On 17 December 2024 the UK Government launched its consultation on potential changes to UK intellectual property law and other laws to accommodate, at least in part, the challenges posed by the use of AI in the UK.
The consultation closed on 25 February 2025. In this blog, we look back at the Government’s key proposals and consider what might come next in AI regulation within the UK.
Key pillars
AI technology relies on and can create a vast amount of data, especially in the case of ‘generative’ AI tools that can produce new text, image or video content from a pre-existing data set or new content introduced by an end-user.
In very short summary, the two main focal points of data use by an AI model (and its developer) are the training data and output data.
The first key pillar of the consultation proposed a new mechanism to clarify the law around how AI developers can access and use training data: rights owners would have to actively opt-out of having their works used for AI training purposes.
The second key pillar of the consultation focused on a proposal to potentially alter how AI-generated output data is treated under intellectual property law and to consider new rules applicable to AI output.
Training data
Potential new exemption under copyright law for AI development
Background
The term training data relates to the original information used to curate and improve the algorithms within AI models. By using 1000s of images of a dog, an AI model can “understand” what a dog looks like – a feature that can be used for various purposes by the AI or the other software in which it sits.
Intellectual property rights (in particular, copyright) can exist in any image. Therefore, especially in the context of ‘large language models’, it is commonly alleged that of the 1000 dog photos, there will be at least one rights holder that did not licence their dog photo to the developer for AI training purposes (if at all). Using an intellectual property right without a licence can amount to an infringement under copyright law in the UK.
If a creator uploaded their image to a media database, which does allow for AI training by third parties, the image can find its way into training datasets even if the original agreement between the creator and media company did not contemplate or permit this use. More commonly, however, the work is available on the internet and the creator is never made aware that their work has been collected or has been used for AI training. End users can also upload content directly to AI software in the form of a command (input data). If the input data is used for training, the issue proliferates. Neither the AI developer nor the content creator fully appreciate what is going on or who is doing what. The issue of transparency over the exploitation of copyright is key – no one will understand their rights and obligations if AI developers don’t know what they are using or are unable to comply with requests for information pertaining to the data that they have used for training purposes.
Still, some AI developers continue to train their models – both under express licence agreements and in scenarios (such as web scraping) where the lawfulness of their conduct is less clear (but not defacto unlawful). Others do not even attempt to navigate the lack of clarity, choosing jurisdictions where the legal position is clearer.
The consultation suggested this state of affairs as a lose-lose for everyone. Creators and rights holders are not getting properly renumerated and lack control of their rights. AI developers, willingly or unwillingly, use copyright protected content on a large scale – potentially exposing themselves to legal liability even with the best intentions. For AI developers based overseas, the UK economy loses out and it is more difficult for a creator to bring enforcement action against them.
Reformed text and data mining (TDM) exemption
UK copyright law already permits use of a copyrighted work without a licence for TDM, as long as the purpose is non-commercial in nature (e.g. scientific research).
TDM is defined by the UK Intellectual Property Office as ‘use of automated analytical techniques to analyse text and data for patterns, trends and other useful information. Text and data mining usually requires copying of the work to be analysed.’
The consultation proposed to expand the TDM exemption to cover commercial use for any purpose (including training), assuming that:
1. The relevant user (the AI developer) already had a lawful access to the work in question. This would cover publicly available websites or online registries. The latter point raises an interesting issue – the current TDM exemption in the EU exempts use of both copyrighted works and databases whereas the UK TDM exemption covers only copyrighted works. The consultation recognised that potential reform would be needed in this area, as a large portion of relevant data would sit within what UK copyright law would view as a database. Databases are another form of intellectual property right in which protection may exist.
2. The copyright owner hasn’t reserved their rights in the relevant work. If this has occurred (a so-called opt-out), the user would have to seek a licence directly.
Rights reservation mechanism
The consultation made clear that ‘giving right holders the ability to reserve their rights under the exception means that they will be able to prevent use of their works for AI training. If a developer were to copy a work to train their AI model, despite a right holder expressly reserving their rights in that work, then this would infringe copyright.’
In other words, the idea is that there would be an automatic opt-in for rights holders to allow AI developers to train on their copyright protected works. If the rights holder opts-out, this would not be permitted.
The consultation’s suggestion raises many unanswered questions. For example, it is unclear whether rights holders could object just to training but permit usage for other purposes (such as non-commercial research purpose).
The consultation made clear that the responsibility for the opt-out will rest with the ‘rights-holder’ – but often, especially in the creative industries, the original creator is not the rights holder. As such, if a large music publishing house does not opt-out – the original creator still loses control and, potentially, further renumeration. This point forms the core of recent high profile objections to the consultation’s proposal, despite the Government recognising the potential imbalance of power within the consultation.
Accordingly, the consultation put forward the idea of applying the concept of collective management organisations. In the creative industries, for example, rights holders of musical works licence them to Phonographic Performance Limited (PPL). PPL manages these collective licences and licences them to businesses such as bars and clubs and collects the fees due to the rights holders.
