How can AI and copyright co-exist? - UK Government re-consults

How can AI and copyright co-exist? - UK Government re-consults

The UK Government has launched a public consultation aimed at copyright law and artificial intelligence (AI). The consultation includes 47 questions on various copyright issues, including clarifying the application of UK copyright law to text and data mining (TDM). TDM refers to the use of automated techniques to analyse large amounts of information. 

Using copyright works for TDM in the context of AI model training is a contentious issue in most jurisdictions. AI developers, particularly those of larger foundation models, often require access to vast quantities of training data from works, which may be protected by copyright. Unless TDM is exempt from copyright infringement, model training may involve reproducing copyright works without permission which may infringe third parties' copyright.

The Government recognises that there is a need to address these issues to support AI innovation in the UK, while also balancing the protection for creators. Several options are presented for regulating TDM, with the preferred and proposed approach being to introduce a new TDM exception. This is despite the fact that the last attempt to introduce a broad exception for TDM for commercial purposes failed, as did attempts to define a voluntary code of practice.  A new exception for TDM in the UK would allow AI developers in the UK to use copyright materials for training purposes, provided that rightsholders have not "reserved" or "opted-out" their rights. The proposed TDM exception would align the UK's approach with the EU.

The proposed TDM exception is coupled with new measures to facilitate the licensing of training data, which will become relevant if/when rightsholders have opted out of TDM. The Government also proposes introducing transparency requirements for AI developers to disclose the training data they have used.

The public consultation also discusses a wide range of other issues, such as copyright protection for computer-generated works, infringement and liability of AI-generated content, and labelling of AI-generated content.

Interested members of the public have until 25 February 2025 to respond (instructions for how to respond can be found at Section A.2, paragraphs 19-23, here).

The proposed approach: a new TDM exception allowing rightsholders to reserve their rights, underpinned by supporting measures on transparency (paragraphs 56-81)

Currently, UK copyright law (s.29A of the Copyright, Designs and Patents Act 1988) only permits computational analysis for the sole purpose of research for a non-commercial purpose. Since most AI model training is unlikely to be solely for a non-commercial purpose, there is a risk that such training activities will infringe copyright, unless AI developers obtain a licence for reproducing each  work that is mined.  A previous attempt to introduce a broad exception for the use of copyright works for TDM for commercial purposes failed.

The Government has proposed to introduce a new TDM exception to address this, which would significantly alter the current UK position. Key features of the TDM exception include:

• It would apply to TDM for any purpose, including commercial purposes.

• It would apply only where the user has lawful access to the relevant works. This would include works that have been made available freely on the internet and those accessed without breaching any applicable contractual terms (including any paywall).

• It would apply only where the rightsholder has not reserved its rights in relation to the mined work.

• It would be underpinned by transparency requirements for AI developers, requiring them to disclose to rightsholders what training data they have used.

The proposed TDM exception would align the UK with the EU's approach under the Directive on Copyright in the Digital Single Market 2019/790 ("the DSM Directive"). Specifically, Article 4.1 of the DSM Directive requires EU Member States to provide a copyright exception for reproductions and extractions of lawfully accessible works and other subject matter for the purposes of text and data mining. Article 4.3 also provides that this exception shall not apply where the rightsholder has expressly reserved its rights in an appropriate manner, such as machine-readable means in the case of content made publicly available online.

So far, only the EU has introduced a TDM exception combined with an opt-out mechanism for rightsholders. Whilst the Government is considering  other options as well in this consultation, such as introducing a broad TDM exception (similar to what has been introduced in Singapore), or requiring mandatory licensing in all cases (which is currently the position in the UK), the EU route is referred to by the Government as a "useful precedent". A different approach, the Government states, may also be difficult to reconcile with the UK's wider obligations under international law (which presumably refers to the so-called three-step test in the Berne Convention and the TRIPS Agreement, which the UK and many other countries have signed and ratified).  The Government is also keen to align trade rules so far as possible with the EU.

Revising data mining rules as part of this consultation to make the UK more competitive (and "at least as competitive as the EU") is also part of the Government's 50-point AI Opportunities Action Plan published on 13 January 2025 (Recommendation 24). The action plan makes it clear that the Government intends to resolve the current copyright position with AI, to avoid the UK falling behind.

