AI & Intellectual Property 2026

Welcome to the Guide

Artificial intelligence (AI) has moved from experimental deployment to operational reality. It is now embedded in professional practice across industries and jurisdictions, shaping how work gets done. This Introduction was itself drafted through an oral conversation with an AI assistant, iterating ideas in real time through dialogue spoken, not typed. We note this not as a disclaimer but as an illustration: AI is no longer something lawyers think about. It is something lawyers now work with, including when creating guidance on AI itself.

The question this guide addresses is how intellectual property law applies to AI. The answer, so far, is: largely through existing legislation and settled precedent. There is no separate comprehensive IP regime for AI. Existing doctrine on patents, copyrights, trade secrets, trade marks, designs applies to AI. The dominant body of law that applies to AI and IP remains existing legislation, not wholly new regimes.

Who This Guide is For

This guide addresses two audiences, and each brings different expertise to the terrain ahead.

If you are an intellectual property lawyer or in-house counsel with deep familiarity across patents, copyrights, trade secrets, trade marks, or designs, you already understand something fundamental: intellectual property is the discipline of applying established legal doctrine to fast-moving technology and culture. Patent law, by definition, addresses inventions that did not exist before. Copyright, by definition, protects expression created within living memory – rarely more than a century old. Trade secret law protects confidential advantage in a world where competitive advantage shifts constantly. IP lawyers are accustomed to the task of decanting new wine into old carafes. So the challenge with AI is not learning a new legal discipline; it is recognising where AI’s particular characteristics – its opacity, its scale, its potential autonomy – strain existing legislation and precedent in ways that may require not just application but recalibration.

If you are an AI technologist or in-house counsel involved in AI, you may know the technology deeply but be less familiar with IP. This guide will show you that intellectual property is not monolithic. It breaks into distinct regimes, each with different doctrines. A problem that appears to be about copyright – training data and outputs – may turn on patent law, trade secrets, or trade mark law. And critically, jurisdiction matters in ways that may surprise you if you are accustomed to thinking of technology as borderless. The internet may be global, but intellectual property law is not.

AI and Intellectual Property is Broader Than Copyright

When most people consider intellectual property and AI, they think of copyright. Specifically, they analyse whether training an AI model on copyrighted works constitutes infringement, and whether AI-generated outputs infringe the works they draw upon. That association is understandable, but it is incomplete.

Intellectual property is a family of distinct legal rights and protections, each creating and protecting a different species of right. Patent law protects technical invention, raising questions about whether AI systems can be named as inventors, whether AI-assisted inventions meet inventive step requirements, and how disclosure obligations apply when an AI system contributes to the inventive process. Trade secret law protects a particular type of highly coveted confidential information – the model weights, training methodologies, and architectural choices that often represent an AI company’s most valuable and most vulnerable assets. And of course, is the input of a trade secret into an AI a breach? Trade mark law protects brands, and AI systems increasingly generate names, logos, and content that may infringe existing marks or raise novel questions about ownership of AI-generated brand assets.

As AI touches every area of commerce, by definition it will touch every area that intellectual property protects within commerce. The guide that follows addresses all of these IP rights, jurisdiction by jurisdiction, because a narrow focus on copyright alone would leave readers unprepared for the breadth of issues AI will raise.

Why Copyright Has Dominated the Conversation

If intellectual property is broader than copyright, why has copyright captured almost all of the public and litigious attention so far? The answer lies in where AI’s commercial value currently sits, and in how that value is generated.

Generative AI models are trained on vast quantities of text, images, and other expressive material, much of it alleged to be copyright-protected. The output these models produce is often plausibly similar to, or directly derived from, that training material. This creates an unusually direct and visible link between the technology’s core function and a specific body of law. Copyright was designed to control the copying and reproduction of expression, and AI training and generation both involve copying and reproduction at scale. The result has been a wave of litigation – across the UK, the USA, and elsewhere – testing whether established copyright doctrine, developed for human creators and human-scale copying, can sensibly be applied to machine-scale ingestion and generation.

