UK Copyright and AI Policy Reset: What the IPA’s Response Means for Creators and Agencies

This revised report sets out the key business context behind UK Copyright and AI Policy Reset: What the IPA’s Response Means for Creators and Agencies, preserving the original facts while improving clarity and flow.

The UK government’s March 2026 report on copyright and artificial intelligence marked a significant reset in one of the most contentious technology-policy debates facing the creative industries. Rather than immediately adopting its earlier preferred approach for text and data mining, the government said it would gather more evidence before changing copyright law. The Institute of Practitioners in Advertising welcomed the retreat from the original preference but warned that major questions remain unresolved for agencies, creators and AI developers.

What the government published

On 18 March 2026, the Department for Science, Innovation and Technology, the Department for Culture, Media and Sport and the Intellectual Property Office published a report and economic impact assessment on the use of copyrighted works in AI development.

The documents were required under the Data (Use and Access) Act 2025 and followed the copyright-and-AI consultation that ran from December 2024 to February 2025. The government said it would not introduce copyright reforms until it was confident they would support both the UK’s creative industries and AI sector.

Why the original proposal caused backlash

The consultation had examined a model under which text and data mining for AI could be permitted more broadly while rights holders retained a mechanism to opt out. Supporters argued that such a system could give AI developers clearer access to training data while allowing creators to reserve their rights.

Many publishers, artists, musicians and other rights holders objected that an opt-out model would place too much burden on creators. They argued that permission should not be presumed and that technical systems for reserving rights were not yet reliable or universal.

What changed in March 2026

The March report did not select the original preferred option for immediate implementation. Instead, the government said further evidence and policy development were needed. It considered the four consultation options and alternative approaches without announcing a final legislative model.

That is why the episode is best described as a policy reset rather than a final settlement. The government moved away from a clear preference but did not decide that the status quo would remain permanently unchanged.

How the IPA responded

The Institute of Practitioners in Advertising said on 20 March that it was disappointed the process still required more time but agreed with shelving the original preferred approach. Richard Lindsay, the IPA’s director of legal and public affairs, said the initial proposal had leaned toward AI developers and had predictably triggered a creative-sector backlash.

The IPA also highlighted issues with direct relevance to advertising, including labelling of AI-generated content, digital replicas of performers and the possibility of a new personality right. Those questions extend beyond training data into how AI-generated material is used in commercial campaigns.

Computer-generated works are another issue

The government’s report considered the UK’s unusual copyright protection for wholly computer-generated works, a provision dating from the Copyright, Designs and Patents Act 1988. The IPA noted that this protection could be reconsidered because it was created long before modern generative AI.

Removing or changing that protection would raise difficult questions about who, if anyone, owns output generated without a human author in the traditional sense. Agencies using AI tools would need clarity because ownership affects licensing, client contracts and the ability to enforce rights against copying.

Training data and output are separate legal questions

Public debate often treats “AI copyright” as one issue, but there are at least two. The first concerns whether protected works can be copied or analysed during model training. The second concerns what happens when a model produces output that resembles or reproduces protected material.

A policy that grants broader access to training data does not automatically determine whether a particular output infringes copyright. Likewise, stronger output safeguards do not settle whether developers had permission to use works in training. Businesses need to examine both stages.

Why transparency matters

Rights holders have repeatedly asked for more information about which works are used to train AI systems. Without meaningful transparency, a creator may struggle to know whether a work was copied, whether an opt-out was respected or whether a licence should have been sought.

Developers respond that detailed disclosure can be technically difficult and may reveal proprietary information. The policy challenge is to create enough transparency for rights to be enforceable without requiring impossible or commercially destructive reporting.

What agencies should do now

Advertising agencies should not interpret the policy reset as a legal free-for-all. Existing copyright, contract, confidentiality, data-protection and advertising rules continue to apply. Agencies need internal policies governing which AI tools staff may use and what material can be uploaded.

Client contracts should also address AI use, ownership of outputs, warranties and the handling of confidential information. Where a campaign relies on a performer’s image or voice, agencies should obtain appropriate permissions rather than assuming a synthetic replica is lawful merely because a tool can generate it.

What creators should understand

Creators should keep records of original works, licensing terms and publication dates. Where platforms or tools provide rights-reservation mechanisms, they may be worth using, but no single technical opt-out should be treated as a complete substitute for legal rights.

Collective licensing may become more important if policymakers seek a system that allows AI developers to obtain large-scale permissions without negotiating separately with millions of creators. Whether such schemes are workable depends on pricing, scope, representation and the ability to identify which works are covered.

Why the government is moving cautiously

The UK is trying to balance two economically important sectors. Creative industries want protection and payment for valuable works, while AI companies want predictable rules and access to large datasets. A rule perceived as too favourable to one side could encourage investment to move elsewhere or weaken incentives for human creation.

The impact assessment acknowledged uncertainty in the evidence base. That uncertainty is a reason for caution, but it also prolongs uncertainty for businesses that need to make investment and contracting decisions now.

What remains unresolved

The central questions include whether there should be a new text-and-data-mining exception, how rights reservations would work, how training-data transparency can be implemented, whether licensing can scale, what happens to computer-generated works and whether additional rights are needed for digital replicas.

Enforcement is equally important. A right that exists on paper but cannot be monitored or enforced across borders may have limited practical value. The international nature of AI development means UK rules will also interact with different approaches in the European Union, United States and other markets.

Bottom line

The government’s March 2026 report did not deliver a final copyright-and-AI regime. It stepped back from the earlier preferred opt-out model and committed to further evidence-gathering before reforming the law. The IPA welcomed that change in direction while stressing that agencies still face important questions around training, labelling, replicas and ownership.

For businesses, the sensible response is not to wait for a perfect future rulebook. Existing law still matters, and organisations should document AI use, respect licences, protect confidential material and write contracts that allocate risk clearly while the policy debate continues.

What the four consultation options were trying to solve

The government’s earlier consultation explored approaches ranging from maintaining the existing legal framework to broader text-and-data-mining permissions with rights reservation. Each option attempted to solve the same tension: AI developers want legal certainty over training data, while rights holders want control, attribution and remuneration.

No option is cost-free. A strict permission-first system can make training expensive and administratively difficult, while a broad exception can transfer value away from creators unless opt-outs and licensing mechanisms are effective. The March report recognised that these trade-offs needed more evidence.

Why small agencies and creators face particular pressure

Large technology companies and media groups can hire specialist legal teams and negotiate bespoke licences. Freelancers, small publishers and independent agencies often cannot. Any new regime that relies on complicated technical notices or individual enforcement may therefore affect small rights holders disproportionately.

The same is true on the user side. A small agency buying an AI service may not know exactly what data trained the model or what warranties are reliable. Clear standard terms and practical guidance could be as important to small businesses as the headline copyright rule.

International rules will influence the UK

AI models are developed and deployed across borders, so UK policy does not operate in isolation. The European Union has its own copyright and AI framework, while U.S. courts continue to address fair-use questions and other jurisdictions are developing different licensing approaches.

A UK rule that is radically different from major markets could increase compliance costs, but simply copying another jurisdiction may not fit the structure of UK copyright law. Policymakers are therefore trying to preserve domestic creative-sector strengths while remaining attractive to AI investment.

What businesses should document today

Companies using generative AI should keep a record of which tools are approved, what data employees may upload, how outputs are reviewed and what licences or contractual promises apply. That documentation can help if a client later asks how a campaign was produced or whether copyrighted source material was used.

It also makes future legal change easier to implement. A business that already knows where AI is used can adapt to new disclosure, licensing or labelling rules far more quickly than one that has allowed uncontrolled use across dozens of teams.