Introduction
UK AI copyright law 2026 looks very different from what Downing Street first proposed, yet AI firms still hold the leverage. The government’s December 2024 consultation sketched a text and data mining exception with an opt-out that would have let developers scrape copyrighted work by default. More than 11,500 responses arrived in two months, and the March 2026 statutory report under Section 137 of the Data Use and Access Act dropped that preferred option. The day-to-day position for creators barely moved, because the 1988 Copyright, Designs and Patents Act still governs commercial training. This article walks through what the Section 137 progress report actually says and why the practical balance of power still favors AI developers. We cover the Getty Images v Stability AI ruling, the Make it Fair newspaper protest, and the three Lords defeats on transparency amendments. Readers get a creator playbook, a reform-option comparator, and three case studies showing how UK rightsholders are acting now while Parliament hesitates on the next step.
Quick Answers on UK AI Copyright Law 2026
What is UK AI copyright law 2026 in one sentence?
UK AI copyright law 2026 keeps the 1988 Copyright, Designs and Patents Act in force while the Data Use and Access Act 2025 requires government reports on transparency, licensing, and remuneration for creators.
Did the UK adopt an opt-out model for AI training?
No. The UK government formally dropped its preferred opt-out copyright model in the March 2026 Section 137 progress report and pledged more licensing and transparency work instead.
Can AI firms legally train on UK copyrighted works today?
Only with permission or under narrow exceptions. Section 29A permits text and data mining for non-commercial research. Commercial training on copyrighted material without a license remains an infringement risk.
Key Takeaways on the 2026 UK Copyright Shift
- The government’s opt-out plus statutory text and data mining exception is dead for now, after 11,500 consultation responses and three Lords defeats.
- Section 137 of the Data Use and Access Act 2025 forces a progress report, an impact assessment, and a wider economic study of AI and copyright.
- Section 29A of the 1988 Act still permits only non-commercial research mining, so commercial scraping without a license remains legally exposed.
- Getty v Stability AI narrowed primary copyright claims but left a trademark finding that keeps litigation risk very much alive in the UK.
Table of contents
- Introduction
- Quick Answers on UK AI Copyright Law 2026
- Key Takeaways on the 2026 UK Copyright Shift
- What Is UK AI Copyright Law 2026
- What the New UK AI Copyright Regime Actually Says
- How the Original Opt-Out Proposal Framed the Debate
- Why the 2024 Consultation Received 11,500 Responses
- The Data Use and Access Act 2025 and Section 137
- Where the Copyright, Designs and Patents Act 1988 Still Governs
- The March 2026 Statutory Report and Its Policy Reset
- Text and Data Mining Rules Under Section 29A
- How Creators Implement Reservation of Rights in the UK
- Transparency Obligations AI Firms Face in the UK Today
- The Economic Weight of the Creative Industries
- Signals From Getty Images v Stability AI
- Make It Fair, McCartney and the Artist Backlash
- How Publishers and Labels Are Rewiring Licensing
- Risks the Current Framework Poses to Creators
- Risks AI Developers Face Operating in the UK
- Ethics of Training Models on Protected British Works
- The Future of UK AI Copyright Law Beyond 2026
- Key Insights on the 2026 UK AI Copyright Debate
- How Each Reform Option Compares on Creator Protection
- Real-World Examples of UK AI Copyright in Practice
- Case Studies in UK AI Copyright Disputes and Deals
- Common Questions About UK AI Copyright Law 2026
What Is UK AI Copyright Law 2026
UK AI copyright law 2026 preserves the 1988 Copyright, Designs and Patents Act while the Data Use and Access Act 2025 requires statutory progress reports, transparency work, and licensing studies. Developers still need permission to train on protected works, and creators retain the full suite of exclusive rights by default.
An Interactive From AIplusInfo
Compare Four UK AI Copyright Reform Options
Pick a reform option and adjust how much weight to put on creator protection, developer certainty and transparency. The scores reflect the live Section 137 debate.
Licensing-first regime
60 percent
25 percent
15 percent
Composite reform score
72
Licensing-first regime balances high creator income with moderate developer certainty, scoring 72 under the current weights.
Likely annual UK licensing revenue by 2030
GBP 6.5 billion
MLex forecast range for a clear licensing framework sits at GBP 5 to 8 billion.
Scoring draws on the March 2026 Section 137 progress report and the MLex analysis of a UK AI licensing market.
What the New UK AI Copyright Regime Actually Says
The UK AI copyright regime is less a single statute and more a layered patchwork of three sources. It layers the 1988 Copyright Act, the Data Use and Access Act 2025, and ongoing Section 137 work. The core legal rule has not moved: training a commercial AI model on protected UK works without a license remains an act of copying that engages the author's exclusive rights. The regime captures the practical point that developers must either secure permission or rely on a narrow research exception. What has shifted is the political temperature, with the Lords, newspapers, and musicians forcing the government to abandon its preferred reform in March 2026. The regime now sits in deliberate limbo while officials gather evidence on licensing, transparency, and remuneration options.
The A&O Shearman briefing on where UK AI copyright stands notes that the government has committed to a Section 137 impact assessment within twelve months of Royal Assent. That timeline points to a decision window stretching into mid-2026, with secondary legislation possible before the end of the year. In practice the Intellectual Property Office is coordinating technical working groups on licensing standards and content provenance. These groups bring together rightsholder bodies, major AI labs, and content infrastructure vendors to road-test workable mechanics. None of the outputs is binding on developers yet, but they signal where enforcement focus is likely to fall.
