Copyright and AI: how is the UK government proposing to strike a balance between the creative industries and AI sector?

United Kingdom

On 17 December 2024, the UK government published a new consultation on copyright and artificial intelligence (“AI”), seeking views on how to ensure the legal framework in the UK for copyright and AI can support both the creative industries and the AI sector.

This is the third consultation on AI and intellectual property (“IP”) in the last five years. The 2020 call for views covered patents, copyright and related rights, designs, trade marks and trade secrets. Based on the response to that call for views, the UK government decided to consult further, in 2021, on how AI should be dealt with in respect of copyright and patents.

We’ve discussed previously (see here, here and here) the interplay between AI and copyright, and the approach(es) proposed by the UK government since the 2020 call for views to try to strike the right balance between rights holders and AI developers. However, as yet, the position under UK copyright law has not changed.

Many of the discussions to date have focussed on the use of copyright works for training AI models. While the UK government has acknowledged that this issue is the primary focus of the latest consultation, the consultation covers a number of other issues related to the interplay between AI and copyright.

The UK government considers that the current UK copyright framework “does not meet the needs of UK’s creative industries or AI sectors” and that action (starting with this latest consultation) is therefore required to deliver a copyright and AI framework that “rewards human creativity, incentivises innovation and provides the legal certainty required for long-term growth” in both the UK’s creative industries and AI sectors.

In this article, we give an overview of the issues covered by the consultation (see here for the consultation).

Although the consultation is called “Copyright and Artificial Intelligence”, the UK government has said that statements and proposals in the consultation that are said to apply to copyright will usually also apply to or have analogues in rights in performances and databases.

1. Text and data mining exception

Most AI models need to be trained on large volumes of data, often contained in works protected by copyright. If a copyright work is copied as part of that training process, permission from the rights holder is needed unless a relevant copyright exception applies.

Currently, rights holders would argue that permission is always needed where training is carried out in the UK for commercial purposes (or for a mixture of commercial and non-commercial purposes) as there is no relevant copyright exception. However, according to the consultation, some AI developers would argue that such training is covered by existing copyright exceptions or that there is no copying in their training workflows.

In practice, rights holders have highlighted a lack of transparency from AI developers about the use of copyright works, making it difficult for rights holders to enforce their copyright or receive appropriate remuneration through licensing agreements. On the other hand, the absence of an express copyright exception for training commercial AI models means that many AI developers do not train their models in the UK.

The UK government would like to establish a fair balance between rights holders and AI developers and, through the consultation, is seeking views on the following four options.

Option 0: Do nothing: copyright and related laws remain as they are.

The UK government considers that doing nothing will “prolong the legal uncertainty of AI model training” for the creative industries and AI sector.

Option 1: Strengthen copyright requiring licensing in all cases.

According to the consultation, this would mean: (a) an AI developer would always need an express licence to train AI models on copyright works in the UK; and (b) any AI firm providing services in the UK would not be able to avoid this requirement by training outside of the UK.

Although this would give rights holders more control over their works and a clearer right to remuneration, the UK government considers that this option would make the UK less attractive for AI developers and would damage the UK AI sector.

Option 2: A broad data mining exception.

This option would involve the introduction of a new exception permitting data mining (on copyright works) for any purpose (whether commercial or otherwise), including for AI training, without rights holders’ permission and with few or no restrictions.

Although this would make it easier for AI developers to access training material and would benefit the UK AI sector, the UK government considers that such an exception is “highly likely to constrain the growth of the creative and media sectors”.

Option 3: A data mining exception which allows right holders to reserve their rights, underpinned by supporting measures on transparency.

This is the UK government’s preferred approach and is similar to the EU opt-out model (introduced by Article 4 of the Directive on Copyright in the Digital Single Market ((EU) 2019/790) (the “DSM Directive”)).

