UKIPO Reopens the AI Copyright Fault Line on TDM and CGWs
Debate over AI and copyright in the UK has flared up again in late June 2026, but the real significance lies in two policy signals that are harder to ignore. In its March 2026 report on Copyright and Artificial Intelligence, the government stepped back from its earlier preference for a commercial text and data mining exception built around rights reservation and transparency. At the same time, it signalled that the UK’s special copyright protection for wholly computer-generated works may no longer deserve to survive.
This is not final legislation. It is, however, a serious change in policy direction. For content businesses, platforms, model developers and companies shipping AI products into the UK, the practical questions are no longer limited to who owns an output. The harder issues are becoming lawful access to training material, whether opt-out systems can work at scale, how much transparency may be expected around crawling and training, and where the legal line will be drawn between AI-assisted works and outputs with no human author at all.
This is not a final statute, but it is a real policy reset
The December 2024 consultation put four broad approaches on the table. The most politically sensitive one was a new commercial text and data mining exception combined with machine-readable rights reservation and transparency measures. By the time the government issued its December 2025 progress statement, support for a licensing-first model had overwhelmed support for the original opt-out approach. The March 2026 report went further still: a broad opt-out exception is no longer the government’s preferred path, and ministers now want more evidence, closer review of alternatives, and greater emphasis on transparency and market-based licensing.
That shift matters because it changes the operating assumption for businesses. The UK has not moved to a simple “everything must be licensed” statute, but neither is it moving quickly toward a permissive AI-training carve-out. For the near term, companies should expect a mixed environment: existing copyright law remains in place, licensing markets continue to develop, transparency pressure is rising, and future reform remains possible but unresolved. Betting on a sweeping exemption now would be a poor reading of the policy mood.
The TDM fight is now about burden allocation, not slogans
The real pressure point in this consultation was not abstract principle. It was who would carry the cost of making any new system work. Supporters of a broader TDM exception argued that AI developers should be able to mine lawfully accessed material, including for model training, unless right holders reserved their rights through machine-readable signals. Opponents answered that this does not reduce friction so much as relocate it. If creators, publishers, image libraries and smaller right holders must build, monitor and enforce opt-out signals across multiple services and formats, then the legal burden has simply been shifted onto the side with less technical leverage.
That is why the live questions in the UK are now more operational than rhetorical. What counts as lawful access in practice? How standardised must rights-reservation signals be before they become meaningful rather than symbolic? What happens when a model is trained in a more permissive jurisdiction and then deployed into the UK market? Until those questions have workable answers, a broad TDM exception will remain more controversial in practice than it may look on paper.
Section 9(3) is losing policy cover
The second major signal concerns section 9(3) of the Copyright, Designs and Patents Act 1988, the provision that attributes authorship of certain computer-generated works to the person making the necessary arrangements. The government’s report does not treat this as a settled or healthy rule. On the contrary, it highlights a structural tension between section 9(3) and the modern originality standard, which is increasingly tied to human intellectual creation and personal creative choices. In other words, the statutory text remains on the books, but its conceptual footing looks weaker than before.
The policy implication is clear enough. The UK is moving closer to protecting works created by humans using AI tools, while becoming less willing to maintain a separate copyright shelter for purely machine-generated text and images. That distinction matters. It affects disclosure practices, platform terms, output-labelling strategies and the economics of businesses built around large volumes of low-cost generated content. If wholly AI-generated outputs become harder to protect, some content models will need redesign rather than minor compliance edits.
What platforms, AI developers and UK filers should do now
This consultation does not itself rewrite UK patent law, nor does it announce a fresh patentability rule for AI inventions. But it should still be read alongside the wider UKIPO picture. In February 2026, following the Supreme Court’s Emotional Perception AI decision, the UKIPO withdrew its earlier examination guidelines for AI inventions. That is a reminder that UK rules affecting AI are still moving, and that businesses should not treat copyright, patent practice and data governance as isolated silos.
The immediate task is not messaging. It is evidence discipline. Companies should be able to show how training data was obtained, which sources were licensed, what signals restricted crawling or reuse, how output provenance is handled, and where meaningful human technical or creative contribution sits in the workflow. Platforms face the same test. The next competitive divide in the UK will not simply be who has the biggest model or the largest library, but who can explain their training inputs, permissions stack, output controls and complaints process without hand-waving. That is where this debate is heading.