Readers familiar with the broader regulatory landscape will recognize the pattern of regulators playing catch-up with faster-moving models. The UK differs from the EU in that it has no AI Act equivalent, so copyright carries more of the regulatory load than in Brussels. It also differs from the US, where fair use doctrine shapes training disputes under common-law reasoning rather than statute. That leaves the UK leaning on the 1988 Act, the DUA Act, and the courts to resolve conflicts case by case. The direction of travel is clearly toward tighter transparency and licensing, yet without statutory force the balance of leverage still rests with developers who already hold trained models.
How the Original Opt-Out Proposal Framed the Debate
Shifting focus to the policy design, the December 2024 consultation offered four reform options. Option three, an opt-out plus a broad text and data mining exception, was the government's preferred approach. That option would have let AI firms train on any UK work by default, placing the burden on creators to actively reserve their rights through machine-readable signals. The closed UK consultation on copyright and artificial intelligence framed the design as a pragmatic compromise. Critics argued the opposite: that opt-out shifts the economics onto thousands of individual creators who lack technical teams. The Lewis Silkin breakdown of the consultation catalogued the sharp split between tech-sector support and creative-industry opposition.
Underneath the detail sat a geopolitical gamble, with ministers suggesting that a looser UK regime could attract AI investment that might otherwise flow to Dublin or Paris. The design borrowed from Article 4 of the EU's 2019 DSM Directive, which permits commercial text and data mining unless rightsholders expressly opt out. UK officials downplayed the risk that machine-readable opt-outs would prove hard to enforce on scraped web data. Rightsholder bodies countered with evidence that most AI crawlers ignore robots.txt and similar reservation signals. That technical reality undermined the ministerial argument before the consultation window closed.
Why the 2024 Consultation Received 11,500 Responses
Turning to the record-breaking response volume, the Intellectual Property Office received approximately 11,500 submissions during the two-month consultation window. That consultation volume was unprecedented for UK intellectual property policymaking, exceeding typical responses by more than tenfold. The Mishcon de Reya progress report briefing from the Mishcon de Reya progress report briefing. Three forces drove the surge: a coordinated creator campaign, high-profile artist advocacy, and open online tooling that lowered the barrier for individual submissions. The result was a dataset the government could not quietly file away, which is partly why the March 2026 report walked the policy back. Critics in Westminster framed the volume as proof that the preferred option lacked a social license.
The composition of the responses mattered as much as the headline count. Rightsholder associations submitted detailed economic analyses, with the Publishers Association, UK Music, and the Society of Authors coordinating evidence packs. Individual creators sent personal stories about scraped work and unlicensed derivative outputs, several of which appeared verbatim in press coverage. AI developers and trade bodies also responded in strength, including TechUK and the major foundation-model labs operating in the UK. The story of UK arts icons demanding AI copyright safeguards captured the public-facing side of the campaign.
Coverage from the Walker Morris analysis of the creative industries response underlined how closely the volume tracked media mobilization. National newspapers ran coordinated editorial coverage in late January and February 2025. Specialist bodies like PRS for Music published submission templates that creators could adapt. That infrastructure helped individual musicians, authors, illustrators, and photographers push past the usual advocacy chokepoints. The resulting dataset gave the Lords ammunition for the amendments that would later derail the preferred policy option.
Behind the volume sat a longer-running concern that AI firms were effectively externalizing data-acquisition costs onto the creators whose work powered the models. The Metas AI training using pirated books coverage put a specific number on that externalization, with the Books3 dataset reportedly containing hundreds of thousands of unlicensed titles. UK authors recognised their own books in the dataset and pressed for transparency obligations. That demand later crystallized into the Kidron amendments that passed the Lords. The sheer scale of unlicensed ingestion made the opt-out proposal politically untenable before the government had finished defending it.
The Data Use and Access Act 2025 and Section 137
Building on the consultation outcry, the Data Use and Access Act 2025 received Royal Assent on 19 June 2025 and became the vehicle for Parliament's response. The Act is mostly a data-protection update rebranded from the earlier Data Protection and Digital Information Bill. Its final sections carry the copyright and AI provisions that ministers accepted under pressure. The Brodies briefing on AI developers versus creators notes that Section 137 is the key operative clause. It requires the Secretary of State to publish a progress report within nine months and an impact assessment within twelve months. That timing locked the government into tangible deliverables rather than open-ended consultation.
Section 137 is not a copyright reform in itself, which is a point creators often miss on first reading. It is a mandatory evidence and reporting mechanism designed to tee up future legislation. The report must address licensing models, transparency obligations, and the economic impact on creators and developers alike. The Act also set an eighteen-month deadline for a broader economic study. The UK is now locked into a published evidence trail through to late 2026. That evidence trail is the primary short-term win for creator advocates, because it prevents ministers from quietly shelving the issue.
The PYMNTS coverage of the three successive Lords defeats captures how fraught the Act's passage became. Baroness Beeban Kidron led the amendments, which initially demanded that AI firms disclose the specific copyrighted works used in training. The government stripped those requirements out, prompting the first defeat, then the second, then the third in April 2025. Ministers eventually accepted a watered-down version that mandates reporting rather than disclosure, which is how Section 137 reached the statute book. The affair proved that the government could pass legislation but could not control the terms of the AI copyright debate.
Where the Copyright, Designs and Patents Act 1988 Still Governs
Stepping back from the DUA Act, the 1988 Copyright, Designs and Patents Act continues to govern substantive rights, infringement, and remedies in 2026. Section 16 of the 1988 Act grants authors the exclusive right to copy, communicate, and adapt their work. Training a model constitutes copying for the purposes of that section. Section 17 extends the restriction to copies made at any stage of a technological process, which captures the ingestion and tokenization pipelines inside modern training stacks. Readers familiar with protecting intellectual property in the age of AI know that this statutory architecture is strong on paper but complicated to enforce against web-scale scraping. The 1988 Act does the heavy lifting while the DUA Act handles process.
