Current UK copyright laws align with a global consensus that materials protected by copyright cannot be used to train AI models like OpenAI’s ChatGPT and xAI’s Grok.
However, this hasn’t stopped vast swathes of materials from finding their way into datasets used to train AI models, with rightsholders largely left unable to prevent their works from being appropriated.
AI developers, however, argue that access to copyrighted materials is vital to improve model development, with OpenAI famously telling a UK House House of Lords Committee it would be “impossible” to train models to the level of its then-flagship GPT-4 models without using copyright materials for training.
The government’s consultation is looking to provide “legal certainty required for long-term growth in both sectors” and is looking for stakeholder views to potentially find a solution that would encourage AI innovation while also protecting human creativity.
Subscribe today for free
'Levelling the playing field'
At the centre of the consultation sits two ideas:
Creating a new exception for AI model training that would allow protected materials to be used for training purposes.
Introducing a new mechanism for rightsholders that would force AI companies wanting to use their materials for AI training to obtain a licence.
Both concepts would “balance rightsholders' ability to seek remuneration while providing a clear legal basis for AI training with copyright material so that developers can train leading models in the UK while respecting the rights of right holders,” the ministerial forward outlining the consultation reads.
Both would likely be unpopular with both sides of the coin — some rightsholders would be unhappy that their works would be able to be used for AI training at all, with some arguing in lawsuits against firms like OpenAI that the models could be used to create potentially competing content.
Meanwhile, AI development firms would likely be unhappy about having to pay to access materials they once freely scrapped from the internet — especially considering the likes of OpenAI are already forking out millions of dollars in licensing deals with Reddit, News Corp, and the AP to access their content for use in training.
John Buyers, head of AI at the law firm Osborne Clarke told Capacity that other aspects of the proposals would still create burdens for AI developers, such as the obligation to make public what copyright content they have used to train their systems.
Buyers said: “Even if the proposed changes do go through, the government's attempt to balance the interests of both sides may mean that it will remain more attractive for AI developers to train AI models in countries with lower protections for content owners.
“We have been here before: the previous government's consultation led them to propose a much wider exception which would have allowed tech companies to train AI without allowing IP rights owners an opt-out, but this had to be shelved following a chorus of protests from the creative industries. It remains to be seen how this new proposal will be.”
We’ve been here before
This isn’t the first time a UK government has considered altering copyright laws in favour of AI training.
The Conservative Government of Boris Johnson introduced plans in the summer of 2022 that would have relaxed UK copyright laws to provide AI researchers with access to use protected materials for research purposes without rightsholder permission.
The move to allow researchers free reign to mine copyright materials was reversed by Rishi Sunak’s Tory government in February 2023 following significant outcry from rightsholders and creative groups.
Technology trade association techUK welcomed the Labour government’s attempt to find a solution to the AI and copyright conundrum, calling for an outcome that “creates a competitive and supportive environment for both for the UK tech and creative sectors”.
"The development and use of AI presents enormous opportunities for all parts of the UK economy,” a techUK statement reads. “However, the current uncertainty over AI and copyright risks holding back both the development and use of AI technology, not least in the creative industries themselves. It is in everyone’s interests to resolve these issues in a way that supports innovation and growth.”
Ralph Arrate, a data protection, AI and cybersecurity partner at law firm Spencer West said this latest consultation must strike a balance between creatives and innovators.
“To foster innovation without undermining creators’ rights, any reform must prioritise transparency and enforceability,” Arrate said. “AI firms should be required to disclose the datasets used for training, ensuring compliance with copyright law.
“Simultaneously, streamlined licensing mechanisms — perhaps through collective licensing or standardised agreements — could enable creators to benefit financially while supporting AI development.
AI, copyright, comic books & lawsuits
The issue of AI and copyright predates the swathe of interest in the subject following the introduction of ChatGPT in November 2022.
As early as 2019, the World Intellectual Property Organisation (WIPO), the United Nation’s IP body, held its first policy conference on the subject in London to identify issues related to AI and rights, covering copyright as well as patentability.
The topic came to a head some years later when the US Copyright Office granted protection to a comic book created using AI-generated images.
Kris Kashtanova, the creator of Zarya of the Dawn, initially received full protection for work in 2023 as the agency was unaware that it included AI-generated art.
After a lengthy review, the Copyright Office let Kashtanova keep some of the protections, limiting the scope to the work but not the generated images.
Since then, we’ve had court rulings in China that found AI-generated images were original enough to obtain copyright protection, and works falling out of copyright being quickly turned into AI training data, like in the case of Mickey Mouse progenitor Steamboat Willie.
As the UK looks to examine the issue with an in-depth review, Robert Guthrie, an IP partner at Osborne Clarke, told Capacity that the proposal truly aims to uncover a balance for both rightsholders and AI developers.
"Many will see these proposals as being AI friendly given the new text and data mining exception would provide AI developers with a new copyright exception they could rely on,” Guthrie said. “However, other aspects of the consultation seem focused on protecting copyright owners' interests when they have reserved their rights.
“In particular, the consultation asks: ‘To what extent does the copyright status of AI models trained outside the UK require clarification to ensure fairness for AI developers and right holders?’
“This is one of the central issues in the Getty v Stability AI case before the UK courts at the moment. In other words, if a model is training in a jurisdiction which may be more AI friendly (which may include the US) but which would have infringed copyright if that training took place in the UK, then should the distribution and use of the AI model in the UK be prohibited under UK copyright law.”
Stock image giant Getty is attempting to take on the startup commercialising the AI image generation model series Stable Diffusion, arguing the model used its library of images without content. It’s a dispute that also found its way into courts in the US, and could have potentially massive ramifications.
Guthrie continued: Accordingly, is it possible that this consultation will in some respects strengthen UK copyright law for rights owners and allow them to object to AI models being commercialised in the UK even when they have been legally training in the US or elsewhere?”
RELATED STORIES
OpenAI officially releases text-to-video model Sora
UK govt unveils £32m funding for AI projects
AI startup gold rush: Inside the high-stakes world of AI investments