This would involve the introduction of a new exception permitting data mining (on copyright works) for any purpose (whether commercial or otherwise), including for AI training. The exception would apply only to the extent: (a) the user or AI developer has lawful access to the relevant copyright work; and (b) the rights holder has not expressly reserved their rights in relation to the relevant copyright work. Where a rights holder has reserved their rights (in an accepted format), a licence would be needed for an AI developer to copy the relevant copyright work to train their AI model.

However, according to the consultation, the UK government will not introduce a new data mining exception unless “robust transparency measures” are also introduced.

The UK government acknowledges in the consultation that further work would be needed to ensure effective mechanisms for reservation of rights and measures on transparency (see below for more detail).

2. Transparency measures

The UK government considers that (increased) transparency by AI developers about the sources of works used to train AI models is key for enabling: (a) rights holders to determine if their works are being used in AI training (so that they can enforce their rights as appropriate); and (b) a better assessment of the associated legal liabilities for users or consumers intending to implement AI tools (which might, in turn, lead to an increased use of those AI tools). However, the UK government acknowledges the practical challenge this may present to AI developers and recognises that disclosure of sources is not always appropriate (for example, where it would compromise trade secrets).

The consultation seeks views on a proportionate approach to ensuring the appropriate level of transparency. This includes asking for views on the transparency requirement introduced by Regulation (EU) 2024/1689 laying down harmonised rules on artificial intelligence (the “EU AI Act”), which requires providers of general-purpose AI models to publish a “sufficiently detailed summary” about training content. [1]

According to the consultation, the UK government will consider “international interoperability” and intends to engage with the EU and other international partners as they develop their approach to transparency.

3. Reservation of rights and technical standards

The UK government decided not to implement the EU’s DSM Directive into UK law. As such, it may seem surprising that the UK government’s preferred approach for a new data mining exception is similar to the EU’s opt-out model under Article 4 of the DSM Directive.

However, the UK government considers that: (a) it is “not always clear” what constitutes a valid reservation of rights under the EU model; and (b) existing tools enabling rights holders to reserve their rights “have significant technical limitations and are insufficiently standardised and adopted”.

For works made available online, the UK government is proposing that any reservation of rights is made using “effective and accessible machine-readable formats, which should be standardised as far as possible”. The consultation also acknowledges that it may be desirable (for rights holders) if a reservation of rights in respect of a copyright work also applies to other copies of the same work.

The consultation seeks views on the approach to reserving rights, the extent to which rights reservation protocols and standards should be improved and standardised, as well as the extent to which regulation is needed to support the adoption and recognition of and compliance with such protocols and standards.

According to the consultation, the UK government intends to engage with industry and international partners in respect of relevant standards initiatives, to ensure a collaborative approach.

4. Treatment of models trained in other jurisdictions

If introducing a new copyright exception for data mining, the UK government would also like to remove any existing ambiguity in copyright law.

One of the clarifications being considered relates to AI models used in the UK but trained in a different jurisdiction. In this scenario, the copyright law applicable to the training of such AI models is the copyright law of that different jurisdiction (rather than UK copyright law), although rights holders may argue that such models will still be caught by the UK’s secondary copyright infringement regime (which is one of the arguments by Getty Images in its case against Stability AI).

The UK government is considering measures to establish a “level playing field” between providers of AI models trained in the UK and providers of AI models used but not trained in the UK. The consultation seeks views on the extent to which clarification of the copyright status for AI models trained outside the UK is needed to ensure fairness for rights holders as well as AI developers.

In particular, for AI developers operating in the UK, the UK government would like to encourage them to comply with UK law on AI model training, regardless of where their models are trained. It’s not clear from the consultation if the UK government is considering requiring, rather than encouraging, such compliance. The UK government does, however, intend to engage with international partners to ensure that the relevant provisions in UK copyright law are “internationally interoperable” and avoid creating “unreasonable burdens” for AI providers operating across multiple jurisdictions.

5. Temporary copies exception

The other clarification being considered relates to the existing copyright exception for the making of temporary copies.