Remedies under the 1988 Act include injunctions, damages measured on a licensing or user basis, delivery-up orders, and account of profits for serious infringement. Section 97 permits additional damages for flagrant infringement, which rightsholders argue should apply where developers knew the dataset was unlicensed. The Simmons & Simmons note on where UK AI copyright stands explains how these remedies translate into practical litigation strategy. The sticking point is proving that specific works entered a specific training run. That evidentiary gap is what Baroness Kidron's amendments were designed to close.
The March 2026 Statutory Report and Its Policy Reset
Turning to the headline event of the year, the Secretary of State published the Section 137 progress report on 18 March 2026. The release came roughly nine months after the DUA Act received Royal Assent. The CEPIC takeaways on the 18 March 2026 report highlight the three core findings. First, the preferred opt-out plus text and data mining exception is formally off the table. Second, no new legislation will be introduced in the current parliamentary session. Third, work continues on licensing standards, transparency mechanisms, and the broader economic assessment required under the Act.
Policy readers describe the March 2026 report as a reset rather than a retreat, because the government kept its options open. Officials avoided ruling out a future text and data mining exception, instead framing the status quo as a holding pattern pending stronger evidence. The report leans on technical working groups to develop licensing templates, provenance standards, and model-card transparency formats before any statutory change. The Lewis Silkin opt-out cop-out analysis argues that the real signal is the government's unwillingness to legislate against the creative industries during an election cycle. That political reading matters because the next administration could reopen the question with a cleaner mandate.
Impact assessments are the second shoe to drop, due within twelve months of Royal Assent, which places the deadline at 19 June 2026. The Stephenson Harwood briefing on the UK impact assessment notes that officials have begun scoping the modelling. Expected analysis covers licensing revenue flows, compliance costs for developers, inward investment effects, and knock-on impacts on the creative economy. Creator bodies worry that the modelling will treat AI growth as a pre-committed benefit while treating creator losses as speculative. That framing dispute is the policy fight that will consume the second half of 2026.
Text and Data Mining Rules Under Section 29A
Returning to the statute itself, Section 29A of the 1988 Act permits text and data mining of lawfully accessible works, but only for the sole purpose of non-commercial research. The exception was introduced in 2014 to support academic research on scientific literature, and it was never designed for commercial AI training. The AI copyright lawsuits in the US coverage shows how different the US fair use path is from the UK's narrower exception. In practice Section 29A excludes almost every commercial training pipeline operated by major AI firms. It also forbids sharing the ingested corpus with third parties, which complicates downstream fine-tuning and model distillation.
Developers often ask whether Section 29A can be extended by contract or by acquiring an academic partner. Legal counsel typically responds that the exception is strict and non-transferable, and that cross-use blurring commercial and non-commercial purposes risks losing the shield entirely. The Hogan Lovells UK AI copyright outlook for 2026 flags the related risk that a court may read Section 29A narrowly to protect the 1988 Act's bargain with rightsholders. The practical conclusion is that commercial labs cannot plausibly rely on Section 29A to legalize training. Licensing remains the only clean path for commercial scale ingestion of protected UK works.
How Creators Implement Reservation of Rights in the UK
Shifting to the practical creator playbook, UK rightsholders can act today without waiting for secondary legislation to crystallize. The first step is publishing a clear and visible rights reservation on every digital surface where protected work appears. Websites should include a machine-readable TDM reservation signal, such as the TDM Reservation Protocol, alongside human-readable copyright notices in the footer. The AIs impact on intellectual property law background article covers the broader toolkit of exclusive-rights framing. A reservation does not block scraping on its own, but it strengthens the evidential trail for any later infringement claim.
The second step is updating robots.txt and the newer ai.txt file with explicit directives aimed at common AI crawlers. UK publishers increasingly target OpenAI's GPTBot, Google-Extended, ClaudeBot, PerplexityBot, and ByteSpider in their crawl-control files. The directives do not create legal rights, but they establish that the rightsholder communicated an unmistakable refusal. That signal matters when a court weighs the knowledge element of flagrant infringement under Section 97. Teams that lack technical help can use template files published by the News Media Association and the Publishers Association. These templates are updated as new crawler user-agents enter the wild.
The third step is active licensing engagement with the AI labs that are willing to pay. Major publishers have signed deals with OpenAI, Microsoft, Google, Perplexity, Mistral, and Anthropic over the past twenty-four months. Coverage of the empty book protest captures the frustration of authors locked out of those deals when their publishers settled without transparent revenue sharing. Smaller rightsholders can route through collective licensing bodies, including PRS for Music, PPL, and the Copyright Licensing Agency, which are building AI-specific collective schemes. Collective licensing lowers the transaction cost for both sides and widens the pool of works that developers can access cleanly.
The fourth step is building the evidential record needed to enforce rights if licensing fails. Rightsholders should log known crawler visits, archive takedown correspondence, and preserve copies of outputs that reproduce protected material. The earlier UK AI copyright consultation coverage lists the evidence categories courts have accepted in recent US and UK matters. Creators should also document the moment of first discovery, because limitation periods in UK copyright run from the date of infringement or discovery, depending on the claim. A disciplined evidence log converts a vague suspicion into a credible letter before action and a well-pleaded particulars of claim.
