If the General Data Protection Regulation (GDPR) was an earthquake, the Data (Use and Access) Act 2025 (DUAA) is more like a series of aftershocks in the sense that DUAA mostly tweaks the existing law rather than making fundamental changes. Nonetheless, it's important to understand how DUAA has altered the status quo, not only to remain compliant with privacy and data protection laws, but also to take to take advantage of a legal framework that has been clarified and relaxed in various ways.
To help you get to grips with DUAA, our IT and data lawyers are publishing a series of articles to explain how it will impact your organisation. This article from Robert Beveridge focuses on the parts of DUAA relating to AI and copyright.
Background
On the face of it, it's strange that a law calling itself the Data (Use and Access) Act is concerning itself with copyright at all. It's only when you think of copyright works as "data" rather than books, songs, films, etc, that it begins to make sense.
In the early days of GenAI, there was a scramble for data to train Large Language Models, such as ChatGPT. It seemed like the leading GenAI companies were racing to train their models on the collective creative works of the human race.
This arms race has rapidly generated some nifty technology that many of us now use every day, but it also created policy problems, which politicians in the UK and elsewhere have not yet decided how to solve. Notably:
- Should GenAI companies be required to disclose the data used to train their models?
- Under what conditions should GenAI companies be permitted to train their models on copyright works?
The UK government kicked off the discussion on how to solve these problems with its AI and copyright consultation earlier this year. Four policy options were considered (paraphrased slightly for clarity):
- Option 0: Do nothing. Copyright and related laws remain as they are
- Option 1: Strengthen copyright, prohibiting the use of copyright works to train AI unless the rights holder has granted a licence.
- Option 2: Create a new rule that allows copyright works to be used to train AI with or without the rights holder’s permission (in other words, weaken copyright).
- Option 3: Create a new regime which (i) requires AI models to be transparent about the material used to train their models – making it easier for rights holders to enforce their copyright, and (ii) permits all copyright works to be used to train AI unless the rights holder marks them with a machine readable notice reserving their rights.
Copyright provisions made their way into DUAA after a fierce campaign in the House of Lords against the government’s favoured Option 3. DUAA requires the government to do the following by March 2026:
- Publish an economic impact assessment of the four policy options outlined above.
- Publish, and lay before Parliament, a report on the use of copyright works in AI development which addresses:
- Technical measures that may be used to control how copyright works are accessed and used to train AI.
- The effect of copyright on the ability of AI developers to access and use copyright works for training purposes.
- The disclosure of information by developers about their use of copyright works.
- The granting of licences to AI developers to do acts restricted by copyright.
- How to enforce restrictions on the use of copyright works in AI development.
- Publish a progress report on the economic impact assessment and AI/copyright report by December 2025.
The economic impact assessment and AI/copyright report should help the government make a more informed policy decision – and decide whether new legislation is needed. They've been clear that they do not intend to rush this process. This means that for the foreseeable future, rights holders and AI developers will be left to muddle through based on their interpretations of current copyright laws. That's not necessarily a bad thing. There are complex and urgent policy problems at stake here – but there does not yet seem to be a fair and workable legislative solution ready to hand.
Comment
AI developers and rights holders who are following the UK policy debate will be hoping that – in the words of the classic Guiness advert – good things come to those who wait. We’ve been closely following the legal and ethical tensions between AI development and copyright law for some time.
The following articles will give you a clear sense of how this debate has developed over time:
Our content explained
Every piece of content we create is correct on the date it’s published but please don’t rely on it as legal advice. If you’d like to speak to us about your own legal requirements, please contact one of our expert lawyers.