Policy paper #22 against new exclusive rights for AI-generated outputs

Background

Generative artificial intelligence (AI) has seen rapid mainstream adoption. Whether or not the technology is overhyped at the moment, it is likely to have a transformative impact on cultural production and the creative sector simply because of the degree of automation that it can provide. And as cultural and creative production are largely governed by copyright law, unsurprisingly the surge of generative AI has raised a number of questions for copyright law. First on the input side: a heated debate continues to rage over the question of the legality of using copyright-protected works during the training stages.¹ A recurring question in this overall debate is: Can AI-generated outputs be copyright-protected? Or should they be subject to a different level or type of protection than human-authored works, if any? This paper focuses on the output side, arguing that no new exclusive rights should be introduced for AI-generated outputs that fall short of the established standard for qualifying as copyright-protected works.

Copyright has accommodated the emergence of a variety of technologies over the years. In many cases, this has been accomplished through changes to copyright law. In response to technological shifts, lawmakers have expanded the scope of exclusive rights from books and paintings to other subject matter where technology is involved in the creative process, such as motion pictures and sound recordings.

While, at the time of drafting, there are no concrete, specific proposals for legislative action at the EU level regarding the protection of AI-generated outputs, the issue has been raised in a number of contexts² and some stakeholders have called for the introduction of new exclusive rights to protect AI-generated outputs that fall short of the criteria for copyright protectability.³ We reject these claims. The EU copyright framework has all the tools needed to establish whether any given expression created with the help of AI is protected by copyright or any of the related rights that are already in place. More importantly, there is currently no evidence that EU copyright policy needs to respond to this technological change by introducing a new exclusive right and taking outputs that are not protected under the existing framework out of the Public Domain.

Using AI for creative expression

Generative AI allows for the creation of a wide range of relatively sophisticated expressive outputs even from simple prompts, among other use cases. The incorporation of technology, including computer-based technology, in creative expression is nothing new from a copyright perspective. In comparison to previous technological innovations, however, generative AI can be used to automate extensive parts of the creative process. This raises questions for the copyrightability of creative expressions generated by or with the help of AI. But before we can discuss the legal implications, we need to understand the process of generating creative expressions with the AI tools available today and the degree of human involvement in the process.

For analytical purposes, the creative process can be divided into three phases, conception, execution and redaction.⁴ When using an AI system to generate an expression, the human usually comes up with an idea for the expression. The human thus remains in control during the conception phase. They typically cede control to the machine during the execution phase and may reclaim creative control again during the redaction stage. While this may change, the use of generative AI for creative expression today roughly falls into one of three categories:

A common way of using AI to generate expressive outputs that affords little creative control to the human is simple prompting. Here, the user enters a single, relatively unspecific prompt in the interface of an existing AI application and selects a result. While the user may have come up with the idea, the user is not in control over its execution, and the outcome will display a high degree of arbitrariness due to the probabilistic (“black box”) nature of today’s AI models.

Iterative prompting, on the other hand, allows for a higher level of creative control. When a human user iterates and refines their instructions to the AI, editing and polishing the output, this may allow them to achieve results that approach their idea for the expression through redaction.

AI can also serve a more assistive function in the creative process, where it may be understood as simply enhancing human creativity by transforming a human’s creative expression. Some artists are modifying, rearranging, or otherwise transforming AI “raw” outputs to create something different. Others are using it as a tool to “brainstorm” ideas, but without using any of the results directly (e.g. AI-generated song lyrics or a proposal for a chord progression). This is a more measured use of the technology where the AI output is only one piece of the puzzle in the entire expressive work. Generally speaking, the category of assistive uses represents a broader spectrum, ranging from medium to high levels of human control, including during the execution stage.

Protecting human creativity with copyright

The existing EU copyright framework offers sufficient clarity for making determinations about whether any expressive output that is AI-generated or created with the help of an AI is eligible for copyright protection. It effectuates the principle, reflected in the Berne Convention and the EU copyright acquis, that copyrightable authorship must come from natural persons, a principle that ought to be maintained in the face of AI.