Transparency Obligations AI Firms Face in the UK Today
Building on the creator playbook, transparency is the single most contested obligation in UK AI copyright policy. The March 2026 progress report accepts that transparency is the foundation of any workable licensing market, and officials have committed to developing standards within the next twelve months. The AI governance trends and regulations overview shows how the UK's pending transparency framework compares to the EU AI Act's codes of practice. The Kidron amendments would have required developers to disclose specific copyrighted works used in training, which was dropped from the final Act. What remains is a weaker reporting mandate that targets the Secretary of State rather than individual developers.
Developers operating in the UK still face transparency pressure from three other directions. First, the EU AI Act's general-purpose model transparency template is likely to become the de facto UK standard too, because major labs will not maintain two separate disclosure regimes. Second, procurement contracts increasingly require model cards, data provenance disclosures, and copyright warranties from suppliers. Third, the Advertising Standards Authority and Competition and Markets Authority have warned about opaque data practices. Those practices can trigger consumer protection and competition enforcement if they distort the creative market. Each of these pressures narrows the practical space for silence about training data.
Litigation adds a fourth transparency vector that developers often underestimate. Pre-action disclosure applications under the UK Civil Procedure Rules can force developers to reveal training-data details once a rightsholder has a prima facie case. The Canadian news outlets suing OpenAI case illustrates how discovery regimes elsewhere have shifted the information asymmetry. UK rightsholders will likely adopt similar tactics as test cases progress. The combination of regulatory, contractual, and litigation pressure is already pushing some developers toward voluntary provenance disclosure.
The Economic Weight of the Creative Industries
Turning to the economics, the creative industries sit at the heart of why UK AI copyright law 2026 remains politically explosive. The Department for Culture, Media and Sport reports that the sector contributed approximately £124 billion in gross value added in 2024 and employed around 2.4 million people. The MLex analysis of a UK AI licensing market estimated that a clear licensing regime could unlock an incremental £5 to £8 billion in annual licensing revenue by 2030. That figure dwarfs the direct economic benefits that AI labs could claim from an opt-out exception, which is one reason the Treasury quietly reconsidered the preferred option. Economic reality now anchors a debate that began as a technology-policy question.
The distributional story matters as much as the aggregate numbers. Royalty income is heavily concentrated among a small number of major publishers, labels, and estates, with the long tail of individual creators earning modest annual sums. Collective licensing bodies like PRS for Music distribute billions of pounds annually, and those flows depend on reliable copyright protection. The AI threat to UK artists piece unpacks the squeeze on mid-career creators. Weak copyright protection would hit that mid-tier hardest, because they have the most to lose from unlicensed generative substitutes.
Signals From Getty Images v Stability AI
Shifting to the courts, the November 2025 High Court ruling in Getty Images v Stability AI remains the most important judicial signal in UK AI copyright law 2026. Mrs Justice Joanna Smith dismissed most of Getty's primary copyright and database-right claims, finding that Stability's training activity occurred outside UK jurisdiction. The Cleary Gottlieb analysis of the Getty ruling highlights that this jurisdictional finding limits the usefulness of UK copyright law against training conducted on US infrastructure. The judgment did find that Stability's Stable Diffusion model infringed Getty's trademarks by reproducing its watermark on generated outputs. That partial win keeps the litigation story alive while narrowing the available legal theories.
The ruling exposed three practical gaps in the UK framework. First, cross-border training with downstream UK distribution escapes the primary infringement net under current statutory interpretation. Second, output similarity claims face a higher evidential bar than many rightsholders assumed. Third, trademark law may do more work than copyright law when models reproduce distinctive branding or style markers. The Bird & Bird coverage of Stability AI defeating Getty's copyright claims explains how the trademark pathway may become a routine fallback. Getty is appealing, with a hearing expected in the first half of 2027. That appeal will test whether the UK courts will fill the gaps that Parliament has left open.
The ruling also sharpened the practical importance of transparency amendments like Kidron's. Without compulsory disclosure of training sources, future claimants face the same evidential uphill battle that limited Getty's primary claims. The Disney lawsuit challenging AI copyright boundaries story shows how US rightsholders have adopted output-centric strategies that may port to the UK. Expect UK litigation strategy to pivot toward output similarity, trademark infringement, and database-right claims that bypass the training-location issue. The Getty judgment did not close the door on AI copyright litigation. It narrowed the hallway and forced claimants to pick smarter theories.
Make It Fair, McCartney and the Artist Backlash
Looking at public campaigning, the Make it Fair campaign on 25 February 2025 remains the defining image of the UK AI copyright debate. National newspapers including The Times, The Guardian, Daily Mail, Metro, i and the Mirror ran coordinated protest front pages. Each demanded that AI developers obtain permission to train on creative work. The American Songwriter coverage of McCartney's warnings captured the moral weight of the argument. Alongside the newspaper protest, more than 1,000 musicians including Paul McCartney, Kate Bush and Elton John released the silent album Is This What We Want on the same day. The gesture gave policymakers a concrete cultural artifact to point to rather than an abstract policy memo.
The campaign was effective because it bridged elite advocacy with mass public attention. The Computing coverage of the silent protest track traced how the single moved up the UK Official Charts in a week. By the time the Lords began sitting in late February, ministers faced sustained public pressure to drop the opt-out proposal. The UK arts icons demanding AI copyright safeguards coverage documented the open letter signed by every major British creative industry body. The scale of mobilization is one reason the government's eventual retreat in March 2026 felt inevitable rather than voluntary.
How Publishers and Labels Are Rewiring Licensing
Turning to the licensing market, UK publishers and labels have not waited for Parliament to settle the rules. News UK, the Financial Times, Dotdash Meredith, Reuters, and Vox Media have all signed multi-year deals with OpenAI, often for £20 to £50 million per year according to industry reporting. The Computerworld analysis of a licensing-first approach notes that the UK legislative signal has reinforced the commercial logic of these deals. Rightsholders prefer licensing because it monetizes existing assets without new product work. Developers prefer licensing because it reduces the litigation tail and improves model safety by sourcing cleaner data.
