The UK government has launched a consultation on copyright and artificial intelligence which will run for 10 weeks, closing on 25 February 2025. While the consultation considers a range of approaches, the key proposal is to permit wider use of copyright materials in the UK to train AI models, while respecting the rights of content owners who wish to “opt out” of this regime. This mirrors the current position in the EU and appears to be part of a trend towards closer cooperation with the EU under the new Labour government.

Background

While the consultation paper suggests that the application of UK copyright law is subject to “legal uncertainty”, there is in fact currently no defence to unauthorised copying of third-party content for text and data mining or other activities required to train commercial AI models. The current UK position is therefore a somewhat anti-AI innovation / pro-content owner position. However, many content owners are also dissatisfied with the current position as, while their IP rights are protected in theory, they need greater transparency from AI developers on their training datasets so that they can monitor and enforce their IP rights in practice.

The UK has been talking about AI and copyright reform since 2021. Its initial (2022) proposal was to introduce a broad exception to allow commercial text and data mining (with no opportunity for rightsholders to opt out), but this was quickly scrapped following a backlash from the creative industries (see our update here). The UKIPO was then tasked with bringing industry stakeholders together to come up with a code of practice re use of copyright content to train AI models, but ultimately no agreement could be reached (see our update here).

With all these false starts, the UK’s approach to copyright and AI has fallen behind. While the intersection between AI and copyright has been developing at pace in other countries (including the US and the EU) whether through legislative developments and/or a docket of case in the courts considering the key issues, the position in the UK has been in stasis. This consultation is therefore very welcome.

Proposals

The consultation describes four options:

• Option 0 - Do nothing: This would maintain the status quo but do nothing to meet the objectives of ensuring control for rightsholders, access for AI firms and transparency between parties.

• Option 1 - Strengthen copyright: The consultation paper says that under this option, models could only be trained on copyright works in the UK with an express licence to do so (as is the current legal position). Interestingly, it also suggests that firms providing services in the UK could not get around his requirement by training their AI models in other countries. However, this doesn’t appear to be a serious proposal in any event as the government quickly points out in the consultation that this would make the UK uncompetitive.

• Option 2 – A broad data mining exception, with no ability for rights holders to reserve their rights: This is the approach proposed by the conservative government in 2022 that met with great opposition from IP rights holders and was quickly shelved. There is presumably little appetite to go down this route again.

• Option 3 – A data mining exception which allows rights holders to reserve their rights, backed up by increased transparency measures: This option would allow AI developers to train on materials to which they have lawful access, including large volumes of web-based material, but only to the extent that rightsholders had not expressly reserved their rights. This would be underpinned by measures to ensure developers are transparent about the works their models are trained on.

The consultation notes that further work would be needed to ensure right reservation and transparency standards are effective, accessible and widely adopted. The consultation paper strongly supports Option 3 and puts it forward as the government’s proposed approach. Given the unattractiveness of the other options, it seems very likely that this is where are headed.

Notably, Option 3 essentially mirrors the EU regime. The EU’s Directive on Copyright in the Digital Single Market provides a text and data mining exception which is wider than that available in the UK (in that it applies also to commercial activities) but allows rightsholders to expressly reserve their rights - meaning that AI models cannot be trained on their materials despite this exception. The EU has tried to turn this EU-wide approach into a de facto worldwide approach through the EU AI Act which requires¹  companies placing AI systems on the market in the EU to ensure that they were trained via a process that respected rightsholders’ reservation of rights (no matter the country in which the AI training took place). AI developers also have transparency obligations regarding their data sources under the EU legislation.²

The UK now seems to be minded to fall in line with this approach. This is consistent with recent suggestions from government that a closer relationship with the EU is in the UK’s interests. It is also commercially sensible as varying copyright rules between the UK and its nearest neighbours is likely to cause difficulties for AI providers and rightsholders alike.

The consultation also recognises that the EU approach is still being developed; that there remains some uncertainty regarding the practical application of rights reservation methods; and that it is unclear whether this approach has led to further licensing. The UK government therefore proposes to explore this approach and the various implementation options in more detail.

The consultation also examines the UK’s approach to the outputs of AI models, particularly whether copyright arises in such outputs. The position here also varies between territories which isn’t ideal, although there is no evidence this is causing significant issues at present. Currently, UK law provides copyright protection for “computer-generated works”. This is anomalous when compared to other major jurisdictions (including the US and most of the EU), and the UK government’s favoured option as described in the consultation is to remove the specific protection for computer generated works – unless the consultation reveals evidence of its positive effects. This sounds sensible given it is unclear what benefit is provided by that protection at present, and given the apparent desire to bring UK copyright law more closely in line with the EU approach.

Comment

This consultation is a good opportunity for industry stakeholders to comment on some of the more detailed technical questions raised in the consultation concerning the technical tools used to reserve or restrict rights (such a “robots.txt” or metadata flags or new “do not train” registries) and whether standardised practices should be encouraged. Responses to this consultation might also credibly inform EU developments given the nascency of the EU regime.

1. EU AI Act, Article 53(1)(c)

2. EU AI Act, Article 53(1)(d)