For a business building commercial value on AI-generated content, the United Kingdom presents a peculiar advantage – and a peculiar trap. English copyright law contains a provision that most jurisdictions lack: statutory protection for computer-generated works where no human author exists. Yet that provision was drafted in 1988, before neural networks, large language models, or generative AI were conceivable. The result is a body of law that is simultaneously ahead of its time and wholly unprepared for the moment it now faces.
AI-generated works in the United Kingdom may attract copyright protection under intellectual property legislation, which specifically provides for computer-generated works assigned to the person who makes the necessary arrangements for their creation. This protection lasts for fifty years from the date the work is made. The critical unresolved question is whether that provision applies to outputs from modern generative AI systems – and courts have not yet delivered a definitive answer.
This analysis covers the doctrinal foundation of UK copyright in AI-generated works, the competing interpretations currently circulating in legal practice and commentary. The practical gap between statute and commercial reality, cross-border implications for businesses operating between the UK and Europe. Additionally, the strategic steps that IP-conscious organisations should take now.
The doctrinal foundation: a provision ahead of its time
UK intellectual property legislation recognises a distinct category of literary, dramatic, musical, and artistic works: those that are computer-generated. The statute defines a computer-generated work as one generated by a computer in circumstances where there is no human author. Ownership vests in the person who makes the necessary arrangements for the creation of the work. This is a departure from the general rule, which requires a human author exercising skill, labour, and judgment.
The provision was originally designed with a modest ambition. Drafters in the 1980s had in mind software that generated routine outputs – financial tables, weather maps, procedurally composed music – where human creative input was real but diffuse. The human who configured the program was close enough to the output to justify a property right. The situation today is fundamentally different. A user who types a prompt into a generative AI system has a much more attenuated relationship with the resulting work. The model's training data, architecture, and parameter weights do the creative heavy lifting. The question of whether the user "makes the necessary arrangements" is genuinely contested.
UK courts have historically interpreted copyright protection strictly. The Supremo Tribunal de Justiça analogy does not apply here – the relevant forum is the High Court of England and Wales and, on appeal, the Supreme Court of the United Kingdom. These courts have consistently held that copyright subsists only in works that reflect human creative choices. They have also shown reluctance to extend protection where the claimed author's contribution is trivial or mechanical. Applying that doctrine to AI-generated works pulls in the opposite direction from the computer-generated work provision. The tension between the two positions remains unresolved.
A further doctrinal complication arises from originality requirements. Under English intellectual property legislation, a work must be the author's own intellectual creation to qualify for protection. That standard derives from EU copyright law, adopted into domestic practice before Brexit. Post-Brexit, UK courts are no longer bound to follow that formulation. However, the standard remains embedded in the case law developed while the UK was an EU member. Whether courts will progressively diverge from EU originality doctrine – potentially expanding or contracting protection for AI outputs – is an open question that no judgment has yet settled.
The Copyright, Designs and Patents Act framework also raises the question of moral rights. Moral rights – including the right of attribution and the right of integrity – vest in the author of a work. Where a work is computer-generated and has no human author, no moral rights arise. For businesses exploiting AI-generated content commercially, this is advantageous: there is no risk of a moral rights claim from a human creator challenging modifications or commercial uses. The trade-off is that the computer-generated work protection is weaker and shorter in duration than standard copyright.
Competing interpretations and the gap between statute and practice
Three broad positions have emerged in legal practice and academic commentary, each with distinct commercial consequences.
The first position holds that the computer-generated work provision applies directly to generative AI outputs. On this view, the person who inputs the prompt, configures the system, or deploys the API makes the necessary arrangements. They receive a fifty-year copyright. This interpretation favours businesses that build products on generative AI. It gives them a defensible proprietary claim over outputs. It also aligns with the legislature's apparent intent to ensure that commercially valuable computer-generated works receive protection. The difficulty is that courts have not endorsed this position in the context of large language models, diffusion models, or equivalent systems. Practitioners in the UK note that relying on this interpretation without judicial confirmation carries meaningful legal risk.