The structure of the deals matters more than the headline numbers. Modern AI content licenses include training rights, retrieval rights for RAG pipelines, output attribution guarantees, and warranties about dataset provenance. Some deals include revenue-share clauses tied to downstream commercial usage, especially where the licensor is a specialist publisher of high-value technical content. Music licensing is the frontier, with PRS for Music and major labels pushing for granular stem-level rights that would allow precise tracking of AI model exposure to specific works. The granularity reflects what Hollywood guild deals pioneered in 2023 and 2024.
Smaller rightsholders face a very different reality from the headline deals. Coverage of the empty book protest captured the anger of authors whose publishers signed opaque training-data agreements without transparent revenue sharing. The Society of Authors and the Writers' Guild have pushed back on these clauses in template contract negotiations. Collective licensing bodies are the likely vehicle for extending reach to the long tail of individual creators. Expect PRS for Music, PPL, ALCS, and CLA to launch AI-specific collective schemes in 2026 and 2027.
Risks the Current Framework Poses to Creators
Shifting to the risk map, the status-quo framework carries real downside for UK creators even after the opt-out retreat. The absence of compulsory transparency means rightsholders cannot easily verify whether their work has entered a training run. The Metas AI training using pirated books story illustrates how far a developer can go before independent researchers surface the fact. Without disclosure, creators are forced into expensive forensic investigation to establish a prima facie case. That evidential cost falls hardest on independent authors, illustrators, photographers, and musicians who lack in-house legal teams.
Substitution risk is the second major threat, with generative outputs increasingly serving use cases that previously went to human creators. Stock photography, children's book illustration, incidental music, and voice-over work have all seen downward price pressure. The AI threat to UK artists coverage documents the industry data on commissioned work volumes. Creators are not fully compensated for the value their work contributes to the models that are now substituting for them. Closing that compensation gap is what collective licensing and transparency rules are supposed to achieve.
Attribution and style capture is the third risk, where models reproduce an author's or artist's distinctive voice without naming them. Style is not directly protected under UK copyright law, which creates a loophole for AI reproduction. A model can be prompted to generate work in the manner of a named living author without any clean legal remedy. That protection gap is why passing-off and trademark theories have become more interesting to UK counsel. A deeper explainer on who owns art created by AI unpacks how style-adjacent claims may evolve. Expect contract drafters to add anti-style-capture clauses in creative-industry licenses throughout 2026.
Risks AI Developers Face Operating in the UK
Turning to the developer side, operating in the UK in 2026 carries different but equally real legal exposure. Primary copyright claims may be narrowed by the Getty ruling, but output-based claims, trademark claims, and database-right claims remain live. The AI copyright crises disrupting livestreams coverage shows how fast an output-level incident can trigger regulatory and reputational consequences. Developers also face procurement risk, with public-sector buyers increasingly requiring warranty-backed training-data representations. Running a UK commercial deployment without defensible training provenance is now a measurable business risk.
Reputational risk is harder to quantify yet proves equally material for AI developers operating today. The UK arts icons demanding AI copyright safeguards campaign demonstrated that creator advocacy can shape public perception quickly. Enterprise buyers in insurance, legal services, publishing, and public health increasingly ask about dataset provenance during procurement. Investor pressure is adding a third vector, with ESG-oriented funds asking foundation-model builders about copyright and labor practices. Expect provenance to become a competitive differentiator alongside benchmarks and inference cost.
Ethics of Training Models on Protected British Works
Stepping back from strict law, the ethical argument against unlicensed training rests on three claims. First, creative work is the output of human labor that deserves compensation when it generates downstream commercial value. Second, consent matters even where the legal defense is unsettled, because consent is the moral anchor of the author-reader relationship. Third, aggregate harm to a professional class can corrode cultural diversity even if no individual creator proves economic loss. The AIs impact on intellectual property law piece examines these arguments in detail.
Developers answer with their own ethical case, which deserves fair engagement rather than dismissal. They argue that models trained on broad corpora deliver public benefits in education, accessibility, and productivity that would be impossible under a narrower rights regime. They also argue that transformative use historically has justified broader access to prior art under both US and UK traditions. The honest position is that both sides have real moral claims, and the right policy balance is unlikely to be found by denying the force of either argument. That is why evidence-based policymaking under Section 137 matters as much as advocacy on either side.
Platform-level ethics is a third layer that often gets lost in the legal debate. The Tollbit fighting AI content scraping coverage shows that even technically-reserved content gets scraped at scale. Developers who respect robots.txt and ai.txt directives differentiate themselves from scrapers who ignore them. Transparent model cards, provenance disclosures, and clear attribution in generated outputs are low-cost ethical practices that most mature labs now adopt voluntarily. Those voluntary standards are where regulation will crystallize once the Section 137 evidence base is complete.
The Future of UK AI Copyright Law Beyond 2026
Looking ahead, The UK AI copyright regime of 2026 is a staging post rather than a settled destination. The impact assessment due in June 2026 will test the government's economic argument, and the broader economic study due in December 2026 will shape the next legislative vehicle. The Hogan Lovells UK AI copyright outlook for 2026 flags secondary legislation under Section 137 as the most likely near-term reform vector. Watch for statutory instruments on transparency and licensing standards before the end of the parliamentary session. The electoral calendar makes any larger primary legislation unlikely before 2027.
