UK Government’s Consultation on Copyright and AI: What’s Next for AI Developers and Creators?
Role of the Copyright and AI Consultation
The UK government's Consultation on Copyright and Artificial Intelligence (AI) has recently concluded. This marks a pivotal moment in the intersection of technology and Intellectual Property (IP) rights. This Consultation aimed to address the challenges posed by developers using third-party copyright within data sets to train AI models. We explore what this means for UK IP law.
Why was a Consultation to review UK copyright law needed?
The UK government is keen to highlight the opportunities for AI, as outlined in its Action Plan paper published in January 2025. AI Opportunities Action Plan - GOV.UK. The plan concludes that increasing AI adoption across the UK will boost economic growth, create jobs and improve lives.
As part of these opportunities, the government has focused on the need for AI technology developers having access to ‘high-quality materials’ to train AI models.
Exclusive rights of a copyright owner
Under the Copyright, Designs and Patents Act 1988 (CDPA), a copyright owner has certain exclusive rights, including the right to reproduce, adapt and distribute the work (section 2(1) CDPA). Any output generated by an AI model will infringe copyright in the UK if it reproduces or adapts a “substantial part” of a protected work.
AI Model Training
Foundation models (FM) and large language models (LLM) are typically trained on three sources of information, data and content (content):
- Publicly available content;
- Third-party content; and
- User generated or supplied content.
Unless the content used in any of these three situations has been appropriately licensed by the copyright owner or rights holder, using it for training a commercial FM or LLM within a generative AI model (Gen AI) is highly likely to infringe copyright. Whilst there are exemptions under the CDPA to use content in some situations (discussed below), these do not apply to commercial use.
Substantial parts of a protected work in outputs
In many instances the output from a Gen AI model will not include a “substantial part” of a copyright work. For instance, a prompt in Midjourney to create an image of a horse against a hill backdrop is unlikely to generate an exact copy of a specific horse or hill image used in training. However, the output might still incorporate elements from the training material.
There is evidence that AI models do produce works which have been copied from training materials. For example, a prompt asking for an image of an athlete which looks like the runner, Keely Hodgkinson can generate an image of an athlete wearing a top with a logo which is protected by copyright, and which appears to be based on existing images of Keely Hodgkinson.
In the Hodgkinson example, both the user of the AI model and the provider of the AI model may be liable for copyright infringement (along with potential trademark infringement in any logo, and potential image rights infringement). As there is no transparency over the copyright works used to train AI models, establishing copyright infringement for owners and rights holders is challenging especially given the vast size of the data sets being used.
Location of data set training
Copyright is territorial. As such, the Government has suggested the Consultation should address the complexity for developers who frequently need to train AI models on copyright protected works in multiple locations. Models which are trained on copyright protected works created in the UK will be governed by the CDPA. However, the government is of the view that to maintain competitiveness globally, there is a need for international interoperability to encourage all AI developers to operate and utilise high-quality copyright protected works from multiple jurisdictions including the UK.
Temporary copies and research
The CDPA contains an exemption for temporary copies (section 28A CDPA) made during technological processes, such as copies held in browser caches. These must be ‘transient or incidental’ and have no ‘independent economic significance’, which is unlikely when works are used to train a Generative AI model or are created as part of an output. The Government is keen to clarify suitable exemptions which support AI innovation and encourage research .
Text and data mining (TDM)
The CDPA contains a limited exemption to copying which permits TDM where there is lawful access and for non-commercial research. The Consultation explores whether this exemption should now be widened to adopt a similar approach to the EU Digital Copyright Directive, where data can be scraped and mined for scientific research purposes (Article 3*) and for commercial purposes where there is lawful access and no express reservation by the rights holder in an ‘appropriate manner’, such as machine readable opt-out (Article 4*). In addition, the EU AI Act 2024 includes obligations on developers to ensure transparency and compliance with EU copyright law including any reservation of rights under the EU Digital Copyright Directive. As a result, where rights holders in the EU reserve rights over their works to prevent TDM ‘in an appropriate manner’ developers and providers of Gen AI models need to obtain express authorisation from rightsholders if they want to carry out TDM.
