A major UK government consultation on how copyrightable material can be used to train AI models, proposing an exception to copyright law for AI training for commercial purposes, has been welcomed as a “step towards clarity” by one senior IP law specialist and a “genuine best effort” by UK publishers. 

The core proposal on training would permit AI companies to use copyrighted material without permission, although rightsholders would still be able to “reserve their rights”, essentially to opt out so that they could still prevent the use of their content to train the models.

By reserving their rights, rightsholders could still license their works for AI training, the government said, giving them more control over how their content is used and supporting them to strike licensing deals. This would also give AI developers greater certainty about what material they can and cannot use and ensure wide access to material in the UK.

The government hopes the plans will give certainty to the creative industries and AI developers in this area as uncertainty about how copyright law applies to AI is “holding back both sectors from reaching their full potential”.

Ralph Arrate, data protection, AI and cyber security partner at UK law firm Spencer West, welcomed the new proposals saying it was a “much needed step towards clarity in the complex relationship between copyright and AI”.

Arrate said that under current law, training AI with such material often constitutes unlicensed use, potentially infringing copyright holders’ rights.

He added: “While licensing deals provide a model for fair remuneration, the lack of uniform legal frameworks creates uncertainty for both creators and developers.”

Dan Conway, CEO of the Publishers Association said the new consultation was a “genuine best effort by ministers at a compromise position between tech and the creative industries”, but pointed out that there had been no “objective case made for a new copyright exception, nor has a water-tight rights-reservation process been outlined anywhere around the world”. 

He acknowledged a fix was needed but added that the “solution needs time and full scrutiny, and a great deal is at stake”.

The consultation also proposes new requirements for AI model developers to be more transparent about their model training datasets and how they are obtained. For example, AI developers could be required to provide more information about what content they have used to train their models. This would enable rights holders to understand when and how their content has been used in training AI. 

Additionally, the consultation addresses other emerging issues, including copyright protection for computer-generated works and the issue of digital replicas.

The government said that previous attempts to agree a voluntary AI copyright code of practice had been unsuccessful and it was “determined to take proactive steps with our creative and AI sectors to deliver a workable solution”.

Announcing the proposed measures, Peter Kyle, secretary of state for science, innovation and technology, noted:  “It’s clear that our current AI and copyright framework does not support either our creative industries or our AI sectors to compete on the global stage.

“That is why we are setting out a balanced package of proposals to address uncertainty about how copyright law applies to AI so we can drive continued growth in the AI sector and creative industries.”

Lisa Nandy, secretary of state for culture, media and sport, said:  “These proposals lay the groundwork for rights holders to strike licensing deals with AI developers when rights have been reserved. For example, a photographer who uploads their work onto their internet blog could reserve their rights, with confidence that their wishes will be respected and generative AI developers will not use their images unless a licence has been agreed.”

She added: “The consultation also recognises issues related to the protection of personality rights in the context of digital replicas, such as deepfake imitations of individuals, and will seek views on whether the current legal frameworks are sufficiently robust to tackle the issue”.   

Global approach 

Arrate noted that to be effective, any response to copyright use in the context of AI model training should be international. “Copyright law is treaty-based law, and a national response would not be very effective,” he said, adding that the consultation was an opportunity for the UK to “lead globally, crafting a framework that respects intellectual property while unlocking AI’s transformative potential”.

Adam Gilbertson, managing associate at IP firm Mathys & Squire, said the proposed solution “on the face of it moves the UK’s position towards that set out in current EU regulations”. 

He noted, however, that “this is by no means a foolproof solution – as evidenced by the ongoing disputes over how the EU regulations should be interpreted. As always, the devil will be in the details of any new legislation”.

The US has already tried to tackle the issue of AI and copyrightable material. In April, the Generative AI Copyright Disclosure Act was introduced to the US House of Representatives. This would require AI companies to disclose all of the copyrighted work they used to train their generative AI models.

The UK consultation will run for 10 weeks, closing on 25 February 2025.