Last reviewed: April 10, 2026

Jurisdictions covered: UK (primary), EU (comparison)

Reading time: 16 minutes

UK Copyright and AI: The Legislative Limbo That Won’t Resolve Itself

On March 18, 2026, the UK government published its long-awaited report on copyright and artificial intelligence. The report’s central finding: “permission would usually be needed to copy protected works at different stages of AI training.” The government then proposed no mechanism for obtaining that permission at scale.

The broad text and data mining (TDM) exception that the government had backed since 2022 — allowing AI companies to train on copyrighted content with an opt-out for rights holders — was abandoned. Approximately 88% of the 11,000+ consultation respondents had opposed it. Only around 3% supported the government’s preferred opt-out model. The creative industries won the argument.

But the alternative is not a licensing framework, a new exception, or any legislation at all. The alternative is nothing. The government’s report describes several possible future paths — a narrower exception for SMEs, a “focused” exception for non-commercial research, a voluntary Creative Content Exchange — without endorsing any of them.

The result: UK copyright law and AI training exist in legislative limbo. The only clear TDM exception is s.29A of the CDPA 1988, which covers non-commercial research only. Everything else requires a licence that the market has not yet produced at scale. Meanwhile, the EU has a functioning framework — and UK companies with EU customers must comply with it regardless of what happens domestically.

Key Takeaways

• The UK government abandoned its preferred TDM exception in March 2026 after overwhelming opposition (88% of 11,000+ respondents favoured mandatory licensing over the opt-out model).
• No replacement legislation has been proposed. The government described several possible paths — narrower exceptions, voluntary licensing — without committing to any.
• The only current TDM exception (s.29A CDPA) covers non-commercial research only. Commercial AI training on copyrighted UK content without a licence is legally uncertain.
• The Getty v Stability AI case is not resolved. The November 2025 judgment found that AI models do not store infringing copies — but permission to appeal was granted in December 2025. The appeal is pending at the Court of Appeal.
• UK companies with EU market exposure face the EU’s copyright framework regardless — Art. 53(1)(c) of the AI Act requires GPAI providers to maintain a copyright compliance policy and respect TDM opt-outs.

What Does UK Copyright Law Currently Say About AI Training?

UK copyright law was not written with AI in mind. The Copyright, Designs and Patents Act 1988 (CDPA) predates the internet, let alone large language models. Two provisions are relevant:

Section 29A — Text and Data Mining for Non-Commercial Research. This is the only statutory TDM exception in UK law. It permits copying of copyrighted works “for the sole purpose of computational analysis” if the researcher has lawful access to the work and the copying is for non-commercial research. This exception does not cover commercial AI training. A company training a model for a commercial product cannot rely on s.29A.

Section 9(3) — Computer-Generated Works. This provision grants copyright to “the person by whom the arrangements necessary for the creation of the work are undertaken” when a work is generated by a computer without a human author. The government’s March 2026 report proposes removing this provision — arguing that AI outputs should not automatically receive copyright protection. This is a proposal, not enacted law.

Beyond these provisions, the general rules apply: reproducing copyrighted works requires permission (a licence) unless a specific exception applies. Since the only applicable TDM exception is limited to non-commercial research, commercial AI training on UK-sourced copyrighted content exists in a legal grey zone — not clearly prohibited, not clearly permitted, and untested at the Court of Appeal level.

What Happened with the TDM Exception?

The timeline of policy reversals tells the story:

June 2022: The Intellectual Property Office (IPO) announced plans for a broad TDM exception covering commercial AI training, with an opt-out mechanism for rights holders. The AI industry welcomed it. The creative industries opposed it.

December 2024 – February 2025: The IPO ran a public consultation. Over 11,000 responses were received — the largest IP consultation in UK history. Approximately 88% of respondents favoured mandatory licensing (the creative industries’ preferred position). Only about 3% supported the government’s opt-out model.

March 6, 2026: The House of Lords Communications and Digital Committee published its report (HL Paper 267), recommending a licensing-first approach and rejecting the broad commercial TDM exception.

