The UK Government has issued its much-anticipated consultation on Copyright and Artificial Intelligence, with the deadline for interested parties to respond of 25 February 2025. This issue has been on the agenda since even before the surge of interest in generative AI following the public launch of ChatGPT in November 2022. Having consulted on the issue in 2021, the previous Government had initially decided to introduce a broad text and data mining exception to allow scraping of copyright-protected work for any commercial purpose (including training of AI tools), without providing any option for right holders to opt their works out. However, following significant opposition from across the creative industries, it later revised its approach to focus on attempting to broker a voluntary code of practice between AI tool developers and rights holder representatives.
With those code of practice discussions having failed to reach a resolution, the new Labour Government has now issued a fresh consultation in which it seeks to reach a balance between the competing interests, and to thereby unlock opportunities for AI training in the UK, whilst also ensuring protection for creative works (described by one Minister as a "win win"). Subject to the responses it receives to its consultation, the Government proposes again to introduce a text and data mining exception allowing copyright works to be used in training, but this time making it subject to rights reservation by right holders (i.e., an opt-out). This is intended to allow them to exercise control over their works by opting them out, or otherwise licensing them for AI training and obtaining payment for their use. Underpinning this would be a requirement of greater transparency from AI developers as to the material used to train their models, how they have acquired those materials and in relation to the content generated by their models.
The responses to the consultation will indicate how straightforward it will be for the Government to ease the passage of what it no doubt sees as a fair compromise between the competing interests, and one that, as widely anticipated, seeks to align with the approach adopted in the EU. However, the Government clearly also recognises the challenges that lie ahead in relation to its preferred approach – in particular, the pressing need for simple and effective technical means to allow creators to exercise their opt-out rights, and the challenges of implementing a resolution which ultimately requires international convergence in order to be truly effective.
Text and data mining (TDM) exception with rights reservation
At its heart, the Government's 'TDM with rights reservation' proposal aims to resolve what it describes in a number of areas in its consultation document as a lack of “clarity” in the current copyright regime, which the Government suggests may currently be leading AI developers to train their models in more permissive jurisdictions. Right holders meanwhile would no doubt argue that the law on this issue is not unclear or ambiguous - indeed, the consultation document also states that copying works to train AI models requires a licence, unless an exception applies – and that the Government is therefore taking a policy decision to implement a new exception in the form of TDM, with a view to providing easy access for AI developers to the creative content they need to develop their AI models, with the onus being on right holders to opt their works out.
The key features of the proposal are as follows:
- It would apply to data mining for any purpose, including commercial purposes
- It would apply only where the user has lawful access to the works (this could be because the works are available on the internet or via a subscription)
- It would apply only where the right holder has not reserved their rights in relation to the work. Where they have done so, a licence could be required.
The Government notes the fundamental issue of trust as between AI developers and right holders, and positions the requirement of transparency as a central pillar to its proposals; indeed, it says that it will not introduce a TDM exception without it. The Government reiterates that there must be robust measures in place to ensure developers are transparent about the works their models are trained on, though further consideration needs to be given as to how to make sure these measures are effective, accessible, verifiable and widely/easily adopted. The Government notes, for example, the approach adopted in the EU AI Act requiring AI providers to make publicly available a "sufficiently detailed summary" of training content.
Also key is a standardised approach to, and proper understanding of, rights reservation mechanisms, as well as an effective resolution of their technical limitations. The EU regime already incorporates an opt-out for right holders, which requires that it be effected by machine-readable means – however, as noted in the consultation report, this requirement has been inconsistently applied, and greater clarity is therefore needed. There are also certain technical issues that would need to be considered including the inability of existing mechanisms such as robots.txt to avoid circumvention measures or to provide granular control over the use of individual works (as right holders typically do not want to have their works restricted from search indexing, as opposed to generative AI training). The issue of effect on findability is also addressed in the first draft General-Purpose AI Code of Practice under discussion in the EU.
The Government also wishes to understand in its consultation whether, and if so how, it should support the licensing framework in relation to training data, including collective licensing. This raises particularly complex issues about the position of individual authors and artists, as compared to, for example, publishing companies.
It will also need to consider carefully how to deal with models that are trained in other jurisdictions, given territoriality considerations, and more generally the challenges presented by a fragmented global approach. It notes, but rejects, the option of restricting from the UK market models that have been trained in other jurisdictions without a licence to do so. It also notes that certain jurisdictions (such as Singapore) provide broad exceptions for TDM, whilst in the US it may be allowed where it meets a 'fair use' standard. Whilst, predictably, the Government has alighted on an option which more closely aligns with the EU, in the form of TDM with rights reservation, there is a point of difference here around extra-territorial reach. Whilst the Government wishes to encourage AI developers operating in the UK (but whose models are trained in other countries) to comply with UK law on AI model training, it is not currently clear how this might be enforced, other than a suggestion of working towards international interoperability to align approaches where appropriate.
Outputs of AI models
Separate to the training of AI models, the consultation considers issues relating to generative AI outputs, both in the context of whether they may infringe copyright (as a reproduction or communication to the public) and whether AI-generated works (as opposed to AI-assisted works created by human authors) may be protected by copyright.
