Generative AI systems now produce commercial quantities of text, images, code, designs, and music. Businesses deploy these outputs in marketing, product development, software engineering, and creative production at scale. Yet the intellectual property status of AI-generated content remains genuinely uncertain across most European jurisdictions. Whether AI-generated works attract copyright protection, who owns them if they do, and what happens when they infringe third-party rights are questions that courts and legislators are only beginning to resolve. For any business that uses generative AI to create commercially valuable output, this uncertainty carries real legal risk.
The Human Authorship Requirement in EU Copyright Law
EU copyright law, harmonised through a series of directives including the Copyright in the Digital Single Market Directive (DSM Directive), the Software Directive, and the Database Directive, does not explicitly address AI-generated works. The CJEU has, however, established through its case law that copyright under EU law requires that the work reflects the author’s own intellectual creation — a personal, human creative choice. In Infopaq (2009), Painer (2011), Football Dataco (2012), and subsequent decisions, the Court has consistently held that the originality threshold for EU copyright protection is met when the author makes free and creative choices that stamp the work with their personal touch.
The implication for AI-generated content is significant. Where a generative AI system produces output without meaningful human creative input — for instance, where a user enters a brief functional prompt and the AI generates a complete article, image, or code module — there may be no human author who made the relevant creative choices, and therefore no copyright under EU law. The output falls into the public domain from the moment of creation, available for anyone to use without restriction or payment.
This conclusion is not universally accepted, and the line between insufficient and sufficient human creative input is not clearly drawn. A user who writes a detailed, specific prompt describing the visual composition, colour palette, stylistic references, and thematic content of an image they want generated is making creative choices that go into the output, even if the execution is mechanical. The question is whether those prompt-level creative choices are sufficient to constitute authorship of the resulting work. No European court has definitively answered this question in the context of modern large language model or diffusion model outputs.
The UK’s Sui Generis Approach and Why It Is Not EU Law
The United Kingdom’s approach to computer-generated works is a useful comparative reference, though it is EU law that governs for EU-based businesses. Section 9(3) of the UK Copyright, Designs and Patents Act 1988 provides that for a computer-generated work — one with no human author — the author is deemed to be the person who made the necessary arrangements for the creation of the work. This provision was inserted in 1988 when computer-generated works meant something quite different from modern generative AI outputs, but it provides a mechanism for allocating copyright in works produced without a human author in the traditional sense.
The UK Intellectual Property Office has consulted on whether this provision remains fit for purpose in the context of generative AI and published a consultation response in 2023 indicating that the existing framework should be retained for now, though further review was anticipated. The UK’s position is therefore that AI-generated works can attract copyright where a human made the necessary arrangements, with the protection term being fifty years rather than the standard life plus seventy years.
EU law has no equivalent provision. The DSM Directive and its predecessors were drafted on the assumption that authors are human and do not contemplate the scenario of a work created by an AI without a human author. Several European commentators have argued for a neighbouring rights approach — akin to the sui generis database right — that would protect AI-generated works without characterising them as authored works, but no such proposal has been formally advanced by the Commission to date.
Training Data and the Copyright Infringement Question
The copyright status of AI-generated output is only one dimension of the IP problem. Equally important — and currently more actively litigated — is the question of whether training a generative AI model on copyrighted content without the rights holder’s consent constitutes copyright infringement. This question is being contested in multiple jurisdictions simultaneously, with cases involving major news publishers, visual artists, music rights holders, and software developers against AI model developers including OpenAI, Stability AI, Midjourney, and GitHub Copilot’s operator Microsoft.
In Europe, the relevant legal framework is Article 4 of the DSM Directive on text and data mining. Article 4 creates a mandatory exception allowing text and data mining of lawfully accessed works for any purpose — including commercial AI training — subject to the rights holder’s right to opt out by appropriate means such as machine-readable reservations. Article 3 provides a narrower mandatory exception for text and data mining for scientific research purposes that cannot be overridden by opt-out.
