Copyright and AI – UK government consults on balancing act, while seemingly backing AI

Its preferred options are: (a) to update existing rules on text and data mining (TDM) to permit it for commercial purposes if the operator has lawful access to the works (e.g. online) and the rights-holder has not reserved rights via an opt-out; and (b) to remove existing rules on computer-generated works (CGWs) altogether, relying instead on other rules for copyright subsistence. The consultation also touches on deepfakes, personality rights and labelling of AI-generated outputs, but makes no firm recommendations on those.

Responses from the creative industries have largely been opposed to the TDM proposal, but the government’s AI Opportunities Action Plan of 13 January 2025 suggests that the government may press ahead with pro-AI measures.

Summary of proposals

Text and data mining

The largest, and most impactful, part of the consultation relates to proposed changes to the current TDM permissions in the UK. Currently, the UK diverges from the EU in this respect. While in the EU the DSM Directive permits TDM for commercial purposes on an opt-out basis,^[2] the UK only provides limited permissions for TDM for non-commercial research (both being subject to a requirement of lawful access to the work).^[3] The consultation also refers to the US “fair use” approach, which is currently subject to litigation. The UK government takes the view that the UK’s position is uncompetitive and inhibits AI investment in the UK, when compared to the EU and US.

The UK government’s preferred option is to allow AI developers to train their models on works using TDM processes. The key element of the proposed approach is the introduction of a broad exception to copyright law for TDM.  It would apply to TDM for any purpose, including commercial purposes. The proposed approach would still require the user to have lawful access to the work in question, e.g. because it has been made available online or by contractual means, such as via a subscription.

The exception would only apply where the rights-holder has not reserved rights using an “agreed mechanism”, and that approach would be accompanied by measures to require increased transparency by AI companies of the works on which AI models have been trained. There are no details proposed for the form of the transparency requirements and “agreed mechanism”, and the government is seeking opinions on those.

Transparency

The transparency measures could include requiring AI developers to disclose details about the sources of content and specific materials used for training, and to provide clear records of which copyright works were accessed and used in AI training, and ensuring that rights-holders can verify that AI developers have complied with their opt-out requests.

While the EU AI Act requires providers of general-purpose AI models operating within the EU to make public a “sufficiently detailed summary” of training content, the UK government could set a more stringent standard for disclosure if it forces AI developers to publish specific works and datasets.^[4] Nonetheless, the consultation states that the UK government intends to engage with the EU on this matter and “consider international operability”.

Opt-out and EU harmonisation

The consultation simultaneously acknowledges the benefits of international alignment with the EU position while noting that the EU approach is still under development.  The DSM Directive refers to an express reservation of rights “in an appropriate manner, such as machine-readable means in the case of content made publicly available online”.^[5]

It has not always been clear what constitutes a valid reservation of rights, which could leave creatives concerned that they have no means of effectively protecting their works from use in AI training. As the consultation states, it must be “as clear as possible to all parties when a work cannot be used under the exception and what the consequences of using it will be”.

The consultation focuses on the reference to “machine-readable”, without acknowledging that there may be other “appropriate manners” in which a reservation in the EU is made. As such, rights-holders will need to wait and see where the government falls on the spectrum between a purposive versus prescriptive approach to the form of opt-outs.

As to machine-readable forms of opt-out, there is no currently preferred standard and current solutions have flaws. For example, the consultation refers to the robots.txt standard, but notes limitations to that standard, which cannot differentiate between different use cases (like search-engine indexations as opposed to AI training) and cannot be used for opt-outs beyond those at a domain level.

Overall, the approach preferred by the consultation seeks to strike a balance between the interests of rights-holders and AI developers, whereby the former can control the use of their works and seek payment through individual or collective licensing agreements, and the latter are able to train on large volumes of online material without running the risk of infringement. The consultation acknowledges the failed attempt in 2022 to expand TDM permissions and proposes a strengthened transparency and opt-out regime as a middle ground.

While the proposed approach appears both to protect the interests of rights-holders and to provide a means for AI developers to train their models without infringing copyright, it would represent a significant change in favour of AI companies from the status quo.

Computer-generated works

Beyond TDM, the consultation also tackles an old dilemma.  Section 9(3) of the Copyright, Designs and Patents Act 1988 purports to provide copyright protection to CGWs “without a human author”.  Yet the provision refers to literary, dramatic, musical and artistic works, for which copyright only subsists where they satisfy the criteria for being “original”. Under English law, it is now established that “originality” means the “author’s own intellectual creation”, reflecting the author’s “personal touch”.^[6] This could be interpreted as requiring a human author, which would conflict with section 9(3). The consultation states “it is unclear in the absence of case law how an ‘original’ yet wholly machine-authored work would be defined”.