The consultation suggests that one way to manage collective licencing for the purpose of AI training would be to create a new collective management organisation, similar to PPL. AI developers would gain access to training data directly from that organisation, and the organisation would pay the relevant licence fee back to the rights holders.
Technical standards and transparency
The key feature of implementing an opt-in/opt-out regime is an ability to easily communicate whether the rights holder opts in or opts out. Collective licensing is one way to access data, but it would not apply to other works as they appear on the internet or other databases.
The consultation put forward three potential solutions:
1. Adoption of robot.txt as the main opt-out mechanism. Robots.txt is a file that is publicly displayed on the backend of the website, which can work to reject the automated crawling (data collection) of the website’s contents (insofar as the crawler respects this).
2. Metadata within the protected work. It is possible to insert terms within the code of the electronic file to indicate whether the rights holder has opted in or opted out.
3. Opt-out databases. The consultation made reference to public databases that can publish the rights holder’s opt-in or opt-out position.
The consultation recognised that there are obvious issues in respect of each approach. Notably, despite permitting TDM for commercial use in the EU, no widely accepted standard has been adopted to communicate the rights holder’s position. It is unclear whether the UK Government could implement or adopt a solution that would make sure the rights holders’ opted-out status is respected by AI developers.
Accordingly, the final supportive measure put forward by the consultation to facilitate a new TDM exception is to require increased transparency from AI developers ‘and others conducting text and data mining.’ The potential effect on such entities would be to disclose the ‘use of specific works and data sets’ or ‘to evidence compliance with rights reservations.’
As the consultation recognised, this has parallels with the transparency requirements placed on developers of general-purpose AI models under the EU AI Act, although, to meet the objectives of the consultation, it is likely that any change to UK law would have to go further.
Output data
The second key part of the consultation focuses on potential changes to the law of copyright and other related issues in respect of content created by AI.
Potential changes to the law of copyright
The consultation referred to three situations in which copyright can arise in relation to AI-generated content:
1. ‘AI-assisted content’ – if a human creator uses AI as a tool to create a work, provided that the original creativity required comes from the human and not the AI. This will potentially qualify for copyright protection, which would be owned by the relevant human involved.
2. ‘Entrepreneurial works’ – newly created sound recording, films, broadcasts or published editions of literary, musical or dramatic works do not require originality. AI-generated music could qualify for copyright protection. The relevant owner is the ‘person who made arrangements for it to be produced’.
3. ‘Computer generated works’ – literary, dramatic, musical and artistic works generated by AI without direct human oversight could qualify for a lesser type of protection under copyright law, provided that these works are original. The consultation considers that the user that inputted the relevant prompt to create the work would be the owner, although this is an area of much contention that was not addressed in the consultation.
The consultation highlights the legal issues with computer generated works. The requirement for originality refers to, in law, human creativity (which AI cannot, in law, possess). But a computer-generated works, by definition, are created without the involvement of a human author.
The consultation discusses whether it is worth clarifying the law around computer generated works.
It suggested that there is no real evidence of organisations relying on the computer-generated work provision in the context of AI, whilst at the same time acknowledge that reform to the definition might achieve value for users of AI tools. Despite this, the Government’s stated preference is to remove the definition of a computer generated work from UK copyright law. It remains to be seen why doing so would be a useful exercise, as opposed to updating the law to reflect the requirement of human involvement in the creation of AI-generated content.
Output labelling
The EU AI Act will require that AI output used within the EU is labelled as such. Such labelling ideally informs the public, prevents misinformation and enhances consumer choice.
The consultation also invited the public’s views on whether AI labelling should be adopted in the UK context, recognising the technical challenges to achieving effective labelling.
From this aspect of the consultation, it appears that AI labelling in some form will arrive in the UK by law.
Digital replicas
There is no straightforward ‘personality right’ in the UK. In the US context, for example, public figures can prevent the use of their image/voice/personality without authorisation.
With generative AI, there has been a rise in AI tools that mimic humans or use their voice or image without authorisation. In some contexts, unauthorised use can be prevented – singers could use their performance rights to control the use of the sound recording. But a straightforward ‘deepfake’ that uses the personality of a celebrity may be more difficult to control under the UK intellectual property law.
The consultation requested the public’s views as to whether the current framework is adequate. There is a hint that the Government is open to creating a new, full blown, personality right in the UK.
Conclusion
There has been recent speculation that the UK’s ‘AI Bill’ – a contemplated piece of legislation to directly regulate AI in the UK – will be placed before Parliament in the summer. Previously, it was expected to arrive sometime in Spring 2025.
Accordingly, it is more likely that the UK Government will have time to produce a formal follow up to its consultation prior to any ‘AI Bill’ being available for public criticism. This would bring welcome clarity to the Government’s proposals, if any are implemented.
At EM Law, we are experts in the legal issues surrounding AI. If you have any legal questions about your development or use of an AI tool, please do not hesitate to contact us here.
