Generative AI and copyright: where are we now?

When the UK Government published its report on copyright and artificial intelligence in March 2026, many expected it to provide a clear answer to one of the most difficult questions now facing copyright law: when, and on what terms, may copyright works be used to train generative AI systems?

Four months later, the significance of the report looks slightly different.  It did not settle the issue.  Nor did it simply preserve the status quo.  Instead, it marked a shift in the way the UK is (currently) approaching the problem.

The Government has moved away from the idea that AI and copyright can be resolved through one legislative lever, such as a new text and data mining exception.  It now appears to view the issue as a broader systems problem, involving transparency, licensing, technical standards, enforcement, outputs, and the future treatment of AI-generated works.

That makes the March report less a conclusion than a staging post.  The UK has paused immediate reform, but the surrounding legal and commercial landscape has continued to move. 

From proposed exception to policy reset

The background is familiar.  The Government’s earlier consultation had proposed a package of measures intended to balance the interests of AI developers and rights holders.  At the centre of that package was a possible new copyright exception for text and data mining, which would have allowed copyright works to be used for AI training unless rights holders reserved their rights.

That would have taken the UK closer to the EU model under Article 4 of the DSM Directive, where commercial text and data mining is permitted subject to an opt-out.

The March 2026 report stepped back from that approach.  The Government confirmed that it would not immediately expand the UK’s TDM exception.  The current UK exception remains limited to non-commercial research under section 29A of the Copyright, Designs and Patents Act 1988.

That was the headline. But it was not the whole story.

What was more striking was that the Government did not replace the abandoned TDM proposal with any clear alternative.  It did not endorse a compulsory licensing regime; it did not propose a statutory collective licensing solution.

In effect, the Government rejected the idea that any of the available models is ready to carry the full weight of reform.

The current UK position

The formal legal position in the UK is therefore unchanged.

There is no broad copyright exception allowing commercial AI training.  There is no general ‘fair use’ doctrine.  In principle, copying copyright works for training purposes requires permission unless an existing exception applies.

However, that statement only takes matters so far.  The unresolved questions remain difficult.  They include whether particular stages of AI training involve acts of copying under UK law, whether any copies are transient or incidental, how UK copyright applies where training activity takes place outside the UK, and when outputs may reproduce a substantial part of an earlier work.

For rights holders, the issue is not only whether rights exist, but whether they can be enforced in practice.  Without meaningful visibility over training datasets, it may be difficult to know whether works have been used, let alone to prove infringement.

For AI developers, the absence of reform preserves uncertainty.  A cautious developer may seek licences, but licensing markets remain fragmented and incomplete. Not all works are available to license.  Not all rights are easy to identify.  Not all rights holders are organised collectively.

The existing copyright framework remains in place, but it is being asked to deal with technical and commercial realities for which it was not designed.

The report’s wider message

To treat the March report as being only about text and data mining would miss much of its significance.

The report identifies a wider set of interlocking issues.

Transparency: Rights holders frequently don’t know whether their works have been used in training datasets.  That lack of visibility affects both licensing and enforcement.  Any meaningful system, whether based on consent, opt-out, or remuneration, depends on some level of information about what has been used.

Licensing: The report recognises that licensing may form part of the solution, and there are already commercial deals between AI developers and content owners.  But the market is uneven.  Large publishers, image libraries and music businesses may be able to negotiate. Individual creators and smaller rights holders may not.  That raises obvious questions about bargaining power, representation and market access.

Technical infrastructure: Even without adopting an opt-out regime, the Government remains interested in machine-readable rights reservation, metadata, content credentials and similar tools.  These may become important whether the UK ultimately chooses a rights-reservation model, a licensing model, or some combination of the two.

Outputs: The copyright debate is not limited to training inputs. AI-generated material may infringe existing works, imitate style, or compete with human-created content.  The Government has also looked at labelling and watermarking of AI-generated outputs, although these tools raise their own practical and technical questions.

Enforcement: Copyright protection is only meaningful if rights can be exercised effectively.  In the AI context, enforcement may involve evidential problems, jurisdictional issues and difficulties in attributing responsibility between model developers, deployers and users.

In view of the above, the March report is not merely a refusal to legislate.  It is an acknowledgement that copyright reform alone may not solve the problem.

Computer-generated works and digital replicas

The report also looked beyond training and licensing.

One issue is the UK’s existing protection for computer-generated works.  The UK is unusual in expressly recognising copyright protection for certain computer-generated works; rather than leaving such works without an author, section 9(3) CDPA deems the author to be the person who undertook the arrangements necessary for the work’s creation. That provision long predates modern generative AI.