The second position holds that AI-generated works receive no copyright protection at all. Under this view, the computer-generated work provision was designed for a different technological context. Modern generative AI systems do not simply execute deterministic instructions; they produce outputs shaped by billions of parameters derived from training data. The human contribution – a prompt – is too thin to constitute making the necessary arrangements. Without a human author meeting the originality threshold, and without a plausible application of the computer-generated work provision, the work falls into the public domain on creation. This interpretation is uncomfortable for businesses that have invested in AI-generated content. It also raises difficult questions about training data: if outputs are not protected, the economic logic of proprietary AI development is partially undermined.
The third position – arguably the most practically influential – holds that the answer depends on the specific facts of the arrangement. The level of human involvement, the nature of the prompting process, the degree of curation applied to outputs, and the extent of post-generation editing all bear on whether the computer-generated work provision applies. On this view, a business that exercises substantive creative judgment in selecting, refining, and assembling AI outputs will be in a stronger position than one that deploys a model in a fully automated pipeline. This intermediate position gives businesses a lever: by documenting and structuring human involvement, they can build a stronger claim to protection.
The gap between statute and practice is most visible in commercial transactions. Technology licensing agreements for AI-generated content routinely proceed on the assumption that copyright subsists – because the alternative, a product built on unprotected content, is commercially unworkable. Yet the legal foundation for that assumption has never been judicially tested in the context of modern generative AI. This creates a latent risk in IP warranties and representations given in M&A transactions, licensing deals, and investment rounds. A buyer or investor relying on the strength of an AI content portfolio may discover, if the point is ever litigated, that the portfolio rests on contested legal ground.
Software liability is a related concern. Where AI-generated content causes harm – defamatory statements, infringing reproductions of third-party material, or misleading commercial communications – the question of who bears responsibility is unresolved. English tort law places liability on the person who publishes harmful content. In an AI context, that may be the deployer, the developer, or both. The High Court has not yet established a clear doctrine of algorithmic accountability in this space. Businesses operating AI-content pipelines should treat liability allocation as an active legal risk, not a theoretical one.
Training data is another fault line. The outputs of a generative AI model are downstream of its training corpus. Where that corpus includes copyright-protected works reproduced without licence, the model's outputs may themselves constitute infringements – or may be tainted by the infringement at the training stage. UK intellectual property legislation provides a text and data mining exception for non-commercial research. It does not provide a general exception for commercial AI training. This is an area where the law has not kept pace with practice, and where litigation is increasingly likely.
For a deeper review of how UK AI and technology regulation interacts with IP protection, see our analysis of AI and technology law in the United Kingdom.
To discuss how intellectual property doctrine applies to your specific AI content pipeline in the UK, contact us at info@ferrazwhitmore.com.
Cross-border implications for European businesses
For businesses operating between the United Kingdom and the European Union, the post-Brexit divergence in IP law creates a set of practical challenges that are easy to underestimate.
The EU has taken a different approach to AI-generated works. EU copyright law does not contain a computer-generated work provision equivalent to the UK's. Under EU doctrine, copyright requires a human author expressing personality through creative choices. An AI system cannot be an author. AI outputs, on this view, do not attract EU copyright protection unless a human has exercised sufficient creative control in their production. The European Parliament's position, reinforced in legislative discussions around the EU AI Act compliance regime, is that AI systems are tools, not authors.
This divergence has immediate consequences for businesses with EU and UK operations. A business that successfully establishes copyright in an AI-generated work under UK law cannot assume that protection extends across the EU. Conversely, a business that structures its AI content production to maximise human creative involvement – to satisfy the EU originality standard – may or may not benefit from the UK computer-generated work provision. The two systems are optimising for different things.
The EU AI Act introduces transparency obligations for AI-generated content, including requirements to label certain categories of synthetic media. These obligations apply to businesses operating in EU member states, regardless of where their AI infrastructure is located. A UK-based business deploying AI-generated content to EU consumers may need to comply with EU AI Act transparency requirements even if it has no EU establishment. The interaction between AI Act compliance obligations and copyright ownership questions creates a two-layered regulatory challenge.
Cross-border technology licensing of AI outputs compounds the difficulty. A licence agreement that assigns or exclusively licenses AI-generated content may be valid and effective in the UK. There. Copyright subsists. However, may transfer nothing of value in the EU. There, the same content may be unprotected. Drafters of AI content licences must address this gap explicitly. identifying the applicable law. The jurisdictions in which protection is claimed. Additionally, the mechanism for allocation of rights if protection is subsequently found not to subsist.