Infringing outputs
In relation to infringing outputs, the consultation suggests that this may not be a significant issue in practice as "although many AI models learn from copyright works, usually their outputs will not include the creative expression in those works". Of course, the issue of infringing outputs is a very live one in ongoing litigation such as Getty Images v Stability AI in the High Court, as well in a number of cases brought in the US. The consultation does recognise that "on occasion, [AI] models can and do output individual copyright works, or substantial parts of them" and that this would be an infringing reproduction or communication to the public, subject to a relevant exception applying. The consultation also notes the question that arises as to where such liability should fall, on both the user and/or the provider of the model? However, the conclusion is that the copyright framework on infringing outputs is "reasonably clear and appears to be adequate". The Government clearly also hopes that its proposals in relation to greater control for right holders, coupled with transparency, will trickle down to limit opportunities for infringing outputs.
Copyright protection for AI-generated outputs
In the previous consultation on computer generated works (CGWs), the then Government ultimately decided to make no changes to the UK law providing protection for such works, not least because it had received minimal evidence as to the potential impact of this provision. Of course, that earlier consultation came at a time before the explosion in use of generative AI for? creative outputs; but also, it is perhaps not surprising that there is limited evidence on the incentive benefits or otherwise of providing protection for CGWs, given that (unlike in the US), there is no requirement to register copyright works in order to enforce them. The new consultation highlights however the central issue around copyright protection for such works, namely the legal contradiction of requiring that CGWs, which are works 'without a human author', should nevertheless be 'original' works, with the test for originality being very closely associated with human qualities.
International alignment also provides the backdrop for this discussion. The consultation notes that many countries, including the US and EU, do not provide for copyright protection for CGWs (though it is available in India, Ireland and Singapore). The Government appears to be moving to a preferred position of removing the protection for CGWs entirely, unless it is satisfied that there is evidence of the economic benefit of providing copyright protection for CGWs. Should responses to the consultation demonstrate these benefits, it may look instead to reform the current protection to clarify its scope, for example by removing the originality requirement for CGWs, or re-defining the originality requirement.
As an aside, it is worth noting that there are similar provisions in both the registered and unregistered design frameworks relating to computer-generated designs. Whilst these have not garnered as much attention as copyright for CGWs, it is an important discussion also and we anticipate it will feature in the expected consultation on the UK's design framework, which will consider a broad range of issues.
Other issues
The consultation also touches on a number of other issues relevant to copyright and generative AI:
Temporary copies exception
The consultation notes that, given that this has been argued as a potential exception for training of generative AI models, there may be a case for clarifying the scope of this exception.
Encouraging research and innovation
The current TDM exception under UK copyright law relates to non-commercial research. In light of revisions in the EU to its TDM regime for the purposes of research by research organisations and cultural heritage institutions, the consultation asks whether the existing UK TDM exception for non-commercial research remains fit for purpose.
Labelling of AI outputs
The Government considers that clear labelling of AI outputs would be beneficial for copyright owners, services providers and consumers, but recognises that it presents technical and consistency challenges. It again looks to the EU AI Act model which requires AI outputs to be machine readable and detectable as having been generated or manipulated by AI. Labelling AI outputs in a consistent manner has the added benefit of making it easier for those outputs to be treated differently to truly original works in training data sets.
Use of AI in education
Whilst much of the debate has focused on the creative industries, the Government also notes the issues that arise in relation to copyright and the use of AI tools in the wider education system (there are of course other issues arising from the use of AI tools in education, particularly in the higher education sector, which go beyond copyright issues).
Digital replicas
Whilst not formally consulting on specific proposals on personality rights, the Government seeks views as to whether the current legal framework provides sufficient control to, for example, performers over the use of their likeness and, if not, whether further intervention is required.
It may, for example, be possible to provide the ability for individuals to reserve their rights at the input stage in order to control the use of their voice or appearance. 'Personality rights' may also give individuals greater control over their voice or likeness. Again, the Government looks here to developments in other countries for inspiration, with the US Copyright Office having proposed earlier this year a Federal Digital Replica Law.
Other emerging issues
Finally, the Government identifies certain other emerging issues, including in relation to AI models that interact with copyright works at inference (i.e., separately from the training process, such as when a user includes a copyright work in their prompt), and how the increasing use of synthetic data to train AI modelsmay also impact the copyright framework.
Next steps
The deadline for responding to the consultation is 11.59pm on 25 February 2025. We expect there to be a significant number of responses to the consultation, representing a broad range of viewpoints, with stakeholders having a range of both overlapping and diverging positions. It is likely to take some time for the Government to consider fully the responses it receives, to conduct further engagement with stakeholders, and to draft appropriate legislation where this is concluded as necessary.
Separately, we had been expecting the US Copyright Office to issue before the end of the year its reports on copyrightability for AI-generated works and on the use of copyright works for the purposes of training AI models. However, it has recently been confirmed that these will now not be published until next year.
In the meantime, Court cases continue to proliferate around the world – our Generative AI & Copyright tracker allows you to track developments in these new and ongoing cases – but the UK Government has clearly decided it cannot wait for those cases to work their way through the judicial system. The Getty Images v Stability AI case will be heard by the High Court in the Summer of 2025 – importantly, as we understand the claim to be framed, that case will consider a number of important issues that go beyond those discussed in the Government's consultation.
If you would like to discuss the consultation in more detail, and the implications of the Government's proposals, please get in touch.