For an AI-generated output to qualify as a copyright-protected work under EU law it would need to meet the criteria that are well established in the acquis and case law: it would need to belong in the “literary, scientific or artistic domain”, be created with human intervention and have the human’s creative choices reflected in its expressive features.⁵

It is safe to assume that many AI-generated outputs belong in the “literary, scientific or artistic domain.” It is also clear that the current generation of generative AI is incapable of fully autonomous creation. All AI-generated outputs, for now, involve some human input, although they display considerable variation as to the extent of human control over the creative process depending on how an AI is used. Thus, when assessing the copyrightability of any AI-assisted output, courts will mostly focus on the originality of AI-generated outputs,⁶ bearing in mind two closely related questions: Is the human author’s original idea sufficiently expressed in the final outcome and was that result achieved through free and creative choices made “in several ways and at various points in [the] production” of a work?⁷

This assessment is far from simple and will need to be determined on a case-by-case basis. Again, the transformation of an idea to an expression may be fully automated. Yet in many cases, the creative process as a whole is iterative. Where a creator refines their instructions to the AI through iterative prompting, a court would need to consider whether this process of redaction allows them to achieve a result that approximates their creative vision and can be considered “their own.” The human needs to make subjective calls over the work (substantial and sufficient creative input) and exercise direct, creative control over its production. Economic investment, intellectual effort, or a requirement of skill alone do not suffice to satisfy the originality requirement.

In sum, there is no single answer to the question of who is in control over the AI-generated expression that encompasses all possible use cases. While some outputs will qualify as works and be eligible for copyright protection, many others will not due to the lack of human control during the execution and redaction phase of the creative process. In some cases still, the applicability of copyright might be limited to certain elements of the work: original elements (e.g. text) contributed by the artist, selection and arrangement of those elements and the AI-generated outputs together, as well as any otherwise unprotectable material that has been edited, modified, or otherwise revised by a human author and consequently presents a sufficient level of originality.

What is clear is that only natural persons, i.e. human beings, can be authors, which is reflected in the Berne Convention and the EU copyright acquis.⁸ This precludes any possibility of AI authorship or AI co-authorship, when the output qualifies for copyright protection.⁹ Neither of the two predominant philosophies of copyright would justify an extension of copyright to non-human authors.¹⁰ Generative AI is not creative in the human sense. It creates new data based on its training data and input. By extension, AI model developers are ineligible as authors for copyright in any content generated by users of the models, as they neither imprint their personal “stamp,” nor do they have sufficient control over the generation of the outputs.

Protecting non-original AI-generated outputs with exclusive rights?

This raises some follow-up questions: Are AI-assisted expressions that do not meet the standards for copyright protection protected by existing related rights or sui generis rights? And is there a need to introduce a new exclusive right?

Most related rights have no threshold for protection. This means that individuals and companies generating non-original expressions with the help of AI may benefit from strong protection from related rights in a variety of sectors already. For example, a user prompting an AI to generate an audio or video output could potentially benefit from the phonogram right or the film producer’s right. Likewise, if a broadcast is automatically produced and transmitted by an AI system, this output might qualify for protection under the related rights of broadcasting organisations. An AI-generated publication published by an European press publisher could arguably fulfil the criteria for protection under the press publisher’s right. The list goes on.¹¹

This is certainly problematic, as the rationale for granting related right protection is to reward effort and protect investments in the production or dissemination of protected works. The fact that non-original AI outputs may qualify for such protection regardless of any underlying creative activity—and often in the absence of meaningful investment—calls into question whether the related rights system is still fit for purpose in the age of AI.

For this reason, we also oppose the introduction of a new exclusive right for non-original AI-generated outputs at this point. The introduction of exclusive rights akin to copyright has a lasting curtailing effect on the Public Domain and should always be conditioned on strict criteria based on the principle of proportionality and clear economic evidence.¹² In this case such a strong evidentiary basis simply does not exist. There is no evidence of a need for additional incentives for using AI technology in the creative process. On the contrary, we should consider the potentially harmful effect on human creators, as the overproduction of AI already drowns out human-generated content from certain platforms.

