AI Training and Copyright – the debate continues

Readers of our blogs will be well aware of the trials and tribulations of what ultimately became the Data (Use and Access) Act 2025 and which passed into law in June of last year.

The origins of the Act date from the time of Brexit and it was originally intended as a piece of legislation to overhaul data protection law in the post-EU world. Certainly, the Act does reform may areas of data protection law (although not to the extent first anticipated) but one of the most controversial areas of the Act proved to be the use and protection of intellectual property rights in the age of AI.

The debate was fiercely contested and the Data (Use and Access) Bill was batted backwards and forwards between the Houses of Commons and Lords before a wording for the Act was finally agreed. In short, the Bill initially proposed that certain exemptions from the law of copyright (which largely prevents unauthorised use of copyright materials) be introduced for the purposes of training AI models. The aim of those provisions was to accelerate the growth and uptake of AI in the UK.

The proposals, however, raised significant concerns for copyright holders whose online work could be used without their consent and without any remuneration.

The compromise which was finally reached when the Act was passed, was essentially to kick the problem down the road – the draft provisions were removed from the Act and the issue would be re-visited another day. The Government would, however, have to issue a progress statement in December 2025 detailing the work undertaken on the issue in advance of publishing an economic impact assessment and report on the use of copyright works in AI development in March 2026.

Despite the extensive media coverage of the final days of progress of the Bill, the progress report arrived in December to almost no fanfare at all, with the report largely rehearsing much of what has already been said of the difficulties in balancing the competing interests of AI and copyright protection.

As matters stand the options proposed by the Government remain the same:

1. do nothing and continue with current law (an approach which is becoming increasing unfeasible);
2. strengthen copyright law, requiring AI companies to obtain a licence to use copyright material;
3. introduce a broad exception to the current copyright requirements for data mining purposes;
4. provide a data mining exemption but allow rights holders to opt out of use of their copyright material.

At present, option (d) is the preferred option of the Government and while it may be seen as a middle ground, it still raises issues as to how rights holders can actually exercise their opt out rights and enforce them.

The Government is due to publish its impact assessment and report before 18 March 2026 but it is unlikely that the answer to how these competing rights are balanced will get any easier between now and then.