In January 2025, we reported on the UK Government's Copyright and Artificial Intelligence consultation, which proposed introducing an EU-style text and data mining exception allowing AI developers to train models on copyright-protected works unless rights holders actively opted out. The article is accessible here.

That proposal is now off the table.

The Government's March 2026 Report on Copyright and Artificial Intelligence confirms that the opt-out model is no longer its preferred way forward. For businesses operating in the publishing, music, media and AI sectors, this provides an important degree of certainty: UK copyright law continues to offer greater protection to rights holders than many had anticipated, and licensing remains the primary route to accessing content for commercial AI training.

Against that backdrop, this article examines what the Government's report means in practice, the opportunities emerging for rights holders, and how AI developers can manage legal risk while the courts continue to grapple with the issues raised by generative AI.

Fair dealing, not fair use

Much of the debate around AI training has been influenced by developments in the United States, where copyright disputes are assessed through the broad and flexible doctrine of fair use. Recent US cases have demonstrated just how unpredictable that analysis can be, with courts reaching different conclusions depending on the circumstances of the training activity and the source material involved.

The UK position is considerably more restrictive.

Unlike US fair use, UK fair dealing operates through a closed set of statutory exceptions. The Government's report expressly acknowledges that the absence of a broad exemption places the UK among the jurisdictions offering stronger protection for rights holders and less flexibility for AI developers.

As matters stand, copying copyright works for dataset creation, pre-processing or model training will generally require permission unless a specific exception applies. In practice, only the limited exceptions for temporary copying under section 28A of the Copyright, Designs and Patents Act 1988 (CDPA) and non-commercial text and data mining under section 29A are likely to be relevant. There remains no general exception for commercial AI training.

The Government's decision reflects the consultation responses it received. While the proposed opt-out regime was intended to strike a balance between innovation and rights protection, it attracted very limited support. Most respondents rejected the opt-out proposal, largely because creative industries feared it would undermine the value of their work and saw an opt-out as impractical, while some AI/research respondents thought it too restrictive relative to other jurisdictions. On licensing itself, many stakeholders in fact argued government should not legislate on licensing at all, preferring the market to develop with transparency requirements as the enabling mechanism.

Getty remains the key litigation to watch

The most significant UK case to date remains Getty Images v Stability AI, which continues to shape the legal debate around AI training and copyright.

One of Getty's arguments was that Stability AI's trained model itself constituted an "infringing copy" under section 27(3) of the Copyright, Designs and Patents Act 1988. Getty contended that the statutory definition was broad enough to capture an article created through an infringing process, even if the finished article did not itself contain the copyright works. The High Court rejected that interpretation, finding that Stable Diffusion's model weights do not store or reproduce the training images.

However, Getty has been granted permission to appeal on this issue, with the Court recognising that it raises a novel and unresolved question of statutory interpretation with potentially significant implications for AI models more generally. By contrast, Stability AI's application for permission to appeal the trade mark infringement findings against it was refused, meaning those findings currently remain in place.

For now, the Government has deliberately chosen not to legislate around the issues raised by Getty. Instead, it is allowing the courts to develop the law, meaning the outcome of the appeal is likely to have greater practical significance for businesses in the short term than any immediate legislative reform.

Licensing remains the commercial opportunity

The absence of a statutory training exception preserves a strong legal position for rights holders, but legal rights alone do not create commercial value. Businesses able to organise, package and license content effectively are likely to be best placed to benefit from growing demand for training data.

The licensing market is already beginning to take shape. To date, most publicly announced AI licensing deals have involved news publishers, with image libraries and academic publishers accounting for a smaller but significant proportion of agreements. Reported deal values suggest that substantial revenues are available where rights ownership is clear and content can be licensed at scale.

That opportunity is not confined to major media organisations. For smaller publishers, creators and content owners, collective licensing may provide a more practical route to market. The Copyright Licensing Agency, together with sector bodies including Publishers' Licensing Services and the Authors' Licensing and Collecting Society, is already exploring collective licensing solutions tailored to generative AI training.