Several external pressures will shape the UK path even if domestic politics stalls. The EU AI Act's codes of practice on general-purpose models already push UK developers toward higher transparency, and divergence costs will push UK rules to follow. US litigation outcomes in the New York Times v OpenAI matter and the Thomson Reuters v Ross Intelligence appeal will set reference points that UK courts can draw on. The AI copyright lawsuits in the US coverage tracks the main US cases. The pattern of settlements and judgments across the pond will heavily influence what the UK Treasury considers a defensible policy position.
Three futures are plausible depending on how the evidence lands and how the next election goes. One future preserves the status quo indefinitely, forcing all compensation through private licensing and collective schemes. A second future reintroduces a tightly scoped text and data mining exception with mandatory licensing, modelled on recent proposals from Switzerland and the Nordics. A third future legislates compulsory transparency and a statutory remuneration right, giving creators a floor of economic protection even where licensing deals fail to close. That third path is the one the Kidron amendments gestured toward, and the one most favored by UK creator bodies.
Whichever future wins, the January 2026 to December 2026 window is the one that will set the terms. The 2026 UK AI copyright framework is being written in Section 137 reports, Lords debates, Getty appeal pleadings, and bilateral licensing contracts rather than in a single statute. The Freevacy coverage of the comprehensive AI bill delay underlines how a planned omnibus AI act was pushed back precisely because of unresolved copyright questions. Readers who want to shape the outcome should watch the IPO working groups, the DUA Act impact assessment consultation, and the Getty appeal filings. Those three threads will set the direction of UK AI copyright law for the second half of the decade.
A Chart From AIplusInfo
UK AI Copyright Reform Options Compared
Composite score and projected 2030 UK licensing revenue by reform option. Use the toggle to switch views.
Composite score weights creator income, developer certainty and transparency equally, drawn from the March 2026 Section 137 report and MLex industry analysis.
Source: March 2026 Section 137 progress report and MLex analysis of UK AI licensing.
Key Insights on the 2026 UK AI Copyright Debate
- The Intellectual Property Office received approximately 11,500 consultation responses during the December 2024 to February 2025 window. That volume ran more than tenfold typical UK IP consultations, confirming political feeling across both AI and creative sectors.
- Section 137 of the Data Use and Access Act 2025 locks the government into a twelve-month impact assessment timeline due by 19 June 2026. The deadline forces an evidence-based decision window that creator bodies now treat as a procedural win.
- The UK creative industries contribute approximately £124 billion in gross value added and employ 2.4 million people, per the Department for Culture DCMS 2024 estimate. That substantial economic weight firmly reframes the entire AI copyright policy debate well beyond pure technology growth narratives.
- In November 2025 the High Court dismissed most of Getty's primary copyright claims in Getty Images v Stability AI. The court found training outside UK jurisdiction fell outside the 1988 Act, exposing a cross-border enforcement gap Parliament has yet to fill.
- Baroness Beeban Kidron forced three successive Lords defeats on the Data Use and Access Bill in 2025. Each demanded that AI firms disclose specific copyrighted works used in training, turning transparency into a Commons-Lords constitutional confrontation.
- MLex industry analysis estimated a clear UK AI licensing framework could unlock £5 to £8 billion in annual licensing revenue by 2030. That potential dwarfs any direct benefit a text and data mining exception could deliver to AI developers alone.
- Section 29A of the 1988 Act still restricts text and data mining to non-commercial research purposes only. No commercial AI training on protected UK works can plausibly rely on this exception as a legal shield.
- More than 1,000 artists including Paul McCartney released the silent protest album Is This What We Want on 25 February 2025. The cultural artifact moved the Commons debate as fast as any formal economic submission could.
These signals combine into a single policy picture that favours deliberate drift over sweeping reform. The government has stepped back from the opt-out model yet refused to legislate a strong transparency regime, leaving creators to rely on contract, campaign, and litigation to defend their position. Developers now operate under heightened reputational scrutiny while retaining real operational flexibility, which is exactly the kind of ambiguity that makes risk management expensive. Economic evidence increasingly supports a licensing-first framework, especially when the £124 billion creative-industries contribution and the £5 to £8 billion licensing opportunity are weighed together. The question for the second half of 2026 is whether the Section 137 impact assessment hardens that case or dilutes it under Treasury-driven growth modelling. UK AI copyright law 2026 is a holding pattern, and the holding pattern is doing most of the work AI firms wanted the dropped exception to do.
How Each Reform Option Compares on Creator Protection
The table below scores each reform option across seven policy dimensions, drawing on the March 2026 progress report and the live licensing-market debate. Columns span the status quo, the dropped opt-out, a licensing-first regime, and a transparency plus remuneration right. Each row isolates a different policy variable so readers can weigh the trade-offs directly. The scoring draws on the published consultation evidence and industry licensing data. It gives a side-by-side view that drove the Lords amendments and the eventual government retreat. Readers can use the table to anticipate where the Section 137 impact assessment is most likely to land in June 2026.
| Dimension | Status quo (CDPA 1988) | Opt-out plus TDM exception | Licensing-first regime | Transparency plus remuneration right |
|---|---|---|---|---|
| Creator income | Moderate, uneven | Low, concentrated | High, concentrated | High, broadly distributed |
| Transparency on training data | Weak, discovery-driven | Weak, opt-out-only | Moderate, contract-driven | Strong, statutory |
| Developer certainty | Low, litigation-exposed | High for compliant scrapers | Moderate, licensing-dependent | Moderate, disclosure-conditional |
| Enforcement cost for creators | High | Very high | Low with collective bodies | Low with statutory audit |
| Speed to effect | Immediate | Fast, secondary legislation | Moderate, 18-24 months | Slow, 24-36 months |
| International alignment | Diverges from EU TDM carve-outs | Aligns with EU Article 4 | Aligns with US licensing deals | Partial alignment with EU AI Act |
| Political durability | High, inertia-driven | Low, politically toxic | Moderate, cross-party support | High, creator-industry backed |
Real-World Examples of UK AI Copyright in Practice
Three enforcement stories show how UK AI copyright has moved from policy theory to measurable commercial consequence. Each example pairs a concrete action with a quantified outcome and a documented limitation.
