The UK government's decision to delay its AI copyright reforms represents a significant victory for the creative industries, but it is a victory that comes with important caveats. The proposed reforms — which would have introduced a broad text and data mining exception allowing AI companies to train on copyrighted material without obtaining licences — have been opposed by an unusually united coalition of publishers, authors, musicians, visual artists, and film producers. The delay signals that the government has concluded that proceeding with the original proposals would be politically and reputationally costly. It does not, however, mean that the underlying policy question has been resolved.

The core dispute is about who should bear the cost of AI development. AI companies argue that training on publicly available data — including copyrighted works — is analogous to a human researcher reading widely before writing, and that requiring licences for training data would make AI development economically impractical in the UK. Rights holders argue that their works have commercial value, that AI companies are extracting that value without compensation, and that a broad training data exception would effectively transfer wealth from creators to technology companies without consent or payment.

The UK's position is particularly significant because it is one of the few major jurisdictions that has explicitly considered introducing a broad AI training exception. The European Union's AI Act took a different approach, requiring AI companies to disclose the copyrighted works used in training and to respect opt-out mechanisms. The United States has left the question largely to the courts, with multiple cases working their way through the legal system. The UK had positioned itself as potentially the most permissive major market for AI training data — a positioning that the creative industries have successfully challenged, at least for now.

The delay creates uncertainty for both sides. AI companies operating in the UK face continued ambiguity about the legal status of their training practices. Rights holders have won a reprieve but not a permanent settlement: the government has indicated that it intends to return to the question with revised proposals, and the direction of travel remains unclear. The most likely outcome is some form of mandatory licensing framework that requires AI companies to pay for training data but does not require individual opt-in consent — a compromise that would satisfy neither side fully but might be politically sustainable.

For the publishing industry specifically, the delay is an opportunity to develop collective licensing infrastructure that could make a licensing framework workable in practice. The UK Publishers Licensing Services has already begun work on collective AI licensing mechanisms, as reported elsewhere this week. The window created by the government's delay should be used to build the institutional capacity to administer licences at scale — because when the government does return to this question, having a workable licensing system ready to deploy will be the publishing industry's strongest argument against a broad exception.