The UK delays its decision on AI and copyright

The government has pushed back its answer on how copyright law should treat AI training, and the delay itself is what creates the legal work.

What happened

• What is the UK actually doing about AI and copyright? Not a great deal — and that is the problem.
• The government launched a consultation on AI and copyright in 2024. It closed in February 2025 and set out four options for how generative AI and copyright law might fit together.
• Option one: do nothing. Leave the existing law to be worked out by the courts.
• Option two: require licensing in all cases. AI developers would need permission to train on protected works.
• Option three: a broad text and data mining exception. Developers could use the material without permission from the copyright holder.
• Option four: a data mining exception with a rights reservation, underpinned by transparency measures. AI companies could train on copyrighted works unless the rightsholder had opted out.
• Option four was the government's preferred position, it proved highly controversial, and ministers have since indicated that the opt-out model is no longer their preference and that licensing and alternative exceptions are back under consideration.
• Further guidance was expected in March. It now looks likely to slip into next year.

How to use this in applications and interviews

• This matters most to IP litigation teams. The delay extends the window in which disputes are fought without legislative clarity, leaving the courts to apply existing copyright law to the question of whether training a model, or its outputs, infringes.
• At a basic level: more IP litigation work for firms, more litigation risk for AI clients. True, but every candidate will say it.
• The stronger version: firms with real IP benches are litigating the cases that will set the precedents. Trainees in those teams work on decisions that shape the area, and the firm builds market knowledge that legislation, when it comes, will be drafted against.
• That in turn is a client-winning position. A firm that argued the leading cases is the obvious firm to advise on the statute that follows them.
• There is a competitiveness angle too. Legal uncertainty makes M&A and investment in UK AI companies harder to price, and gives developers a reason to move training activity offshore or ring-fence their UK operations.
• If the UK does land on licensing, expect a large-scale data licensing market to accelerate — deals letting AI developers use a dataset for a fee, structured differently depending on the stage of the development lifecycle. That is new, recurring work for IP teams rather than one-off litigation.

Key terms

• Text and data mining exception is a carve-out in copyright law allowing protected material to be analysed at scale without a licence; the fight is over how wide it should be and who can opt out.
• Rights reservation (opt-out) puts the burden on the copyright holder to signal that their work may not be used for training, rather than on the developer to seek permission.