The proposal under the Consultation seeks to align the UK position on TDM with the EU position and the Government has suggested it expects there to be a standard technological solution for rights holders to adopt to show they have reserved their rights. The government is confident these technical measures will be available and has recently stated that it will not ‘legislate until the tech companies can prove that the technology can deliver the transparency that they have said that they can, and that we will find ways for the creative arts industry to make money in the digital age’. At this stage, we are yet to see concrete proposals to achieve this.
This issue has led to many in the creative industries calling the proposals for an opt out to TDM as ‘unworkable’ and ‘shortsighted’**.
AI outputs, digital replicas, and deepfakes
Another area of review is around the status of Gen AI outputs and whether these are all protected by copyright.
Works such as literary, dramatic, musical, and art work require originality (section 1(1)(a) CDPA), which traditionally involves human input to meet the originality threshold. When Gen AI tools are used to edit or create these works, the Government views the outputs as protected by copyright, provided the underlying work is original. This approach is supported by most Gen AI terms, which state that outputs are generally owned by the user.
Works such as sound recordings, films, broadcasts and published editions do not require originality to be protected by copyright. Regardless of this distinction, the Government seems to have taken the same approach and assumed these outputs are protected by copyright. In reality, Gen AI sound recordings are more likely to be protected by copyright due to the lack of any originality threshold. However, the point is that the strict interpretation of the CDPA does not necessarily align with the Governments approach.
The CDPA creates specific protection for literary, dramatic, musical or artistic works which are computer generated works (CGWs). The author is the person “by whom the arrangements necessary for the creation of the work are undertaken” (section 9(3)). If Gen AI is used, this is the person who has made the prompts, again subject to any terms stated in the AI tools used. The Government Consultation focuses on section 9(3) and the perceived contradiction between the use of original works (literary etc) which require a human ‘personal touch’ and references to works without a human author being included within section 9(3) as copyright protected works. The Consultation suggests this contradiction should be resolved to ensure ownership in CGW, including prompts and outputs, is clear.
Copyright is aligned to moral rights (such as the right to object to derogatory treatment of your work) and rights to privacy. The Consultation has considered the legal ramifications of digital replicas and deepfakes, which can infringe upon copyright and personal rights, posing ethical and legal challenges.
AI’s Impact on International Copyright law
Any changes to the CDPA could infringe international law, specifically the Berne Convention for the Protection of Literary and Artistic Works. The UK, as a signatory, must ensure that any changes to its copyright laws remain compliant with the Convention's principles. This includes the protection of authors' rights and the recognition of works irrespective of their form or method of creation. The Government has recently sought to reassure rights holders in the creative industries that the new copyright regime will be ‘fully compliant’ with international law.
Rights holders are unable to easily enforce their rights
The Consultation explores concerns that rights holders in the creative sectors struggle to effectively charge for their works and enforce their IP rights. This is partly due to the lack of transparency from AI developers on how works are used to create data sets for FM and LLMs, and because IP infringement claims are complex, cross-jurisdictional, and involve vast amounts of data. The interim judgment from the Getty Images (US) v Stability AI Ltd case in the US shows how difficult and complex these cases will become.
What does the UK Government’s Consultation on Copyright and AI Consultation propose?
The UK government's Consultation on AI and copyright has considered four options to address the legal uncertainties surrounding AI model training and copyright..
Option 0: Do Nothing
This option would leave existing copyright laws unchanged. The Government acknowledges the current lack of clarity, which poses challenges for both rights holders and AI developers. Rights holders struggle to enforce their rights and seek remuneration, while AI developers face legal risks, particularly impacting smaller firms and new entrants.
Option 1: Strengthen Copyright Requiring Licensing
This option proposes that AI models could only be trained on copyright works in the UK with an express licence. While it would provide a clear route to remuneration for creators, it risks making the UK less competitive compared to jurisdictions like the EU and US, potentially reducing investment in the AI sector. It could also limit innovation and consumer choice, as some of the most capable AI models might not be available in the UK.
Option 2: Broad Data Mining Exception
This option suggests introducing a broad exception allowing data mining on copyright works without permission, including for AI training. While it would enhance access for AI developers and boost investment, it would not meet the needs of rights holders, who would lose control and remuneration opportunities. Such changes could impact the growth and survival of creative and media sectors and may conflict with international legal obligations.