The practical significance of the Article 4 opt-out right is considerable. Rights holders who have implemented opt-out mechanisms — through robots.txt files, licensing terms, or other technical means — may be able to argue that training on their content after a valid opt-out was implemented constitutes infringement. The legal validity and technical effectiveness of different opt-out mechanisms has not yet been tested in European courts, and the obligation on AI developers to respect opt-out signals and to maintain records of training data provenance sufficient to verify respect for opt-outs is an area of active regulatory attention under the EU AI Act’s transparency requirements for GPAI models.
AI-Generated Inventions and Patent Law
Patent law adds a further dimension to the AI-generated IP question. European patent law, under the European Patent Convention administered by the EPO, requires that a patent application name the inventor, and the inventor must be a natural person — a human being. The EPO has consistently rejected patent applications that designate an AI system (specifically the DABUS AI system developed by Stephen Thaler) as the inventor, on the ground that no natural person is named as inventor. The UK Supreme Court reached the same conclusion in 2023 in the UK litigation concerning the same patent applications.
The consequence is that inventions generated autonomously by AI systems — without a human inventor who made the inventive step — cannot be patented under current European patent law. They fall into the public domain unless the business can identify a human whose contribution to the AI’s training, architecture, or deployment constitutes the inventive contribution. For pharmaceutical companies, semiconductor manufacturers, and other R&D-intensive businesses that are beginning to use AI to accelerate invention discovery, this gap in patent protection for AI-generated inventions represents a genuine strategic risk that has not yet been resolved by either litigation or legislation.
Practical Risk Management for Businesses
Given the uncertainty, businesses deploying generative AI for commercially valuable output should consider several practical risk management approaches. First, they should document human creative involvement in AI-assisted workflows as thoroughly as possible: who specified the prompt, what iterative choices were made in refining the output, what human editing or selection occurred among multiple AI-generated alternatives. This documentation supports a claim to copyright protection by establishing the human creative choices that went into the output.
Second, businesses should review their agreements with AI service providers regarding IP ownership in outputs. Some providers’ terms of service purport to assign output IP to the customer; others disclaim any warranty of non-infringement and leave the customer with both the output and the legal risk. The contractual allocation of IP risk between AI provider and customer is an important dimension of vendor selection and procurement for any business producing AI-assisted commercial content.
Third, businesses should assess whether their AI deployments touch on content categories where third-party IP infringement risk is elevated. AI systems trained on large corpora of images, text, or code can produce outputs that reproduce elements of their training data in ways that may constitute copyright infringement — particularly where the training data included highly distinctive or recognisable works. For brand-sensitive applications such as advertising, product design, or software development, the risk that AI-generated output infringes third-party copyright or trade marks warrants specific diligence and possibly contractual indemnity from the AI provider.
The Policy Trajectory
The EU AI Act’s provisions on transparency for GPAI models — requiring disclosure of training data and compliance with EU copyright law — signal that the Commission intends to address the training data copyright question through regulation rather than waiting for the courts to resolve it through litigation. ESMA, the EPO, and the European Commission have all published consultation documents or work programmes indicating that AI-generated IP is a priority policy area for the coming years.
The most likely legislative trajectory is a clarification or extension of the DSM Directive to address both the training data question and the status of AI-generated outputs, potentially combined with a sui generis protection right for AI-generated works comparable to the existing database right. Whether such a framework would be adequate to address the full complexity of AI-generated IP — which spans copyright, patents, trade secrets, and database rights — is a question that will occupy European IP lawyers and policymakers for years to come.
Conclusion
The ownership and protection of AI-generated intellectual property is one of the most consequential unresolved questions in European business law. The existing framework was designed for human creators and does not map cleanly onto AI-generated works. Businesses that ignore this uncertainty and treat AI-generated output as straightforwardly owned and protected are taking legal risks that may materialise in enforcement actions, licensing disputes, or invalidation of registered rights. A thoughtful approach — documenting human involvement, reviewing vendor agreements, assessing infringement risk, and monitoring the legislative and judicial developments — is the foundation of defensible AI-assisted IP practice.