The consultation asks for responses on this topic, and sets out the government’s preferred option is to remove section 9(3) altogether, citing a lack of economic data to support a conclusion that it encourages any investment into the UK, and arguing that existing law, such as copyright subsisting in AI-generated “entrepreneurial works”, like sound recording and films, already fulfils the purpose of the legislation.

Interestingly, the consultation notes “in the case of general-purpose AI which generates output in response to a user prompt, the ‘author’ will usually be person who inputted the prompt” [sic]. This point is not settled and is usually dealt with in the terms and conditions of the platform, and so it is interesting to see the UK Intellectual Property Office take this stance.  If the government’s preferred approach is adopted, there is a potential that this would generate a gap in copyright protection for AI-generated text and images where a human author cannot be identified. For example, it is unclear whether copyright would subsist in a chatbot’s responses in a conversation either with another chatbot or a human.  It remains to be seen whether the government will settle on taking any action in this respect.

Other points raised

The consultation has also raised some significant questions around the interaction of IP law and AI, but does not address those in any great detail, nor provide any recommendations.

AI output labelling and liability

The consultation briefly considers liability for AI output that infringes on existing copyright, noting that either the user or the provider of the AI platform could be responsible, taking the view that “the copyright framework in relation to infringing outputs is reasonably clear and appears to be adequate”, while also asking for responses on the topic.

Personality rights and “deepfakes”

The consultation also touches on the issue of deepfakes and the calls of creatives to have greater “personality rights” in the UK to protect their image or voice. The consultation makes it clear that reform in this area is not being consulted on specifically, but asks for views on whether the current legal framework remains fit for purpose.

Licensing models

Finally, an interesting point raised in the consultation is whether creatives are put at a disadvantage as they are often required to agree to standard terms and conditions of various technology platforms, such as social media platforms. Those often provide for non-negotiable licences to those platforms, which, in turn, reduces the amount of control that creatives have over their work.  For example, Reddit recently licensed the data contained on its platform, which will almost all be user-generated, for AI training purposes.^[7] The consultation does not make any specific recommendations to address such concerns, but asks: “Should measures be introduced to support good licensing practice?”

Comment

Brief consultation period

The consultation was launched shortly before Christmas, running for 10 weeks until 25 February. Given that the introduction of a broad exception for TDM would amount to a major overhaul of copyright protections in the UK, the consultation period appears brief, especially accounting for the holiday period. The UK government states that it will try to engage with stakeholders during the consultation and has pledged to publish a summary of responses, but rights-holders in particular will be concerned to ensure that their voices are heard.

“Unrealistic burden” on rights-holders

Responses to the consultation, including from the Authors’ Licensing and Collecting Society, have raised the point that a broad TDM exception with an opt-out places an “unrealistic burden” on rights-holders to keep track of, and to take action against, the unlawful use of their copyright works.^[8] There is also uncertainty as to whether the existing technical means, such as metadata instructions, for opting-out are sufficient.

Bodies representing thousands of rights-holders published a joint statement last month rejecting the proposed approach in the consultation. The Creative Rights in AI Coalition said that existing copyright laws must be respected and enforced rather than degraded.^[9]

There are concerns that the process of reserving rights, licensing works and seeking payment for them may only be beneficial to the largest rights-holders, and that small- and medium-scale creative entities or individuals are unlikely to be aware of the importance of reserving their rights.

Critics argue that an opt-out is less effective if intermediaries have already taken opted-out content and uploaded it elsewhere (whether that infringes the rights of creatives or not). The devil will lie in the detail, and rights-holders will want to know how they can enforce their rights in such a situation.

Is the writing on the wall?

The consultation forms part of the wider political landscape in the UK. On 13 January 2025, the UK government launched its AI Opportunities Action Plan. The Prime Minister, Sir Keir Starmer, said that “the AI industry needs a government that is on their side, one that won’t sit back and let opportunities slip through its fingers”.^[10]   Recommendation 12 of the AI Opportunities Action Plan states that the government should “actively incentivise and reward researchers and industry to curate and unlock private data sets”.  Further details are expected in summer 2025.^[11]

Recommendation 24 of the AI Opportunities Action Plan refers back to the consultation, recommending the UK government “reform the UK text and data mining regime so that it is at least as competitive as the EU”, specifically noting that “the current uncertainty around intellectual property (IP) is hindering innovation and undermining our broader ambitions for AI”, and that “this has gone on too long and needs to be urgently resolved”. Critics from the creative industries would no doubt argue that, in fact, the law in the UK is currently reasonably clear that using copyright materials without permission to train AI models is infringement of those rights, and so references to “uncertainty” in this space lack substance.

Nonetheless, the UK government has now indicated that it believes that alignment with the interests of AI providers is in the public interest, and the creative industries may need to learn to navigate that new landscape.