Option 3: Data Mining Exception with Rights Reservation and Transparency
This approach aims to balance access for AI developers with control for rights holders, supported by transparency measures. AI developers could train on materials unless rights holders expressly reserve their rights. It would enable licensing agreements and remuneration opportunities, whilst ensuring transparency over model inputs and outputs. This option is aligned with some aspects of the EU approach. It is not clear how rights holders would expressly opt out and reserve their rights for historic and new works.
The Consultation seeks to explore these options in detail, considering their implications for innovation, rights holder control, and compliance with international standards.
Initial views on the Options
Practical Implementation of Opt-Out Measures for AI Model Training
The Consultation explores the feasibility of implementing opt-out measures for creators who do not wish their works to be used in AI training datasets. Such measures would require robust mechanisms to ensure that creators can effectively control the use of their copyrighted materials, balancing the interests of innovation and protection. Questions remain as to how these technical measures enabling opt out will be applied to existing and new works and why the presumption of consent has been altered to favour developers and providers of AI. There is also a question of whether these technical measures go far enough in that they also fail to offer a technical method of compensation for those who have not opted out.
Commercial Considerations
The Consultation is considering how AI technologies could undermine traditional revenue streams for creators and whether new business models could emerge. High-profile cases, such as those involving Getty Images, highlight the existing tensions between AI developers and copyright holders, underscoring the need for clear guidelines on enforcement, revenue and rights management. The timing of any legislative changes to the CDPA also requires consideration as many rights holders are in extensive talks with AI providers to license their rights; dramatic changes to the CDPA could undermine these negotiations.
Culture, Media and Sport (CMS) Committee and the Science, Innovation and Technology Committee – Joint Letter – Fair Remuneration
Since the Consultation there has been a joint letter from the CMS Committee and the Science, Innovation and Technology Committee which has emphasised the need for fair remuneration for creators whose works are used in AI training.
Next Steps for UK Copyright Law
The UK government's AI Consultation on copyright law represents a critical juncture in addressing the challenges posed by AI technologies. As the legal landscape evolves, it is imperative to strike a balance between fostering innovation and protecting the rights of creators, ensuring that the UK remains at the forefront of creativity and technological advancement.
For now, the CDPA remains but we await the Government’s next steps once it has digested the detailed responses to the Consultation.
* DIRECTIVE (EU) 2019/790 on Copyright And Related Rights In The Digital Single Market and amending Directives 96/9/EC and 2001/29/EC
** Quote from Sony Music in Times article on 9 March 2025
FAQs
What was the purpose of the UK Government’s AI Consultation on Copyright?
To address concerns raised by the creative and tech sectors around the training of data sets for generative AI models. The purpose of the Consultation is to:
- Supporting right holders’ control of their content and ability to be remunerated for its use.
- Supporting the development of world-leading AI models in the UK by ensuring wide and lawful access to high-quality data.
- Promoting greater trust and transparency between the sectors.
What are the potential legal implications for AI developers using copyrighted works?
AI developers are at risk of copyright infringement by using copyrighted works as training material for their AI models and also infringement of trade marks (such as a logos) being used in the output of AI models.
What are the key differences between the UK and EU approaches to AI and copyright?
The EU has a wider exception for text and data mining which allows developers to scrape data and text for commercial purposes providing that the rights holder has not expressly reserved their rights. Currently in the UK, developers are only permitted to scrape data and text for non-commercial purposes.
Under the EU AI Act, developers must be more transparent about their data sources that they are using to train their AI models.
What are the challenges for rights holders in enforcing their copyright against AI-generated works?
There is a lack of transparency from AI developers on the training materials used in AI models. Copyright infringement is therefore difficult to enforce as the rights holder must show that a ‘substantial part’ of their work has been copied in the AI-generated works. This is difficult to show when such vast amounts of data are being used to train these models.
Infringement claims can often be complex as they can involve multiple jurisdictions. Enforcement proceedings are not aligned in all jurisdictions, and this can therefore lead to increased costs and uncertainty for the rights holder attempting to enforce their rights in the works.