This exception permits the making of temporary copies during technological processes in certain scenarios. However, the UK government considers that it is not entirely clear if the temporary copies exception applies to the training of generative AI models.

The consultation seeks views on the extent to which the scope of the temporary copies exception should be clarified in respect of AI training.

6. Contracts and licensing

In principle, rights holders can choose whether they grant a licence for the use of their work with AI. In practice, some rights holders have complained that they do not always have sufficient control over the use of their work with AI, including where terms relating to use with AI are contained in standard terms (with little or no room for negotiation) or are “bundled in with other contractual terms or framed in broad or vague language”.

The consultation seeks views on the current practice in relation to the licensing of copyright works for AI training and whether measures should be introduced to support good licensing practice.

The UK government recognises the importance of collective management organisations being able to: (a) grant collective licences to use their members’ works for AI training; and (b) reserve the rights of their members effectively.

The consultation also seeks views on how collective licensing and data aggregation/brokering services can play a role in providing access to copyright works for AI training and fair remuneration for rights holders.

7. Use of AI outside the creative industries

The UK government recognises that other sectors and industries may be affected by any changes to copyright law or policy in respect of AI.

The UK Intellectual Property Office is already intending to support the Department for Education in taking into account the views of pupils, students, parents, carers and teachers when determining the appropriate approach to copyright and the use of AI for pupils, schools and the wider education system.

The consultation seeks views on issues relating to copyright and AI relevant to other specific sectors, bodies, and individuals.

8. Research exception for data mining

UK copyright law already contains an exception for data mining. The existing exception allows a person with lawful access to a copyright work to copy it to carry out a computational analysis of anything recorded in it, but the exception is not available unless the analysis is “for the sole purpose of research for a non-commercial purpose”. [2]

Article 3 of the EU’s DSM Directive contains a different research exception for data mining. For example, the EU exception is not limited to non-commercial research, extends to databases (rather than being limited to copyright works), and applies to research organisations and cultural heritage institutions (whereas the UK exception applies to individual researchers and research institutions).

The consultation seeks views on the extent to which any changes to the current UK exception are needed.

More generally, the UK government recognises that, as with the approach adopted in the EU AI Act, regulatory requirements may depend on the size of the AI firm or the purpose and effect of the relevant technology. The consultation seeks views on whether a similar approach should apply for UK copyright law relating to AI. Although the question is included in the section on the current UK exception for data mining, it seems to apply more generally.

9. AI output labelling

The UK government considers that labelling AI outputs is beneficial to rights holders and enables consumers to make more informed choices. However, it acknowledges the practical and technical challenges for AI developers to ensure consistent labelling and to ensure labels are resilient to manipulation.

The UK government also recognises that there are questions over: (a) the degree or level of AI which should trigger any labelling requirement; (b) how AI outputs should be labelled; and (c) the information that should be provided.

The consultation seeks views on a proportionate approach to ensuring that AI outputs are labelled as AI generated. This includes asking for views on the transparency requirements introduced by the EU AI Act, which (in general) require the detection and labelling of AI-generated or AI-manipulated audio, image, video or text content.

10. Computer generated works

As discussed previously (see here), UK copyright law contains specific protection for literary, dramatic, musical or artistic works which are “computer-generated” (in other words, generated by computer in circumstances such that there is no human author).

To be capable of copyright protection, such works must be original. However, it’s not clear how an AI-generated work would satisfy the current legal test for originality (which requires a work to be the “author’s own intellectual creation”, reflecting their “personal touch”). [3]

According to the consultation, there are also arguments that the protection of computer-generated works is not needed because such protection: (a) has “little or no positive incentive effect”; and (b) is “to the detriment of human creation”.

The UK government consulted in 2021 on whether to change the law in relation to computer-generated works. Commenting that use of AI was “still in its early stages”, the UK government decided not to make any changes to the law but said it would keep this under review.