Rights holders should also consider implementing technical rights-reservation measures now. Existing UK copyright law already protects technological protection measures and rights management information, and the Government has expressed support for industry-led standards designed to communicate licensing preferences in a machine-readable form. Tools such as enhanced robots.txt protocols and the emerging Really Simple Licensing framework may become increasingly important in both enforcement and licensing negotiations.

Transparency also remains a critical commercial issue. Although the Government has stopped short of mandating disclosure of training datasets, it recognises that greater transparency assists rights holders in understanding how their works are being used and enforcing their rights where necessary. Businesses entering licensing negotiations should therefore continue pushing for meaningful transparency obligations as part of any agreement.

Practical considerations for AI developers

For AI developers, the report provides clarity in one important respect: the Government is not introducing a broad statutory defence for commercial training activity.

Businesses should therefore assume that training, fine-tuning or processing copyright-protected material in the UK will usually require a licence unless a specific exception applies.

Equally important is the need to avoid treating offshore activity as a complete shield from UK copyright claims. The Getty litigation shows that a UK court will scrutinise whether an AI model trained overseas can itself be an "infringing copy" when imported into the UK, even where no copy is stored in the model. That question is now before the Court of Appeal, so its scope remains unsettled. Organisations should maintain clear records of training data provenance and be able to explain what a model was trained on, how information was obtained, and what the model can and cannot reproduce.

The Government has, for now, rejected calls to extend UK copyright law more aggressively to offshore-trained models deployed domestically. However, it has not ruled out future reforms and continues to monitor developments in other jurisdictions, particularly the EU.

Similarly, the Government has declined to impose any form of compulsory licensing or statutory levy scheme. For AI developers, that creates an opportunity to secure access to valuable datasets on commercial terms now, rather than waiting for future regulatory intervention.

Particular attention should be paid to fine-tuning and retrieval-augmented generation (RAG) systems. Unlike large-scale pre-training that may occur overseas using broad datasets, these activities frequently involve more targeted use of specialist content and are often carried out closer to the point of deployment. As a result, they are likely to represent a greater area of UK copyright exposure and should be prioritised when assessing licensing requirements.

A period of continued uncertainty

The Government's report also identifies two areas for potential future reform. First, it proposes removing copyright protection for works generated entirely by AI without a human author, while retaining protection for works created with AI assistance. If implemented, this could affect businesses currently relying on copyright to protect wholly machine-generated content.

Secondly, the Government is exploring how best to address AI-generated "digital replicas" of a person's voice or likeness, including the possibility of introducing a new personality or digital replica right. While no consensus has yet emerged on the form such protection should take, businesses operating in media, entertainment, advertising or AI-generated voice and avatar technologies should monitor developments closely, as they could result in an entirely new statutory right rather than an extension of existing copyright or passing off protections.

The Government's report ultimately adopts none of the options consulted upon. Rather than selecting a preferred model, it has opted for further evidence gathering while monitoring the Getty appeal and developments under the EU's emerging AI copyright framework.

That means businesses on both sides of the debate cannot afford to wait for legislative certainty. The more immediate challenge is operational: understanding rights ownership, implementing appropriate technical measures, negotiating licences where required and maintaining sufficient records to demonstrate compliance.

Key takeaways

• UK fair dealing, not US fair use, remains the legal framework governing AI training in the UK.
• The proposed opt-out text and data mining exception has been abandoned, leaving no imminent statutory exemption for commercial AI training.
• Rights holders should focus on licensing opportunities, monitor collective licensing developments and implement technical rights-reservation measures where possible.
• AI developers should assume that UK copyright issues may arise wherever UK content is used in training, fine-tuning or RAG systems, and should maintain clear records of training data provenance.
• The most significant developments to watch are the Court of Appeal's decision in Getty Images v Stability AI and the Government's ongoing work around transparency and industry standards.

If you would like to discuss the implications of these developments for your business, or explore content licensing strategies and risk management approaches for AI development, please contact a member of our Intellectual Property team.