The Protection of AI Prompts in EU Copyright Law and its Implications for the Public Domain

In recent years, the global emergence of generative artificial intelligence (‘AI’) tools has given rise to a number of unprecedented issues in copyright law, some of which have only started to be addressed in the European Union (‘EU’). Examples of these issues include whether the use of copyright-protected materials as ‘training data’ for AI models may amount to copyright infringement or may fall under the ‘text and data mining exception’ in the DSM Directive—a matter that is addressed in the recently enacted Regulation (EU) 2024/1689—and whether AI-generated outputs that lack a human author may qualify for copyright protection. However, an issue that has received little coverage to date is whether the instructions (or ‘prompts’) that a user gives to an AI model to generate an output (such as text or an image) are themselves capable of being protected by copyright in EU law.

This presentation contends that this issue is deserving of greater attention as it raises several challenges for EU copyright law. These challenges concern, inter alia, how far (if at all) the copyright protection of AI prompts might be impacted by the relationship (or connection) between a prompt and its output (such as the extent of control that the user’s prompt exerts over the output) and whether copyright in a prompt might prevent someone from copying the output, irrespective of whether the output itself is or is not capable of receiving copyright protection. For example, might the author of a (copyright-protected) prompt successfully claim that direct copying of the output amounts to indirect reproduction (and thus infringement) of their prompt? Or might we characterise a prompt merely as the ‘instructions’ for its output, like the recipe for a cake, with the result that copying the output (or the cake) would not indirectly reproduce the ‘intellectual creation’ of the prompt (or the recipe) and thus would not constitute infringement?

As of yet, however, the issue of whether prompts themselves are capable of being protected by copyright has only been touched upon in the Czech Republic, appearing in obiter remarks in a 2023 decision by the Municipal Court in Prague and only in the context of the Czech Copyright Act. In this case, the claimant alleged that the defendant had infringed his copyright in an AI-generated image by publishing the image online without authorisation. Although the Court’s decision centred on the ‘authorship’ requirement, what is significant about this case is the Court’s obiter suggestion that a prompt would not be a ‘work’ under Czech law because, applying the idea-expression dichotomy, a prompt is the ‘subject’ of the work or ‘possibly an idea’ and thus incapable of being protected by copyright.

Drawing upon this case and recent developments in copyright law, both within and outside of the EU, concerning generative AI, this presentation examines two core issues. These are the following: first, whether prompts are capable of being protected by copyright in EU law, focusing in particular on the requirements for protection under the Software Directive and the Information Society Directive; and second, what the implications of such protection would be for the public domain, particularly in the light of what appears to be an emerging market for ‘engineering’ prompts that generate a desired output.