The Government has not proposed immediate reform, but its position is more definite than a simple watching brief.  The March report proposed that the specific protection for wholly computer-generated works under s.9(3) should be removed, while copyright protection would be retained for works created with AI assistance.  That proposal remains unimplemented, but it narrows the field of likely outcomes considerably.

Another issue is digital replicas: AI-generated uses of a person’s voice, image or likeness.  This is not purely a copyright issue.  Passing off, performers’ rights, data protection and privacy may all be relevant, but none provides a simple general personality right of the kind found in some other jurisdictions.

The Government has not committed to introducing such a right and it would be too strong to say that a new UK personality right is expected imminently.  However, the Government indicated alongside the March report that it intends to launch a separate consultation on digital replicas, examining how to address harms caused by their unauthorised creation while protecting legitimate innovation.  That consultation was anticipated for summer 2026 and will be watched with interest by performers, musicians, actors and public figures whose voices or likenesses may be replicated without consent (see a recent example HERE).

The courts: Getty v Stability AI

In the absence of legislative reform, the courts are increasingly important.

The leading UK case remains Getty Images v Stability AI.  The High Court judgment was important, but it did not answer every question many had hoped it would answer.  By the time judgment was given, the case had narrowed, and the court was not required to decide the full legality of training a generative AI model on copyright works.

The litigation has now moved to the appeal stage.  Permission to appeal was granted on certain copyright issues, including issues relating to secondary infringement and the meaning of ‘infringing copy’.  That matters because many AI models are developed outside the UK but made available in the UK.  The appeal may therefore shed light on how UK copyright law applies to models and software entering the UK market after training has taken place elsewhere.

Even so, the appeal is unlikely to solve every issue.  The law may develop incrementally, through particular disputes and particular factual patterns.  That is a very different process from comprehensive legislative reform.

The EU comparison

The contrast with the EU is increasingly important.

The EU has already legislated for text and data mining under the DSM Directive.  It has also introduced the AI Act, which entered into force in August 2024.  Obligations for general-purpose AI model providers began applying from August 2025, including requirements to maintain technical documentation and to publish a summary of the content used for training.

Separate transparency obligations concerning the labelling of AI-generated content apply on a different timetable, with those provisions becoming applicable from August 2026.

The EU approach is not free from uncertainty.  Questions remain about how rights reservations must be expressed, how AI providers should identify and respect opt-outs, and how much detail training-data summaries must provide.  But the EU has chosen to create a structure first and refine it through implementation.

The UK has chosen a different route.  It is preserving flexibility while waiting for litigation, licensing markets, technical standards and international practice to develop.

Neither approach is obviously risk-free.  The EU may face difficulties in implementation.  The UK may face continuing uncertainty.  But the divergence is now clear.

Four months on: what has changed?

Since March, the central point has become sharper.  The UK Government has paused, but the underlying pressures have not.

The Getty appeal continues to matter.  The EU AI Act is no longer a distant prospect for general-purpose AI providers.  Licensing deals are developing, but unevenly. Creative-sector pressure remains strong. Technical tools for transparency and rights reservation continue to evolve.

The Government has not said that copyright law is fit for purpose. Nor has it said that AI developers should be given a broad statutory freedom to train on copyright works.  Instead, it has effectively said that the system needed to support reform is not yet sufficiently mature.

That is a cautious position, but not a neutral one.  It leaves rights holders with strong formal rights, but ongoing enforcement challenges.  It leaves AI developers without a broad exception, but also without definitive answers on the boundaries of existing law.

Where next?

The next developments are likely to come from several directions rather than one.

The Court of Appeal may provide important guidance in Getty v Stability AI.  The EU’s implementation of the AI Act may influence expectations around transparency.  Licensing markets may mature, particularly in sectors where rights are concentrated and commercially valuable.  Technical standards may make rights reservation and content identification more workable.  Further pressure may also build around digital replicas and the absence of a general UK personality right.

For now, the UK’s position is best described as controlled uncertainty.  The Government has chosen not to legislate prematurely, but it has also recognised that the existing framework is under strain.

Four months after the March report, the question is therefore not simply whether the UK will introduce a TDM exception. The larger question is whether the UK can build a functioning copyright ecosystem for generative AI at all.

That ecosystem is still taking shape.

All the thoughts and commentary that HLaw publishes on this website, including those set out above, are subject to the terms and conditions of use of this website.  None of the above constitutes legal advice and is not to be relied upon.  Much of the above will no doubt fall out of date and conflict with future law and practice one day.  None of the above should be relied upon.  Always seek your own independent professional advice.

Humphreys Law