The UK government has now decided to revisit this issue in the current consultation, seeking views on three options:

Option 0: No legal change, maintain the current provisions.

The UK government considers that the current copyright protection for computer-generated works has little (if any) effect on the production of content or on AI development, and that retaining the current provisions may perpetuate legal uncertainty around their scope. As such, the UK government is proposing to reject Option 0, unless responses to the consultation suggest otherwise.

Option 1: Reform the current protection to clarify its scope.

This option would involve the UK government clarifying the existing copyright protection for computer-generated works, whether by removing the originality requirement altogether or by clarifying when a computer-generated work can be considered original.

Option 2: Remove specific protection for computer-generated works.

If, following responses to the consultation, there is insufficient evidence of positive effects from the protection for computer-generated works, the UK government’s preferred approach is to remove such protection.

UK law on registered and unregistered designs mirrors the protection for computer-generated works, setting out who is the author or designer of a computer-generated design. However, as the current consultation is limited to the interplay between copyright and AI, any changes to UK designs law will be considered separately in the future.

11. Liability for infringement of copyright by AI output

We have discussed previously (see here) whether AI-generated output can infringe someone else’s copyright.

The UK government considers that the current UK copyright law is reasonably clear and adequate in relation to copyright infringement by AI outputs. However, the consultation seeks views on any gaps or deficiencies in UK copyright law or any barriers to effective enforcement of copyright, in each case in relation to infringement by AI outputs.

The consultation also seeks views on the steps providers of generative AI services should take to avoid AI outputs infringing copyright, such as filtering prompts and outputs.

12. Digital replicas

Currently, there is no specific legislation in the UK for the protection of personality rights. (Here, we refer to “personality rights” (sometimes referred to as “image rights” or “publicity rights”) as an individual’s right to prevent the (unauthorised) use of their voice, image, likeness or other aspect of their personality.) Instead, there is a patchwork of causes of action which can be used to protect personality rights but only in certain scenarios.

The current consultation does not contain any specific proposals on personality rights, but the UK government recognises that some individuals would like more control over whether content can be created that includes their likeness or personality.

As such, those responding to the consultation are invited to share their reviews on whether, under the current legal framework, individuals have sufficient control over their personality.

The consultation also seeks views on the use of AI to create “digital replicas” or deepfakes (meaning “images, videos and audio recordings created by digital technology to realistically replicate an individual’s voice or appearance”). In particular, the consultation asks if enabling individuals to reserve their rights for copyright works at the input stage would provide them with sufficient control over the use of their image and voice at the output stage.

13. Emerging issues

The UK government recognises that developments in AI will continue to raise new questions about the application of UK copyright law.

While the consultation focuses on the “most urgent and contested” issues, it also seeks views on how to address such emerging questions.

For example, it seeks views on: (a) the implications of synthetic data being used to train AI models; and (b) whether it is sufficiently clear how UK copyright law applies to AI products (which have already been trained) interacting with copyright works in order to generate outputs (including where a user prompt includes a copyright work or where an AI product interfaces with ‘live’ data sources as a result of user prompts).

Next steps

The UK government is keen to ensure that the full range of views is heard, describing this consultation as an opportunity for anyone with an interest in the issues summarised above to share their views. It’s obviously important that stakeholders from both sides of the AI/copyright fence respond to the consultation, as the UK government will use the consultation responses to shape its proposed approach to the relevant issues.

The consultation runs until 11.59pm on 25 February 2025. Individuals or organisations wishing to respond to all or part of the consultation can do so online (see here) or by email (to [email protected]).

 

[1] Article 53(1)(d) of the EU AI Act

[2] Section 29A of the UK Copyright, Designs and Patents Act 1988

[3] This is the EU test for originality, introduced in Infopaq C-5/08. In a recent UK case, it was confirmed (by the Court of Appeal in THJ Systems Ltd v Sheridan [2023] EWCA Civ 1354) that this is the correct test to apply in the UK.