copying data protected by copyright to a server occurs after the act of scraping itself, and potentially to servers outside the UK; and
if so, what meaningful options rightsholders have in practice to address any such infringing acts. Without a clear statutory right to point to, rightsholders must fall back on their contractual rights for any use of the content on the terms made available, and enforcement of rights, whether in English courts or overseas, becomes a riskier and less attractive prospect.

The result is that the UK copyright and AI narrative is increasingly playing out through litigation and commercial licensing rather than recourse via legislation, and without the clarity that policy reform could deliver.

Recent developments: the UK government's Report on Copyright and Artificial Intelligence

The most significant recent development in this area is the UK government’s Report on Copyright and Artificial Intelligence, published on 18 March 2026 pursuant to sections 135–136 of the Data (Use and Access) Act 2025 (the Report).

The key takeaways from the Report are:

Nothing will immediately change. The government has expressly declined to introduce new copyright legislation or regulatory oversight at this stage, opting instead to maintain the status quo while continuing to gather evidence and monitor international developments in its effort to support both the creative industries and the AI sector (and attempting to reconcile their competing demands).
The government has abandoned its previously preferred option of an exception to copyright infringement which would permit commercial text and data mining, due in no small part to the strong backlash from the creative industries (highlighting the battle lines between the two sectors), but has not yet proposed an alternative way forward.
The Report acknowledges strong consultation support for a licensing‑first approach to regulate AI-related use and exploitation of data and content but stops short of endorsing it in law, making clear that the government is distancing itself (at least for now) from intervening directly. Instead, the government will monitor the licensing market. Indeed, the Report references the Creative Content Exchange (the CCE) – a proposed UK government-backed marketplace designed to facilitate the legal licensing of digitised cultural and creative assets for AI training, with the aim of facilitating access to high-quality data for AI developers while ensuring content creators are fairly remunerated and their copyrights respected. We are pleased to be advising the Natural History Museum – the host of a closed pilot of the CCE expected to launch in summer 2026 – on its planning and implementation of this ambitious framework.

The Report also cautiously signals several areas where reform remains live:

Computer‑generated works: the government indicates that section 9(3) CDPA protection (which assigns authorship of certain 'computer‑generated works' in circumstances where 'there is no human author' to 'the person by whom the arrangements necessary for the creation of the work are undertaken') should likely be removed, subject to further monitoring. That position comes as little surprise. Section 9(3) does not sit easily within the copyright framework, having been introduced to address early computing systems rather than technologies now capable of generating sophisticated outputs from limited human prompts, and with little evidence that it is being used in practice. A replacement provision that addresses the application of copyright to modern AI technologies with greater clarity and coherence would be a welcome development.
Transparency: the Report explores transparency obligations around AI training data and outputs but defers final decisions on the matter. Instead, it proposes developing best practice on input transparency by working with industry, which may inform future legislation (we consider transparency obligations more generally in section 4 below).
Digital replicas and deepfakes: the government suggests stronger protection may be required but has not committed to a preferred model. While English law does not currently have a distinct body of law or rights relating to one's image (unlike the USA, for example), with high profile individuals such as Matthew McConaughey and Taylor Swift turning instead to trade marks to protect their voices, likeness and catchphrases, there is certainly room for updates to English copyright law to address this area expressly.

Overall, the tone of the Report is deliberately non-committal. Making a piece of legislation that is approaching 40 years old work for a technological landscape that is evolving at speed is plainly no small task. The government’s decision to take stock and gather further evidence to get this right is understandable.

In practice, however, this caution does not serve either side of the debate particularly well. Rightsholders are left to negotiate licensing arrangements, often with significant resource limitations and against counterparties with greater bargaining power and deeper pockets, assuming those counterparties come to the table in the first place. At the same time, AI developers continue to operate with the potential risk of infringement looming overhead.

Although the government has said that it will legislate once it is confident reforms will meet its objectives for the economy and UK citizens, there is a real question as to whether waiting for the right moment is realistic in a sector defined by technologies that move rapidly and test legal boundaries as they do so. In such an environment, moving slowly in pursuit of certainty risks falling behind and watching the framework fracture.

Legislation out: licensing and litigation in

In the absence of statutory reform, market behaviour is filling the gap in two principal ways: licensing and litigation.