March 18, 2026: The government published its Section 136 DUA Act report on copyright and AI (CP1205). The key sentence: “a broad copyright exception with opt-out is no longer the government’s preferred way forward.” The government acknowledged that “permission would usually be needed to copy protected works at different stages of AI training” — but proposed no mechanism for delivering that permission.

What the report proposed instead:

• A “focused” exception for SMEs, non-commercial research, or task-specific AI — discussed but not endorsed
• A voluntary Creative Content Exchange (CCE) — targeted for summer 2026, no confirmed launch date
• Further evidence gathering and monitoring of international developments
• Removal of s.9(3) copyright protection for computer-generated works — proposed but not legislated

The Design and Artists Copyright Society (DACS) described the outcome as “a reset, not a resolution.” Lewis Silkin called it “kicking the issue down the road.” The BBC reported the government had “backtracked” on its AI-friendly copyright plans.

What Did Getty v Stability AI Decide?

Getty Images v Stability AI is the most significant UK court case on AI training and copyright. Getty sued Stability AI for using millions of Getty-licensed images to train Stable Diffusion without permission.

The November 2025 judgment (High Court) addressed two key questions:

Can an AI model be an “infringing copy”? No. The court found that the trained model (Stable Diffusion) does not store copies of individual training images. The model’s weights encode statistical patterns, not reproductions. This meant Getty’s primary infringement claim — that the model itself was an infringing copy — failed.

Did secondary infringement occur through training? The court addressed whether the act of copying images during the training process (which undeniably involves reproduction) was infringing. Some secondary infringement claims failed on jurisdictional grounds — the training occurred outside the UK. The full reasoning on whether UK copyright was infringed by the training process itself was not definitively resolved.

Permission to appeal was granted in December 2025. The appeal is pending at the Court of Appeal. The outcome could significantly clarify whether AI training on copyrighted content requires a licence under UK law. Until the appeal is decided, the legal position remains uncertain.

What practitioners should take from this: The High Court’s finding that AI models are not “copies” is significant but not final. The question of whether the training process itself requires permission remains open. Do not treat the November 2025 judgment as clearance to train on copyrighted UK content without a licence.

What About the EU? A Framework Already Exists

While the UK deliberates, the EU has a functioning copyright framework for AI training:

Directive 2019/790, Article 4: Establishes a TDM right for commercial purposes, with an opt-out mechanism for rights holders. Rights holders can “reserve” their rights through machine-readable means (robots.txt or equivalent). If no opt-out is expressed, TDM is permitted.

EU AI Act, Article 53(1)(c): Requires all GPAI providers — including open-source models — to implement a copyright compliance policy and respect TDM opt-outs. This is a binding obligation enforceable by the AI Office from August 2, 2026.

Kneschke v LAION (Hamburg Higher Regional Court, December 10, 2025): The first appellate court test of TDM and AI training. The court held that pre-processing for AI training qualifies as TDM under the Directive. Machine-readable opt-outs must be respected. Human-readable disclaimers on websites are insufficient — only machine-readable signals like robots.txt qualify.

Dimension	UK	EU
Commercial TDM exception	None (s.29A is non-commercial only)	Art. 4 Directive 2019/790 — commercial TDM with opt-out
Opt-out mechanism	Not applicable (no exception to opt out of)	Machine-readable (robots.txt); human-readable insufficient per Kneschke
AI Act obligation	None	Art. 53(1)(c) — copyright compliance policy + training data summary mandatory
Court precedent	Getty v Stability AI (appeal pending)	Kneschke v LAION (TDM for AI training is lawful with valid opt-out)
Government position	“Permission usually needed” but no mechanism proposed	Framework in place, enforced from August 2026

For UK companies with EU market exposure: The EU framework applies to you regardless of UK domestic law. If you provide a GPAI model that reaches EU users, you must comply with Art. 53(1)(c) — copyright policy, TDM opt-out respect, training data summary. The UK’s domestic limbo does not exempt you from EU obligations.

What Licensing Options Are Emerging?