It is not the first time that a new technology has attracted litigants focusing mainly on one IP right. When search engines emerged, the early intellectual property battles concentrated on keyword advertising and trade mark law, before broader issues of liability and fair use came into focus. With AI, copyright has played that opening.

But practitioners should not mistake the opening act for the whole performance. The patent, trade secret, and trade mark questions raised above are no less significant – they are simply earlier in their litigation life cycle, with fewer decided cases and less public attention to date. This guide therefore attempts to predict outcomes from first principles rather than from cases of first impression.

Courts, Not Legislators, Are Setting the Pace

Some jurisdictions have started to introduce targeted rules for AI – the EU’s text and data mining exceptions, for instance, or disclosure requirements in certain patent offices for AI-assisted inventions. These are narrow interventions, not a wholesale new body of intellectual property law. That is as it should be. Legal systems do not typically create a separate treatise for each new technology; they apply existing principles and legislate only where a genuine gap appears. AI is no exception.

In the absence of comprehensive tailored legislation, it is courts, not legislators or regulators, that are setting the pace of development. Courts have no choice in the matter: faced with a dispute, a court must decide it, even where the law was not written with the facts before it in mind. Litigants pursuing AI-related claims are often asking judges to apply decades-old laws to genuinely novel facts, and the judges must answer, even where the answer is imperfect or invites legislative correction.

This judicial pace-setting has two important features for practitioners to understand. First, until a dispute reaches the highest appellate courts in a given jurisdiction, key uncertainties remain open. A first-instance decision settles the parties’ dispute but does not necessarily settle the law, and lower-court reasoning cannot be relied upon to predict outcomes elsewhere with confidence. Second, court decisions are inherently fact-bound. A judgment on whether AI training infringes copyright in one set of circumstances does not necessarily answer the same question on different facts, even within the same jurisdiction.

Why Precedent Does Not Travel Cleanly Across Borders

The fact-dependency described above becomes more acute still when a dispute crosses jurisdictions. Given how AI systems are built and deployed, this is now the norm rather than the exception. A model may be trained on servers in one country, by a company headquartered in another, using data scraped from sources located worldwide, and deployed to users everywhere. When a dispute arises, the first legal question is not “what does the law say?” but “which jurisdiction’s law applies, and why?”

This is a conflict of laws question. A court must first decide what kind of legal question it is actually facing: is this a copyright question, a database rights question or a contract question? Only then can the court identify which country’s court should hear the case and which country’s laws are applicable to resolve the matter. Recent UK litigation concerning where AI training activity actually takes place illustrates the stakes vividly; the outcome may turn less on what the law says about copyright infringement than on where, geographically and legally, the relevant copying is deemed to have occurred.

The practical lesson is this: it should not be assumed that because an organisation is based in a particular territory, that such territory’s law governs its AI activities. The location of the relevant offices, servers, training data, and users may all point in different directions – and which one matters depends on the conflict of laws rules of the court that ultimately hears the dispute. A ruling from one jurisdiction, on one set of facts, may simply not transfer to a different jurisdiction facing superficially similar facts.

How to Use This Guide

The chapters that follow apply the themes set out above to a range of major jurisdictions, each addressing the same core questions: how existing intellectual property law treats AI systems, training data, outputs, and AI-assisted invention, and where courts, regulators, and legislators are beginning to mark out new ground. Patent, copyright, trade secret, trade mark, design, and related rights are all in scope, reflecting the breadth of intellectual property’s engagement with AI rather than the narrower copyright-only frame.

Readers should approach what follows with two principles in mind. First, treat each jurisdiction’s analysis as exactly that – an account of that jurisdiction’s law, on its own facts, which may not transfer cleanly elsewhere. Second, expect change. Courts are deciding cases now that will narrow today’s uncertainties, but new ones will open as the technology and its commercial deployment continue to evolve. This guide is a snapshot of a fast-moving subject, offered as a reliable starting point for analysis rather than a final word.