Getty Images and the Watermark Pathway
Getty Images filed its UK claim in January 2023 and ran the matter through to a High Court judgment on 4 November 2025 before Mrs Justice Joanna Smith. Getty argued that Stability AI copied approximately 12 million watermarked Getty images to train Stable Diffusion, pleading primary copyright, database rights, and trademark infringement. The court dismissed the primary copyright and database claims on jurisdictional grounds because the training occurred outside the UK, which is detailed in the Bird & Bird case analysis. It did uphold trademark infringement where Stable Diffusion reproduced the Getty watermark on generated outputs. The court awarded injunctive relief and damages, cutting primary copyright claims by roughly 90 percent. The limitation is that the trademark pathway only helps rightsholders whose branded assets travel into outputs, which is a narrow slice of the UK rightsholder base. Getty filed an appeal in December 2025 with a hearing expected in the first half of 2027, and the ruling now sets the default playbook for UK AI copyright strategy.
The Make It Fair Newspaper Protest
On 25 February 2025 nineteen UK national titles ran a coordinated Make it Fair front-page campaign. Participating papers included The Times, The Guardian, Daily Mail, Metro, the Mirror, Daily Express and i. The campaign was timed to the final day of the consultation window and the launch of the McCartney-backed silent album. The measurable outcome was immediate political pressure that fed into the Lords debates and the eventual Kidron amendments, with ministers acknowledging the campaign in Commons responses the following week. The limitation was that newspapers hold a specific commercial stake, so critics argued the campaign conflated proprietor interests with wider creator welfare. Reach data from Enders Analysis estimated the campaign hit 29 million UK adults in one day, a 45 percent lift that converted editorial attention into political capital in days. The protest proved that AI copyright had moved from an elite legal debate into mainstream political territory within three months.
The Financial Times OpenAI Licensing Deal
The Financial Times signed a multi-year content partnership with OpenAI in April 2024 that was expanded in 2025 to include training data and retrieval augmentation rights. Industry reporting placed the annual value between £7 and £15 million for FT content ingestion, model training, and attributed ChatGPT answers. The arrangement is projected to grow at 20 percent year-on-year through 2027, delivering a template other UK publishers have since adapted. The News UK and Dotdash Meredith deals signed later in 2024, detailed in the Computerworld licensing-first analysis, followed quickly. The limitation is that the FT is unusually well positioned, with proprietary business content, a global subscriber base, and existing B2B licensing infrastructure. Smaller UK publishers face harder negotiations because they lack the FT's data assets. The FT model also raised internal journalist questions about by-line attribution in generated outputs, which the deal addressed through specific guarantees.
Recommended by AIplusInfo
Books to go deeper on UK AI copyright law 2026
Three essential titles that frame the policy, legal, and ethical arguments covered above.
As an Amazon Associate, AIplusInfo earns from qualifying purchases.
Book
The Cambridge Handbook of Generative AI and the Law
The scholarly reference for how UK, EU, and US copyright regimes are adapting to generative AI, directly relevant to the Section 137 debate.
Buy on AmazonBook
The Alignment Problem: Machine Learning and Human Values
Brian Christian's primer on how machine learning systems inherit the values of their training data, which anchors the UK training-ethics debate.
Buy on AmazonBook
Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence
Kate Crawford's investigation of AI's labor, extraction, and policy footprint frames the political economy driving the UK copyright fight.
Buy on AmazonCase Studies in UK AI Copyright Disputes and Deals
Three UK case studies demonstrate the live gap between creator ambition and current enforcement reach. Each case documents a problem, a solution, measurable impact, and the limitation that still constrains the outcome today.
Case Study: Baroness Kidron's Transparency Amendments
Baroness Beeban Kidron tabled a sequence of amendments to the Data Use and Access Bill between January and April 2025. The amendments would have required AI firms operating in the UK to disclose specific copyrighted works used in training. The problem she targeted was the evidential asymmetry that leaves creators unable to prove ingestion of their work by any specific developer. Her solution combined mandatory disclosure and a regulatory enforcement role for the Intellectual Property Office, with billions of licensing revenue at stake across the creative sector. The Lords passed her amendments three times, inflicting three successive defeats on the Bill, as covered in the PYMNTS reporting of the April 2025 vote. The limitation is that the Commons stripped the amendments out each time, substituting the weaker Section 137 reporting mandate that eventually passed. The controversy exposed a constitutional split, with the Lords acting as the creative industries' effective voice while the Commons protected ministerial room for manoeuvre. The outcome falls short of the mandatory disclosure Kidron sought, yet it embedded transparency as a live UK copyright question that will return in the next session.
Case Study: News UK's OpenAI Content Partnership
News UK publishes The Sun, The Times, and The Sunday Times. It signed a multi-year content and licensing partnership with OpenAI in May 2024, reportedly worth £10 to £20 million annually. The problem News UK faced was the risk of unlicensed ingestion of its paywalled journalism, combined with the opportunity cost of not monetizing training rights that competitors were already selling. The solution was a direct bilateral deal with explicit training rights, retrieval augmentation rights, and attribution guarantees in ChatGPT responses. The impact included a protected revenue stream and new ChatGPT collaborations citing News UK journalism, which industry commentators treated as the UK template for broadsheet AI licensing. The limitation highlighted by the empty book protest coverage was that staff journalists were not consulted on by-line usage before the deal was signed. The case study illustrates both the commercial logic of licensing-first strategy and the labour-relations risks when creators are not partners in the negotiation.
