Copyright information around generative AI

Introduction

Generative AI (GenAI) tools can generate large amounts of text or images in an instant. But from the point of view of copyright law, there are important consequences you should take into consideration when using GenAI.

Who has copyright on AI-generated texts?

The unchanged text output of generative AI (GenAI) systems, such as ChatGPT, is not protected by copyright. These unchanged outputs do not have (enough of) a creative human element in their creation. Therefore, neither the model’s creator nor the prompt’s writer can claim copyright on the output. This also means that anyone you share this output with is free to use, copy, modify and share it with others.

If you substantially alter the AI model’s output by adding creative human input, this final work can be subject to copyright. For example, if you use the text output as inspiration for a publication, that publication will be protected by copyright.

Note: Be careful what kind of information you input into GenAI systems, as these prompts can be reused by the tool for other purposes. For this reason, do not input personal data or company-sensitive information.

Can an AI-generated text infringe on copyright?

Yes, a text-based generative AI (GenAI) may produce output that violates copyright by being too similar to an existing work. If you use any text from a GenAI system in your work, you, as the author, are responsible for ensuring that it does not violate any confidentiality, licencing or copyright laws.

How should I cite the usage of text-generating AI systems?

Publishers and journals have different rules for using generative AI (GenAI) text systems. Some do not allow GenAI to be used in their publications, while others require clear disclosure of use. Always check what is allowed or required for a given publication. However, the following recommendations are a good baseline to use when publishing an article for which GenAI has been used:

• When AI is used to edit or to adjust a text, use of the GenAI text system must be disclosed in an authorship statement or CRediT statement.
• A GenAI system cannot be the (co-)author of a publication, as it cannot take on all the responsibilities that come with being the author of a work.
• When used as a data analysis tool, AI is considered part of the methods and should be discussed in that part of the publication. 
• GenAI is, by definition, not a primary source of information, so AI-generated text usually cannot be cited as a source of information.
• The authors of a publication are always responsible for the publication’s accuracy, completeness and coherence.

Who has copyright on AI-generated images?

As with text, the unchanged output of GenAI systems, such as Dall-E, is not protected by copyright, as they do not have any (or enough) creative human element in their creation. This means neither the system’s creator nor the prompt’s writer can claim copyright on the image. This also means that anyone you share this image with is free to use, copy, modify and share it with others. 

Note: Be careful what kind of information you input into GenAI systems, as these prompts can usually be reused by the tool for other purposes. For this reason, do not input personal data or company-sensitive information.

Can an AI-generated image infringe on copyright?

Yes, generative AI (GenAI) may produce images that infringe on copyright by being too similar to an existing work. This is particularly relevant for images, as the GenAI models for images can easily create similar materials.

As it is virtually impossible to guarantee that a GenAI’s image output doesn’t infringe on other work, we strongly discourage using GenAI-generated images in any official capacity.

What data can I use to train an AI system at the WUR?

As a research institute and university, we are allowed to mine text & data sources even if they are protected by copyright or explicitly prohibited from data mining by commercial institutions. This right does have some restrictions, so always check with the copyright information point or your legal department if these apply to your situation. In addition, the creators of certain AI systems or models are required to show the training data used to train their systems. We recommend that you keep track of the training data you use for your AI system or model, so that the AI system or model meets the legal and scientific prerequisites and can be released publicly in the future.

Do I need to disclose what material was used to train an AI system I created?

Yes, you may be required to share what training data was used to train your AI system under EU law when you release/share your system. Therefore, it is important to keep track of the data used to train your AI system or model.

Questions & support

Do you have questions about this service, or would you like personal support? Feel free to contact us. You can send an email to copyright@wur.nl.