Technical standards for rightsholders' opt-outs (paragraphs 77-89)

The TDM exception in the EU, combined with the opt-out mechanism, was negotiated between Member States and introduced before the EU AI Act. It is currently an open question how practical this legal mechanism will be for large-scale TDM practices and where a growing number of rightsholders may reserve their rights, and in different ways. Article 4.3 of the DSM Directive does not harmonise exactly how rightsholders may reserve their rights (so long as it is done "in an appropriate manner, such as machine-readable means"), which has caused legal uncertainty about the mechanism in the EU.

The Government recognises these concerns, and considers that technical standards are needed to enable internet users to reserve their rights on the web easily, and to allow AI developers to respect these decisions.

The most widely-adopted standard today is using robots.txt, which is a file that can be used to manage crawling traffic on the internet and which AI developers generally respect. The Government refers to robots.txt as standard practice, but points out its deficiencies, such as that it covers the use of web crawlers at the site level, rather than allowing rightsholders to reserve their rights in respect of individual works. Other standards, including protocols and metadata, are further discussed and members of the public are invited to comment on the appropriate technical standard for opt-outs.

The Government notes that it intends to engage with standardisation initiatives being taken forward by industry and by other international partners, such as the EU, to ensure a collaborative approach.  The Second Draft of the General-Purpose AI Code of Practice, which was published on 19 December 2024 (for providers of general-purpose AI models to demonstrate compliance with the EU AI Act), proposes that Signatories will commit to making "best efforts that are proportionate to their size and capacities in accordance with widely used industry standards to identify and comply with, including through state-of-the-art technologies" (Measure 2.7). It is expected that the Government will be paying close attention to subsequent drafts of the Code of Practice, the final draft of which is meant to be published by April/May 2025.

Licensing of training data (paragraphs 90-98)

The Government recognises that rightsholders have complained about inequal negotiation power in the context of TDM licensing, and that they need more control. No particular measures are proposed at this stage, but the public is asked whether measures should be introduced to support "good licensing practice" (a term which remains undefined).

It is suggested that the collective licensing of copyrighted materials for TDM purposes should be facilitated, through collective management organisations (CMOs). It is not discussed how this will work in practice, however. For example, some rightsholders of literary works comprised in datasets will not belong to a particular CMO. Extended collective licensing is not specifically discussed, but is not excluded from the consultation. So far, only Denmark has introduced extended collective licensing laws in the context of AI and TDM, and there is no EU-wide harmonisation requiring other Member States to do the same.

Relatedly, the Government's AI Opportunities Action Plan also discusses creation of a National Data Library to licence public sector data assets (including potentially NHS data) as well as the establishment of a "copyright cleared British media asset training data set, which can be licensed internationally at scale" (Recommendation 13). Further details on the National Data Library and its data access policy will be published separately.  The Action Plan does not address in much detail  how the cleared media asset training data set will be accomplished, but it has been suggested that this could be done through partnering with bodies that hold valuable cultural data like the National Archives, Natural History Museum, British Library and the BBC to develop a commercial proposition for sharing their data to advance AI. 

Transparency requirements for training data (paragraphs 103-112)

The Government recognises that it is difficult for rightsholders to determine whether their works are being used to train AI models. This becomes relevant to rightsholders who have opted out of TDM for their works, if such a legal mechanism is introduced.

The Government proposes new transparency requirements for disclosing the training data used in TDM practices, which would facilitate rightsholders' seeking licences and/or enforcing their rights against AI developers and/or providers.

The Government refers in this regard to the EU AI Act and the California's Assembly Bill 2013 (AB 2013), and states that it intends to engage with the EU and other international partners as it  develops its approach and it will consider international interoperability. Specifically, Article 53(1)(d) of the EU AI Act requires AI providers to make publicly available a "sufficiently detailed summary" of training content. The nature and format of that summary will follow a template, yet to be published by the EU AI Office. It is expected that the Government will pay close attention to how the EU AI Office implements this requirement, however members of the public are invited to share their views of the EU's approach to transparency.