Case Study: The Meta Pirated Books Discovery
Court filings in late 2024 and early 2025 revealed that Meta had used the LibGen pirated-books dataset to train its Llama 3 family of models. The problem for UK authors was straightforward and well documented through court filings. The LibGen archive contains more than 7.5 million titles ingested without permission, including works by UK writers under active copyright, representing tens of millions of pounds in lost revenue. The solution advanced by UK author bodies combined US litigation support, UK-side regulatory pressure, and a Society of Authors advocacy campaign documented in Metas AI training using pirated books. UK author organizations cited LibGen in consultation responses, Baroness Kidron referenced it in Lords debates, and Books3 became shorthand for the opacity creators were fighting. The limitation is that no UK court has yet ruled on cross-border infringement claims against Meta, so remedies remain theoretical for most UK rightsholders. The controversy also surfaced an uncomfortable question for publishers, since some of the pirated texts came from legitimate editions whose licensing terms had been evaded in acquisition. The outcome has been renewed political support for the transparency obligations that the Kidron amendments proposed, grounding the policy debate in a concrete and ongoing wrong.
Common Questions About UK AI Copyright Law 2026
The UK AI copyright regime in 2026 keeps the 1988 Copyright, Designs and Patents Act in force. Section 137 of the Data Use and Access Act 2025 adds a required progress report and impact assessment. The government dropped its preferred opt-out exception in March 2026. No new primary legislation is expected in the current session. Creators should treat 2026 as a holding year, acting on licensing and transparency rather than waiting for new statutes.
There is no new primary statute that explicitly favors AI firms. The December 2024 consultation proposed an opt-out model that would have, but Parliament rejected that approach. The current framework still requires permission or a narrow exception for lawful training. The imbalance comes from weak enforcement rather than from any new provision tilted toward AI developers.
No. Section 137 requires the Secretary of State to publish a progress report and impact assessment, not AI firms to disclose training data. The Kidron amendments would have mandated such disclosure but were stripped from the final Act. Compulsory transparency on training data remains a live legislative question in the UK Parliament today. Any binding disclosure obligation would need either secondary legislation or new primary legislation.
Jurisdiction is the main obstacle, as the Getty v Stability AI ruling confirmed. UK courts generally require the infringing act to occur within the jurisdiction, and training outside the UK falls outside the 1988 Act. Claims may still succeed on output similarity, trademark, or database-right theories. Specialist UK counsel can advise on the strongest theory for your facts.
Section 29A of the 1988 Act allows text and data mining of lawfully accessible works for the sole purpose of non-commercial research. It cannot support commercial AI training, cannot be transferred to third parties, and cannot be relied on where any commercial purpose is involved. Commercial labs need licensing or a different exception to train lawfully.
Creators should publish a visible rights reservation, update robots.txt and ai.txt to block common AI crawlers, and work through collective licensing bodies where possible. These steps do not create new legal rights but strengthen the evidential record for any later claim. They also signal to compliant developers that the content is unavailable without a license. Technical steps and legal rights-reservation steps work best together as a complete creator defence.
Make it Fair was a coordinated newspaper and artist campaign on 25 February 2025 demanding that AI developers obtain permission to train on creative work. The campaign coincided with the McCartney-backed silent protest album and the Society of Authors mobilisation. The campaign did not change the statute, but it created the political pressure that pushed the government to drop the opt-out proposal. It also helped embed transparency into the Section 137 reporting mandate.
Getty lost most of its primary copyright and database claims but won on trademark infringement. The court found Stability AI's training occurred outside UK jurisdiction, which took the primary copyright pathway off the table. The trademark finding keeps litigation alive and sets a template for future claims. Getty has appealed, with a hearing expected in the first half of 2027.
No. The UK has no equivalent to the EU AI Act and does not have an Article 4-style text and data mining exception with an opt-out. UK developers face a more restrictive statutory baseline but softer practical enforcement. Alignment pressure is real, with UK developers tending to adopt EU AI Act transparency templates to avoid maintaining two regimes.
Not always, because outputs may reproduce protected elements of training data. Outputs may reproduce protected elements of training data, which can trigger infringement claims against commercial users. UK law also treats originality cautiously, so purely AI generated work may lack copyright protection in its own right. Commercial users should perform output screening, keep provenance records, and secure warranties from the AI provider.
The impact assessment is due by 19 June 2026 and the broader economic study by late 2026, so a clearer policy direction should emerge by early 2027. Any primary legislation is unlikely before 2027 given the parliamentary calendar. Secondary legislation on transparency and licensing standards is possible before then. The Getty appeal in 2027 will also shape the position.
News UK, the Financial Times, the Guardian Media Group, Reach, and several specialist B2B publishers have signed deals with OpenAI, Microsoft, Google, or Perplexity. Reported values range from £5 million to over £20 million annually. Deals typically cover training rights, retrieval rights, and attribution guarantees across both ingestion and generated output. Smaller UK publishers often negotiate through collective licensing bodies rather than running direct bilateral deals.
Developers should license training data where feasible, keep documented provenance records, respect robots.txt and ai.txt directives, and avoid ingesting known pirated datasets. Publish a model card describing data sources and known limitations. Prepare for disclosure obligations by maintaining an audit trail now, before statutory rules bite. Legal counsel should review training pipelines against the current Section 29A limits.