United Kingdom

From Opt‑Out Ambition to Industry Pushback

In December 2024, the government launched a consultation on how copyright law should apply to AI model training. Its preferred option at that time was a model intended to:

• Support AI innovation by allowing training on copyright works to which AI developers had lawful access, while
• Giving rightsholders “greater practical control” through the ability to opt out of such use.
• The most contentious element was this opt‑out structure. A broad coalition of creators, publishers, media organizations and other rightsholders criticized the proposal as unworkable in practice, particularly given:
• Limited transparency over which works were being used in AI training; and
• Uncertainty over how rightsholders could realistically exercise an opt‑out at scale.

These concerns are reflected explicitly in the March 2026 statement, where the government notes that the opt‑out proposal was “overwhelmingly rejected by the vast majority of the creative industries.”

The March 2026 Shift: No Preferred Option, No Immediate Legislative Fix

In its written statement to Parliament on 18 March 2026, issued alongside a Report on Copyright and Artificial Intelligence and an Impact Assessment under the Data (Use and Access) Act 2025, the government confirmed that:

• It has reconsidered its position and
• It “no longer has a preferred option” for copyright‑AI reform.

Instead of advancing the earlier opt‑out model, the government has opted to:

• Continue evaluating policy options from the 2024 consultation and alternative approaches; and
• Monitor domestic and international developments and market practice.

While this does not close the door on future legislative reform, it does represent a meaningful recalibration. The direction of travel has changed, but there is no replacement regime in place.

What the Government Is Prioritizing Now

Although the opt‑out framework has been set aside, the government has not abandoned the field. It has outlined a broader work programme focused on incremental, targeted measures.

1. Ongoing Evidence Gathering and Policy Development

The government will continue to:

• Assess how existing copyright law applies to AI development; and
• Examine both the options previously consulted on and new or hybrid approaches, taking account of evolving market practices.
• For businesses, this means the policy environment remains fluid, with the possibility of further consultation and refinement rather than a single “big bang” reform.

2. Greater Transparency Around AI Training and Outputs

• Transparency has emerged as a core theme of the March 2026 report. The government is placing increased emphasis on:
• Transparency regarding access to and use of copyright works by AI developers (“input‑side” transparency); and
• Transparency concerning AI outputs, including how they are generated (“output‑side” transparency).

Future reforms may therefore focus less on broad exceptions and more on mechanisms that, improve rightsholders’ visibility into how their works are used; and strengthen their ability to understand and enforce their rights.

3. Taskforce on AI‑Generated Content Labelling

The government has announced a task force on AI-generated content labeling, with an interim report expected in the autumn.

This initiative is intended to explore how AI‑generated content should be identified and labelled, which may have implications for:

• Platforms hosting AI‑generated material;
• Businesses deploying generative AI in consumer‑facing products; and
• Rightsholders concerned about confusion between human‑created and AI‑generated works.

4. Consultation on Digital Replicas

The government will also launch a summer consultation on digital replicas, focusing on harms associated with unauthorized replication of a person’s likeness or identity.

Issues likely to be in scope include:

• Misuse of an individual’s image, voice or persona in AI‑generated content; and
• The extent to which existing rights and remedies adequately address such uses.
• Businesses involved in synthetic media, virtual influencers, or personalized content should expect closer scrutiny in this area.

5. Monitoring – Not Regulating – the Licensing Market (For Now)

Finally, the government appears inclined, at least in the near term, to:

Observe rather than intervene in emerging licensing models for AI training on copyright works.

The March 2026 report discusses the development of licensing markets, but stops short of imposing a specific legislative solution. This suggests a preference for Market‑led licensing arrangements, while Retaining the option of more targeted statutory intervention if needed.

Practical Implications for AI Developers

For AI developers and businesses deploying generative AI in or from the UK, the key implications are:

• No new statutory “safe harbour”: There is still no broad, legislated permission to train on copyright works simply because they are lawfully accessible.
• Continued copyright risk assessment: Companies should continue to evaluate copyright exposure in line with existing law and should not assume that a permissive text‑and‑data mining‑style regime will be introduced in the short term.
• Importance of transparency and documentation: Given the government’s focus, maintaining robust records of training data sources and usage practices may become increasingly important in both regulatory and commercial negotiations.
• Licensing remains central: In the absence of a statutory solution, contractual licensing and related technical measures are likely to remain the primary tools for clearing rights and managing risk.

These implications flow from the government’s decision not to proceed with its previously preferred model and the absence of any enacted legislative reform in the March 2026 materials.

Practical Implications for Rightsholders and Creative Industries

• For rightsholders, the government’s announcement is both a political signal and a commercial reality check:
• Policy win, but no comprehensive new regime: The government is not currently prepared to proceed with a default AI‑training‑permitted framework subject only to opt‑out. However, rightsholders do not yet have a replacement system that fully addresses transparency, licensing leverage or enforcement.
• Negotiating landscape remains unsettled: In the near term, rightsholders will continue to rely on existing copyright law, contractual terms, and emerging technical measures to manage and monetize AI training uses of their works.
• Focus on transparency and labelling: Upcoming work on transparency and AI‑content labelling may offer new tools for distinguishing and tracking uses of creative works and AI‑generated outputs, albeit on a gradual basis.
• Opportunity to shape future reforms: Continued consultations – including on digital replicas – provide a further opportunity for creators and industry bodies to influence the ultimate design of any licensing‑led, statutory, or hybrid framework.
• These conclusions follow from the government’s withdrawal of its preferred option and its decision to carry on policy development rather than legislate immediately.

Comparison: March 2026 U.S. Federal Policy Position

A useful point of comparison is the U.S. Administration’s March 20, 2026 National Policy Framework for Artificial Intelligence: Legislative Recommendations, which takes a notably different approach to the copyright-and-AI training issue from the recent UK position. The White House framework is a policy blueprint for Congress, not binding law, and any implementation would require congressional action.

Key U.S. IP-Related Takeaways

• AI training and copyright: the Administration states that it believes training AI models on copyrighted material does not violate copyright law, while acknowledging that contrary arguments exist. It therefore recommends that courts should resolve the issue, and that Congress should not legislate in a way that affects judicial resolution of whether such training constitutes fair use.
• Licensing frameworks: the Administration suggests that Congress should consider enabling licensing frameworks or collective rights systems through which rightsholders could collectively negotiate compensation from AI providers without incurring antitrust liability. At the same time, it recommends that such legislation should not determine when or whether licensing is legally required.
• Digital replicas: the framework recommends that Congress consider a federal regime protecting individuals against the unauthorized distribution or commercial use of AI-generated digital replicas of their voice, likeness, or other identifiable attributes, while preserving exceptions for parody, satire, news reporting, and other First Amendment-protected expression.
• Ongoing monitoring: the Administration further recommends that Congress continue to monitor how copyright case law and enforcement develop in the courts and assess whether additional action may eventually be needed to address AI-specific gaps or strengthen protections for creators.

Summary:

The UK and U.S. are currently signaling different policy instincts on AI and copyright. The UK has retreated from its earlier preferred opt-out model for AI training on copyrighted works and is shifting toward further consultation and policy development centered on transparency, creator control, labelling, and digital replicas. The March 2026 U.S. federal policy framework, by contrast, states that AI training on copyrighted material does not, in the Administration’s view, violate copyright law, but recommends that courts should resolve that question, while Congress considers collective licensing mechanisms and digital-replica protections. The result is a notable divergence: the UK is reassessing legislative reform, while the U.S. framework is more focused on judicial resolution of the core copyright issue and targeted legislative action around the edges.