With licensing, major publishers, image libraries and collecting bodies are entering into bespoke, express contracts with AI firms and tech companies. Their content is made available subject to various conditions, for example clauses requiring disclosures and accountability about training use, model scope and downstream outputs.

Licensing addresses the uncertainty of asserting rights through the statutory framework of the CDPA. It enables rightsholders to safeguard and commercially exploit their content via contractual means. If a licensee misuses the content, rightsholders, as licensors, could claim against them for breach of contract.

However, this approach is not infallible. Entering into bespoke licensing arrangements for the permitted use or data scraping of a rightsholder's content can be a resource-, cost-, and time-intensive process for both sides. This is made more difficult by the fact that there is often an imbalance in negotiating power due to AI developers' commercial leverage (as noted above).

On the litigation side, Getty v Stability AI (which at time of writing, is on appeal), shows both the potential and the limitations of enforcement through the courts. In its November 2025 judgment, the High Court rejected Getty’s central claim for statutory copyright infringement of their vast image library, largely due to territorial and evidential grounds, and provided only limited guidance on whether training AI models on copyright-protected works infringes UK copyright law.

Although the decision confirmed that AI models can be 'articles' for copyright purposes (an important prerequisite for certain infringement claims), it left open many fundamental questions which rightsholders were hoping the courts would answer. The case also highlights the difficulties for rightsholders of enforcing their rights without clear transparency obligations on AI developers. Getty was not able to provide sufficient evidence that Stability's models had been trained on Getty's data in the UK, meaning it had to drop its primary copyright infringement claim. While litigation can produce tactical leverage and incremental clarification, it is slow, fact‑specific and jurisdictionally constrained (not to mention expensive).

For many cultural institutions, charities, and not-for-profits, full‑scale litigation is simply not a realistic course of action. This leaves the focus on risk governance rather than definitive legal answers – ie documenting decision‑making, adjusting access strategies and engaging collectively through trade bodies.

Transparency and comparisons with the EU approach

The UK government's more hands-off approach (at least for the time being), is an interesting contrast to the position adopted by the EU, which has passed a comprehensive and comparatively interventionist Artificial Intelligence Act. The Act came into force in 2024 and will become fully applicable in August this year.

As well as introducing prohibitions on various higher-risk AI practices, the EU's AI Act introduces transparency requirements for AI developers who train their models on third-party materials and, to that end, has also published a template form for developers to use to meet those requirements.

The form contains a forensic breakdown of the various categories of data scraping-related information which must be provided. This includes the scraper type or identifier, period of collection, comprehensive description of scraped content, modalities covered and domain names used.

The form sets a high standard of compliance for AI developers in terms of transparency and could address problems encountered by rightsholders who struggle to monitor AI-based scraping and exploitation of their data. However, the effectiveness of these measures remains to be seen and some have argued that the rigid, pro-forma approach may prove impractical in reality.

Next steps

Until legislative change and certainty arrive, rightsholders should operate strategically by:

Pursuing licensing opportunities early, particularly collective or framework licences that scale more effectively than one‑to‑one negotiation against potentially big players in the AI industry. Consider including transparency and disclosure requirements to enable more effective monitoring of a counterparty's activities.
Controlling access to sensitive or high‑value content, including through paywalls, authentication layers and other technical measures, in addition to contractual terms. Consider whether particularly sensitive or material data should be made available.
Engaging with consultations and sector initiatives, which continue to shape government thinking despite the lack of immediate reform.
Implementing technical anti‑scraping measures, such as robots.txt files, rate‑limiting, watermarking, metadata labelling and crawler detection tools, while acknowledging their practical limits.
Documenting internal AI and copyright positions, so that organisational choices can be justified.

Stability may avoid short-term disruption, but it also leaves rightsholders, creators, and institutions bearing much of the risk themselves. While AI developers need access to vast quantities of high quality and diverse data to train models – with huge potential to drive UK economic growth – rightsholders deserve to be fairly compensated for use of their proprietary works which already generate enormous value for the UK creative industry and wider culture.

For now, UK copyright law and AI remains a landscape shaped less by statute and more by contracts, courts and caution. Where do we go from here? For now at least, the UK government's position is 'we'll have to wait and see'. We hope they do not leave it too long.

This publication is a general summary of the law. It should not replace legal advice tailored to your specific circumstances.