Introduction
As AI continues to evolve, the intersection of AI and copyright law remains a hot topic for content creators and technology platforms alike. One of the key areas of focus is the EU’s general text and data mining (“TDM”) exception (contained in the EU Digital Single Market (“DSM”) Directive)[1], intended to provide a safe harbour for AI developers, while providing content creators with a right to opt out from their works being used to train AI. The EU AI Act[2] contains copyright-related provisions referring to this general TDM exception, but there remains significant uncertainty in respect of its application to TDM and the opt-out mechanism.
In December 2024, the UK government published a consultation on copyright and AI which is open for comment until 25 February 2025. The UK government is seeking views on a range of topics but the primary object of the consultation is to consider whether the UK should follow the EU and adopt a similar TDM exception.
This clear policy ambition was further confirmed by the publication in January 2025 of the UK’s “AI Opportunities Action Plan” authored by Matt Clifford working with the Department for Science, Innovation & Technology (“DSIT”). In this plan, DSIT sets out to adopt a “pro innovation” approach and specifically recommends that the UK government “reform the UK text and data mining regime so that it is at least as competitive as the EU”.[3] The UK government has responded to this recommendation by referring to its consultation and stating that it will act to ensure the UK has a competitive regime supporting both the AI sector and creative industries.[4] However, there is concern within the UK’s creative industries that inclusion of this recommendation means that the UK reform of TDM is a fait accompli.[5]
In this article, we discuss some of the issues with the practical application and implementation of the EU’s general TDM exception, as highlighted in discussions by the Working Party on Intellectual Property (Copyright) of the Council of the EU (the “EU Working Party”), and then consider the approach being considered by the UK government in the context of its current consultation.
The EU’s general TDM exception: in force but requiring guidance to ensure consistent implementation
Across the member states of the EU, Article 4 of the DSM Directive requires national governments to implement a TDM exception for commercial purposes (“General TDM Exception”) (in addition to a TDM exception for scientific research under Article 3 of the DSM Directive). Article 4 says that the exception does not apply where the use of the relevant work has been “expressly reserved by their rightholders in an appropriate manner”. Although Recital 18 of the DSM Directive states that, in the case of content made publicly available online, it is only appropriate to reserve rights using machine-readable means, the DSM Directive does not prescribe the exact way in which rights should be reserved. As such, while Member States were required to offer an exception within the scope of Article 4, they had some discretion in how to implement the “opt out” from the exception.
In December 2024 the Hungarian Presidency of the Council of the EU hosted a policy discussion to explore EU member states’ perspectives on the relationship between generative AI and copyright. On 20 December 2024, the EU Working Party published a summary of their discussions.[6] The publication gives an insight into member states’ concerns on important issues, such as: (i) how the General TDM Exception applies to AI training, (ii) how to exercise the opt-out mechanism effectively, and (iii) concerns about transparency and the inability of rights owners to verify whether their opt-out is being respected.
Unsurprisingly, a significant number of member states mentioned the need for more clarity and certainty as regards the practical application of the General TDM Exception and the opt-out mechanism introduced by the DSM Directive.
On the issue of training of AI models, several member states were of the view that copyright uses for AI training went beyond the scope of the General TDM Exception, arguing that the General TDM Exception only permits acts of reproduction and extraction, but doesn’t extend to the later publication or commercialisation of copyright works as part of the output of a generative AI tool.
In addition, a number of member states raised concerns about the lack of guidance for rights owners who wish to exercise their right to opt out in respect of certain copyright works, pointing to issues such as identifying the “machine readable” opt-out form to be used, and how such opt-out might be exercised in respect of copyright works already published online.
Under the EU AI Act, providers of general-purpose AI (“GPAI”) models are required to: (a) put in place a policy to comply with EU law on copyright and related rights, and in particular to identify and comply with an opt-out from the General TDM Exception; and (b) to publish a “sufficiently detailed” summary of the training content for the GPAI model. The European Commission is currently calling for tenders in relation to a study to assess the feasibility of a registry supporting the “effective expression” of TDM opt-outs by rights owners and facilitating their identification by AI developers.[7] The AI Office is in the process of creating a template for this summary, although a draft has not been made publicly available yet. GPAI model providers can rely on codes of practice to demonstrate compliance with these obligations, until a harmonised standard is published. A draft GPAI code of practice has been published, with the final version expected in April 2025.
However, the EU AI Act’s requirement for a summary of training content has not fully addressed concerns held by member states about an information deficit, with many member states of the view that rights owners do not have the relevant control tools to verify whether their reservation of rights is being respected. In addition, a number of member states also referred to the challenge presented by the fact that the majority of AI systems are trained outside of the EU and questioned how EU solutions can apply to such systems. However, one member state referred to recital 106 of the EU AI Act, which says that any AI providers who place their GPAI models on the EU market will be caught by the obligation to put in place a policy to comply with EU copyright law and comply with opt-outs from the General TDM Exception, regardless of where the training took place.
While member states expressed a variety of views, there was support for the introduction of practical tools to overcome the challenges of implementing a practical and efficient opt-out, including the development of a code of practice by the AI Office under the EU AI Act (as mentioned above) or the introduction of common standards. No consensus was reached on the creation of an EU-wide database to facilitate the opt-out process, with concerns that this could be unduly burdensome for rights owners given the huge quantities of works and the need to keep such database updated.
The UK consultation: what clues does it provide about how the UK government might legislate and balance the competing interests of content creators and AI developers?