Digital services operating across the UK-EU border also face the question of digital services regulation. The UK's approach to platform liability and content moderation has diverged from the EU Digital Services Act regime. Where AI-generated content is hosted, distributed, or monetised through digital platforms, the applicable regulatory regime will depend on where the platform is established and where its users are located. Businesses with dual-market exposure should not assume that compliance in one jurisdiction satisfies the other.
For businesses with cross-border IP portfolios, our analysis of intellectual property law in the United Kingdom provides a foundation for understanding how UK doctrine interacts with equivalent EU protections.
A comparative perspective on how similar questions have developed in a civil law jurisdiction is available in our deep analysis of AI-generated works and IP in Portugal.
For a tailored strategy on AI-generated content protection and cross-border licensing, reach out to info@ferrazwhitmore.com.
Strategic implications for IP-intensive businesses
The uncertainty described above is not a reason to defer action. It is a reason to structure action carefully. Businesses that wait for judicial clarity before building IP strategies around AI-generated content will find that competitors who acted earlier have established stronger positions – in licensing relationships, in transactional representations, and in commercial precedent.
The first strategic priority is documentation. Businesses should maintain detailed records of human involvement at every stage of AI content production. This includes records of prompt design, selection criteria, editorial review processes, and post-generation modifications. These records serve two purposes. They support a claim that the computer-generated work provision applies – because they demonstrate that identified persons made the necessary arrangements. They also support a higher-tier copyright claim if human creative input reaches the level of originality required under standard copyright doctrine.
The second priority is contractual clarity. Technology licensing agreements, employment contracts, and commissioning arrangements should explicitly address ownership of AI-generated outputs. Under UK intellectual property legislation, copyright in works created by employees in the course of employment vests in the employer. Where external contractors or AI service providers are involved, ownership does not transfer automatically. Businesses that have deployed AI tools without updating their standard contracts may have created ambiguity about who owns what. Practitioners note that this ambiguity is particularly acute where AI tools are integrated into creative workflows – design, marketing, software development – without legal review of the resulting ownership structure.
The third priority is risk allocation in transactions. M&A due diligence, investment documentation, and licensing warranties should reflect the current uncertainty in AI copyright protection. A blanket warranty that the target company owns copyright in all AI-generated content is legally vulnerable. A more defensible approach is to warrant that the company has taken commercially reasonable steps to establish and document its claims to protection, and to carve out residual uncertainty explicitly. HMRC and the Financial Conduct Authority (FCA) do not currently impose specific reporting obligations tied to AI copyright status, but the position is evolving.
The fourth priority is training data hygiene. Businesses that develop or commission bespoke AI models should obtain clear representations from developers about the composition and licensing status of training data. Where a model has been trained on unlicensed copyright material, the outputs carry infringement risk. This risk is not extinguished by the computer-generated work provision – that provision addresses ownership of outputs, not the legality of how the model was built. Companies House filings and corporate records do not capture training data provenance, so businesses must establish their own internal documentation standards.
The fifth priority is monitoring the legislative trajectory. The UK Intellectual Property Office has consulted on whether changes to copyright law are needed to address AI. The government has considered options ranging from extending text and data mining exceptions to overhauling the computer-generated work provision. No legislative change has yet been enacted. But the direction of travel is clear: Parliament will need to address these questions. Businesses that have engaged with the consultation process and understand the competing policy positions will be better placed to adapt when legislation emerges.
A self-assessment for businesses building commercial strategies on AI-generated content in the UK:
- Can you identify, by name and role, the person who made the necessary arrangements for each category of AI-generated output your business relies on?
- Do your standard contracts address ownership of AI-generated works, including works produced by third-party AI tools integrated into your workflows?
- Have you obtained representations from AI model providers about training data provenance and licensing?
- Do your IP warranties in current transactions reflect the doctrinal uncertainty, rather than assuming protection subsists?
- Are you monitoring UK legislative developments and EU AI Act compliance obligations in parallel?
If the answer to more than two of these questions is no, the business is carrying unquantified IP risk that a structured legal review can address.
Outlook: where the law is heading
The UK's position on AI-generated works is at an inflection point. The computer-generated work provision gives the UK a distinctive starting position – but that provision was never designed to carry the weight now placed on it. Courts will eventually be asked to rule on whether generative AI outputs qualify. When they do, the answer will depend on how judges resolve the tension between the statutory text, the originality requirements developed through decades of case law. Additionally. The policy question of whether copyright protection advances or impedes beneficial AI development.