Two parallel licensing developments are worth monitoring:

Creative Content Exchange (CCE). The government’s proposed voluntary marketplace where AI companies can negotiate licences with rights holders. Targeted for summer 2026. No confirmed launch date, no published specifications. The concept: a centralized platform that reduces transaction costs for individual licensing negotiations. The risk: voluntary participation means AI companies may simply not engage, and rights holders may set prices that make licensing commercially impractical.

Publishers Licensing Services (PLS) Collective Scheme. PLS launched its own collective AI licensing scheme in March 2026 — ahead of the CCE. This is a practical market response: publishers offer blanket licences for AI training on their content. Whether major AI companies will subscribe is unclear, but the scheme demonstrates that the market is not waiting for government action.

The structural problem: Even if licensing frameworks emerge, they cannot address content already used for training. Models trained before any licensing scheme existed — which is most current models — used copyrighted content under the previous legal ambiguity. No retroactive licensing mechanism has been proposed.

What Should Practitioners Do?

1. Do not assume AI training on UK copyrighted content is legal. The government’s own report says “permission would usually be needed.” The only clear exception (s.29A) covers non-commercial research only. Until the Getty appeal is decided or legislation is enacted, commercial AI training on UK-sourced content without a licence carries legal risk.

2. Document your training data provenance. Whether or not UK law eventually requires it, the EU AI Act does — Art. 53(1)(d) mandates a training data summary. Building this documentation now serves both jurisdictions.

3. Implement machine-readable opt-out respect. The EU’s Kneschke ruling confirmed that robots.txt and equivalent signals are the standard. Even if UK law does not yet require it, implementing opt-out respect positions you for compliance in both markets and reduces litigation risk.

4. Monitor the CCE and PLS schemes. If a viable collective licensing mechanism emerges, early participation demonstrates good faith and may provide a defence against infringement claims.

5. Watch the Getty appeal. The Court of Appeal’s decision will be the most significant UK copyright/AI ruling to date. If it upholds the finding that models are not “copies,” the focus shifts entirely to whether the training process requires permission. If it reverses, the implications for every model trained on UK-sourced content are substantial.

6. If you serve EU customers, comply with Art. 53(1)(c) now. This is not optional. The EU AI Office gains enforcement powers on August 2, 2026. Copyright compliance policies and training data summaries should be in place before that date.

Disclaimer: This content is for informational purposes only and does not constitute legal advice. UK copyright law as applied to AI training is unsettled and actively evolving through litigation and potential legislation. Organizations should consult qualified IP counsel. Reg Intel is not a law firm and does not provide legal services.

Last verified: April 10, 2026

Sources

Official Sources

• UK Government Report on Copyright and AI (CP1205), March 18, 2026
• Copyright, Designs and Patents Act 1988, s.29A
• House of Lords Communications and Digital Committee, “AI, copyright and the creative industries” (HL Paper 267), March 6, 2026
• Data (Use and Access) Act 2025, s.136

Case Law

• Getty Images v Stability AI [2025] EWHC (Ch) — November 2025, High Court. Permission to appeal granted December 2025. Appeal pending.
• Kneschke v LAION, OLG Hamburg, Case 5 U 104/24 — December 10, 2025. TDM for AI training qualifies under DSM Directive; machine-readable opt-out required.

Analysis and Commentary

• Reed Smith: “The opt-out is dead — but what comes next?”, March 26, 2026
• International Publishers Association: Copyright Update, March 31, 2026
• DACS: “A reset, not a resolution” — response to government report, March 2026
• Lewis Silkin: “Kicking the issue down the road” — analysis, March 2026
• PLS: Collective AI licensing scheme launch, March 2026

Compare: EU vs UK

For the comprehensive comparison across twelve dimensions — structural divergence, risk classification, the 19 UK regulators vs the EU AI Office, enforcement penalties, the Data (Use and Access) Act 2025, AISI vs the EU AI Office, and a five-step dual-market compliance baseline — see EU vs UK AI Regulation: Precaution vs Innovation Compared (2026).