AI and copyright
The rise of AI affects copyright in two ways. AI models are trained on existing work, often without the authors knowing that their texts, music or images are being used for it. At the same time those models generate new material, which raises the question of who ends up holding the rights in the result. For authors, producers and publishers these are not theoretical questions, because they bear directly on the value of their work.
Liaise advises authors and creative businesses on the point where AI and copyright meet. This page sets out the legal position on training AI on protected work, on the reservation of rights you can make as an author, and on the copyright status of what AI produces.
Do you need legal advice straight away? Call 020 675 88 21.
May an AI model train on work protected by copyright?
To train an AI model, providers need large quantities of material. They often collect it automatically from websites, a method known as scraping. Much of that material, texts, photographs, illustrations and music, is protected by copyright. Anyone who copies protected work without the rightholder’s permission acts unlawfully in principle.
Even so, that training is not simply prohibited. The legal framework for it is the text and data mining exception in the DSM Directive, which has also been incorporated into the Dutch Copyright Act. Machine learning is regarded as a form of text and data mining. The copying required for the training therefore does not amount to infringement, provided two conditions are met. The party carrying out the data mining must have lawful access to the work, and the rightholder must not have reserved its rights.
Reservation of rights: shielding your work from AI training
This is where the most important lever lies for authors. The text and data mining exception applies only for as long as you, as rightholder, have not expressly reserved your rights. Where you do reserve them, an opt-out in practical terms, your work may not be used for commercial data mining, and therefore for AI training, without your permission.
For work online, that reservation of rights has to be expressed in a machine-readable way, so that automated systems can recognise it. In practice that means a technical instruction attached to your website or files. We advise you on how to record the reservation properly, so that it holds up and your position is strong if you later want to act against unauthorised use.
Transparency about training data
A recurring problem with AI and copyright is proof. Even where you have reserved your rights, it is hard to establish that your work in particular ended up in the training data. The AI Act changes this. Providers of large AI models have to make available a summary of the material they trained on. That gives rightholders more insight into which data was used, which will make enforcing copyright easier over time. This transparency obligation forms part of the phased introduction of the AI Act.
Does copyright subsist in work made with AI?
For the output of an AI system the ordinary starting point of copyright applies: protection exists only where the work is the result of free and creative choices bearing the personal stamp of a human being. That has consequences for various situations.
Work generated wholly autonomously by an AI system, without creative direction by a human, does not in principle qualify for copyright protection. There is then no human author making the decisive choices. Where, by contrast, you use AI as a tool and make the creative choices yourself, by deliberately steering, selecting and editing the result, copyright can indeed arise in you. The greater and more decisive your own creative input, the stronger your position.
Are prompts protected by copyright?
A separate question is whether the prompt itself, the instruction you give an AI system, can be protected. An extensive, detailed prompt containing clear creative choices can in some circumstances be regarded as a work. Whether it is depends on whether it contains sufficient free and creative choices bearing a personal stamp. Where AI results are worked on further, editing a generated image for instance, the end result is more readily accepted as deserving protection. This is a subject still very much in development.
How Liaise helps you
The questions surrounding AI and copyright are new and the case law is still taking shape. That is precisely why it matters to secure your position rather than wait. We help you make a valid reservation of rights against AI training, assess whether AI output infringes your work, record your rights in work you make with the help of AI, and take action where your work has been used without permission. We do this with knowledge of the creative sector, so that the advice fits the way you work.
Do you have a question about AI and your copyright? Get in contact and you will have an answer within one working day.