The Supreme Court of the United Kingdom has shown willingness to depart from EU doctrine in IP matters post-Brexit. It has also shown caution about extending property rights in ways that Parliament has not clearly authorised. Both tendencies are relevant here. A restrictive ruling – one that treats AI-generated outputs as unprotected unless substantial human creative input is demonstrable – would place the UK closer to the EU position than the statute implies. An expansive ruling – one that treats prompt engineering and model deployment as making the necessary arrangements – would give the UK a more AI-friendly IP regime than any other major jurisdiction currently offers.
The EU AI Act compliance timeline is running in parallel. As transparency and accountability obligations for AI-generated content take effect across EU member states, UK businesses with EU market exposure will face regulatory pressure that indirectly shapes their AI content strategies. Businesses that build compliance structures for EU purposes will, in many cases, also be building the documentation infrastructure that strengthens their UK copyright claims.
The Financial Services Authority and FCA have begun engaging with questions of algorithmic accountability in regulated industries. Where AI-generated content intersects with financial promotions, investment advice, or regulated communications, an additional layer of sector-specific regulation applies. This layer operates independently of IP law but interacts with it in practice: content that cannot be attributed to a responsible human author may also fail to meet financial services content standards.
Practitioners in the UK note a convergence of three pressures: commercial demand for clear IP rules around AI outputs. Legislative hesitancy pending broader AI governance decisions. Additionally, judicial reluctance to resolve doctrinal tensions without clear statutory guidance. The result is a period of deliberate legal ambiguity. For businesses, ambiguity is not neutral. It systematically advantages those who plan for multiple outcomes over those who assume a single legal position will prevail.
Frequently asked questions
Q: Does UK law currently protect AI-generated works under copyright?
A: UK intellectual property legislation contains a specific provision for computer-generated works, under which copyright may vest in the person who makes the necessary arrangements for the work's creation. Whether this provision applies to outputs from modern generative AI systems has not been definitively resolved by the High Court or Supreme Court. Businesses relying on this protection should document human involvement carefully and take legal advice on the strength of their specific claims.
Q: How long does copyright last for AI-generated works in the United Kingdom?
A: A common misconception is that AI-generated works receive the same duration of protection as works by human authors. They do not. Under UK intellectual property legislation, computer-generated works are protected for fifty years from the date they are made – significantly shorter than the life-plus-seventy-year term that applies to works by human authors. Businesses building long-term commercial value on AI-generated content should factor this shorter term into their IP strategy. Engaging a lawyer in the United Kingdom with experience in technology licensing can help structure arrangements to maximise the available protection.
Q: How does the EU AI Act affect UK businesses using AI-generated content?
A: The EU AI Act imposes transparency and labelling obligations on certain categories of AI-generated content distributed to users in EU member states. A UK-based business with no EU establishment may still need to comply if it targets EU consumers. These obligations operate independently of copyright ownership questions but interact with them in practice: documentation maintained for AI Act compliance purposes can also support UK copyright claims. As an international law firm in the United Kingdom and European markets, Ferraz & Whitmore advises on both dimensions in an integrated manner.
About Ferraz & Whitmore
Ferraz & Whitmore is an international law firm based in Lisbon, advising business clients across 46 jurisdictions. Our team combines Portuguese civil law expertise with English common law tradition to deliver cross-border legal solutions in AI and technology law, intellectual property protection, and technology licensing. We advise technology companies, institutional investors, and in-house legal teams navigating the intersection of AI regulation and IP strategy across European and international markets. The firm's IP and technology practice covers AI Act compliance, software liability, algorithmic accountability, and digital services regulation across both civil law and common law systems. Our practitioners have advised on AI-related IP matters before the High Court of England and Wales and in cross-border licensing transactions spanning EU and UK jurisdictions. To discuss your AI-generated content strategy and IP position in the United Kingdom, contact us at info@ferrazwhitmore.com.
Disclaimer: This publication is provided for informational purposes only and does not constitute legal advice. The information herein should not be relied upon as a substitute for professional legal counsel tailored to your specific circumstances. Ferraz & Whitmore assumes no liability for actions taken or not taken based on the contents of this material. For advice regarding your particular situation, please contact info@ferrazwhitmore.com.