Other points for consultation

• Extraterritoriality (paragraphs 114-118): The Government recognises that UK copyright law does not apply where the training activities take place outside the UK, and that currently, many AI models used in the UK have been trained outside the UK. Indeed, this issue is currently being litigated before the English High Court in Getty Images v. Stability AI, where Stability AI has  argued that it cannot infringe UK copyright, if the model training activities took place outside the UK.

  The Government expresses an interest in making the UK's copyright regime internationally competitive and it wants to attract AI developers to train their models in the UK. However,  the Government also "wants to encourage AI developers operating in the UK to comply with UK law on AI model training, even if their models are trained in other countries." It is not clear if the Government is referring to voluntary or mandatory compliance with UK copyright law, extraterritorially, for foreign model training activities. The UK Government may be inspired by the EU's position, which has been interpreted as seeking to extend extraterritorially EU copyright law standards to overseas training activities (see, specifically, Recital 106 and Article 53.1(c) when read together with Article 2.1(a) of the EU AI Act).

• Clarifying the "temporary copies" exception (paragraph 119): The public is asked whether the temporary copies exception (s.28A of the Copyright, Designs and Patents Act 1988) requires clarification in relation to AI training, as some AI developers have sought to rely on this exception for training AI models, but it is not clear if it would apply.

• Protection for AI-generated content (paragraphs 126-155): It is recognised that it is possible for AI-generated outputs to be protected by copyright under current UK law (for example, as "computer-generated works" under s.9(3) of the Copyright, Designs and Patents Act 1988). The Government is sceptical about maintaining the copyright protection that may arise for AI-generated output under this provision (which was previously reviewed by the Government in 2021) as it considers that copyright protection should not be extended to works without a human author.  Further views are welcomed from the public.

• Infringement and liability of AI-generated content (paragraphs 157-162): The Government recognises that it is copyright infringement to reproduce a protected work in the form of AI-generated content, or to communicate that work to the public, without permission. Depending on the circumstances, both the end-user and the AI model provider may be liable for copyright infringement in this context . No changes are currently proposed to the existing copyright framework, which is considered "reasonably clear and appears to be adequate", but other views are welcomed.

• Labelling of AI-generated content (paragraphs 163-168): It is proposed that regulation may be needed to require AI-generated output to be labelled. This is already a requirement under Article 50 of the EU AI Act, as well as in a growing number of other jurisdictions across the world.

• Digital replicas (paragraphs 169-179): It is noted that AI tools have been used to create digital replicas (so-called deepfakes) of individuals, without their permission. The public is asked whether the current legal framework, notably the reliance on performance rights, passing off and data protection, is adequate to protect individuals.

• Synthetic data (paragraph 184): It is noted that synthetic data is increasingly used to train AI models. The Government invites the public to share their views on the implications of this, including how it may affect the licensing ecosystem.

Comment

This is a very important consultation about potential new legislation, not just for AI developers and the creative industries, but also the many businesses that are/will be using AI tools which may have been trained using third party data (where the AI developers currently provide limited warranty and indemnity protection against the risk of infringement).

After the fiasco of the last Government consultation on a proposed wide TDM exception which was abandoned after sustained criticism and lobbying from the creative industries, it remains open whether this consultation will have better success and result in actual change and, if so, in what timescale. However, what is apparent is whether clarity is provided to industry by legislative change, voluntary industry codes or from the decisions of the courts in current and anticipated litigation, this significant issue is not going to go away without intervention. There will also need to be detailed discussions around how opt-outs will work and be respected, as well as how transparency and reporting is carried out by AI developers.

What is clear from the Government's AI Opportunities Action Plan is that the Government is keen to turbocharge investment in AI development in the UK and it is concerned not to fall behind other countries by maintaining a restrictive approach to copyright law.  The UK Government has sought to balance competing interests in this space by stating in its Action Plan that it will appoint AI Sector Champions in key industries such as the life sciences, financial services and the creative industries to work with industry and Government departments.  Further details and discussions on these proposals are expected as the consultations progress and we will report regularly.