The UK consultation paper published in December 2024 considers three main options when evaluating how to clarify UK copyright law to meet the challenges presented by AI innovators. The UK government argues that the current lack of clarity is unacceptable so doing nothing (although listed as Option 0 in the consultation) is not an option in practice. This clearly signals reform of UK copyright law is on the way, and the choice being presented by the UK government is tripartite:
Option 1: strengthen copyright law by requiring comprehensive licensing; or
Option 2: legislate for a broad TDM exception which would allow data mining of copyright works including for AI training without rights owners’ permission (and with few or no restrictions). This would mirror the approach provided by some pro-innovator countries such as Singapore (or the US, where there is a “fair use” caveat – which does not come without litigation risk) and would be widely welcomed by the tech community but risks constraining the growth of the UK’s important creative and media sectors; or
Option 3: provide for a TDM exception which seeks to balance the need for certainty among the AI developer community while giving rights owners the ability to reserve their rights underpinned by transparency measures.
A full analysis of the consultation’s proposals can be found in our Law-Now here.
Having read the consultation and the many column inches of commentary from a range of stakeholders, it does appear that the UK government is favouring option 3, which is similar to the EU’s General TDM Exception. However, there are some important differences, including in relation to a number of outstanding issues still being grappled with by our EU neighbours.
Scope of the proposed UK TDM exception
The UK government envisages a broad exception which would apply to data mining for any purpose (whether commercial or otherwise). However this exception would only apply to someone who has lawful access to the copyright work, either because a work is available online or via a paid subscription. Further it would only apply to works where there has not been an express reservation of rights by the rights owner in an accepted format. Where these requirements are not met, the AI developer would need a licence to copy the relevant work to train their model.
Currently, there is a TDM exception in the UK contained in s29A of the Copyright, Designs and Patents Act 1988 (“CDPA”). This is a relatively new exception to copyright infringement, which is available where someone has “lawful access” to copyright works, however only for non-commercial research. When contrasting the TDM exception proposals in the UK 2024 consultation with those of the 2021 consultation, it seems that the previous criticisms by the creative industries have been taken into account. Both consultations include as an option a copyright exception allowing TDM for any purpose without giving rights owners the ability to “opt out”. However, the favoured option in the current consultation (option 3) represents a compromise between the current narrow TDM exception under the CDPA and the broader TDM exception that the UK government had proposed introducing following the 2021 consultation (before lobbying from the creative industries forced the UK government to abandon this plan).
Practical implementation of an opt-out
The UK government reinforces the need to ensure rights owners have an effective means to opt out from having their copyright works used for AI training. Some commentators in the creative sector have expressed dissatisfaction that the onus is on rights owners to ensure that their rights are not infringed (rather than the potential infringers themselves). Of course, the effectiveness of any TDM exception in protecting rights owners relies on content creators being aware of the TDM exception and taking the required steps to “opt out”. It is interesting that the UK government expressly states that it wants to ensure a rights owner who has opted out has the ability to negotiate directly with an AI developer regarding a licence for particular activities while potentially precluding use for others. In addition, the UK government acknowledges that it may be desirable (for rights owners) if an opt-out in respect of a copyright work also applies to other copies of the same work, although this would create a practical burden for AI developers.
The UK government also recognises the need for development of common technical standards to allow for standardisation which would deliver benefits to AI developers as well as conferring on rights owners the means to manage and police the use of their copyright works (and the consultation seeks suggestions on how to tackle this). Where possible, tools should enable rights owners to distinguish between different types of uses, such that a content creator might permit web crawling for search indexing but opt out of use for generative AI. The UK consultation references the invitation-only workshop organised by the US-dominated Internet Architecture Board in Washington DC in September 2024 which considered mechanisms to communicate the opt-out choice of content creators, including a discussion about extending the robots.txt standard to communicate more granular opt-out choices.
Transparency concerns
In its consultation, the UK government acknowledges concerns expressed by rights owners about the lack of transparency and considers the possibility of requiring AI developers to disclose the sources of works used for AI training. Some proposed transparency measures are set out in paragraph 108 of the consultation and include disclosing the detail of web crawlers and requirements for AI developers to evidence compliance with rights reservations. The risk that any UK reform may go beyond what is currently required by the EU AI Act in terms of transparency has caused some concern among the tech community. Given the global nature of AI, of paramount importance to AI developers is harmonisation: for the UK and EU to agree a single standard for transparency which applies throughout Europe and beyond, not least given the complexity of the legal regime applicable in Northern Ireland.
Conclusion
Organisations have until 25 February to respond to the UK consultation. Although the UK government has expressed its preferred option for a new UK TDM exception, we’ll have to wait to see how the UK government responds to the consultation. We predict that there will be significant political pressure to adopt the same or a substantially similar approach as in the EU. However, the UK government may decide to take steps to address some of the issues identified with the EU approach.
[1] Directive (EU) 2019/790
[2] Regulation (EU) 2024/1689
[3] See the AI Opportunities Action Plan here
[4] See the UK government’s response to the AI Opportunities Action Plan here
[5] See, for example, the statement from Dan Conway, CEO of the Publishers Association here
[6] See the summary of the EU Working Party discussions here
[7] See the call for tenders here
Social Media cookies collect information about you sharing information from our website via social media tools, or analytics to understand your browsing between social media tools or our Social Media campaigns and our own websites. We do this to optimise the mix of channels to provide you with our content. Details concerning the tools in use are in our Privacy Notice.