Neither is there a need for creating new forms of exclusivity for AI model developers. The existing legal framework provides sufficient incentives to innovate, including patents, trade-secret protection, unfair competition, and software copyright protection. There does not appear to be any shortage of incentives for investments into generative AI and it is unlikely that this already thriving market is in any meaningful way driven by the hope of acquiring exclusive rights for the outputs generated by AI one day.

The case for registration

A number of problems remain. In many cases, the human creator does not have an interest in the commercial exploitation of AI-generated content, even if it is protected by copyright or related rights. It is also nearly impossible for an individual user to ascertain whether an AI-generated output is copyright-protected or not. This places a disproportionate burden on users of these outputs, first to determine their copyright status and, should the output be considered protected, to identify and locate the right holder (who might not even be interested in commercially exploiting their works). Making copyright protection conditional upon registration would constitute a more appropriate solution for users, particularly if this information were to be stored as part of the metadata and/or in a public database.

The conditions for copyrightability, including originality, would still apply, and a registration would only constitute a declaration of interest in commercial exploitation. Like in the US, a European copyright office could still reject a registration or revoke protection if an investigation of the facts reveals that these criteria are not met.

COMMUNIA has long advocated the introduction of a registration requirement for all kinds of works to avoid overprotection if a creator has no interest in commercial exploitation.¹³ We argue that unregistered works should receive only moral rights protection (which would not be applicable in the case of non-original AI-generated outputs). Any legal mechanism intended to strengthen copyright protection beyond what is required in binding international treaties (such as additional terms of protection, additional rights or additional enforcement mechanisms) should be made conditional upon registration.

Conclusions and Recommendations

The European copyright framework already has all the tools needed to deal with AI-generated outputs. The output may be protected by copyright provided it meets the bar for copyright protection. When an AI-generated output falls short of the bar for copyright protection, it may still benefit from protection under related rights. If no rights apply, it belongs in the Public Domain, free for all to reuse.

There is no need for a new exclusive right for outputs that fall short of the standards for copyrightability. No additional incentives are needed to promote the use of generative AI. A new related or sui generis right would not provide for greater legal clarity either since the copyright status of any output would still need to be determined. Instead, the European legislator should consider the introduction of a registration requirement to make it easier for users to understand whether a creator has an interest in the commercial exploitation of their AI-generated output or not.

RECOMMENDATION 1

No new exclusive rights

We call on the EU to refrain from introducing any new related or sui generis right for AI-generated content that falls short of the standards for copyright protection, and from creating new forms of exclusivity for AI model developers.

RECOMMENDATION 2

Full copyright protection should only be granted to works that have been registered by their authors

We renew our call for a copyright system based on registration where full copyright protection is only available to works registered by their authors to prevent unnecessary and unwanted protection. Limiting the registration requirement to AI-generated outputs that qualify as copyright-protected works would raise many fundamental and practical concerns. Thus, formalities should apply to all categories of expressions that are protected by copyright.

Endnotes

1. We addressed this question in the European context in our policy paper #15 on teaching the machine published in April 2023.
2. In 2024, the Hungarian Council Presidency circulated a questionnaire to the Working Party on Intellectual Property (Copyright), section III of which covered “the protection of outputs.” The tenth session of the WIPO Conversation on Intellectual Property and Frontier Technologies held in 2024 focused on the topic of “generative AI-IP and outputs.” Most recently, the US Copyright Office issued a report on the copyrightability of AI-generated outputs under US copyright law, opposing both the expansion of copyright law and the creation of a new sui generis right: see United States Copyright Office (2025), Copyright and Artificial Intelligence: Part 2: Copyrightability, at p. iii.
3. For an overview of arguments brought forward in favor of new exclusive rights for AI-generated outputs, see pp. 32 ff. of Copyright and Artificial Intelligence: Part 2: Copyrightability.
4. See P. Bernt Hugenholtz, João Pedro Quintais, and Daniel Gervais (2020), Trends and Developments in Artificial Intelligence: Challenges to Copyright. See also chapter 3 of European Commission: Directorate-General for Communications Networks, Content and Technology et al. (2020), Trends and developments in artificial intelligence – Challenges to the intellectual property rights framework – Final report, Publications Office of the European Union.
5. For an overview of the relevant EU case law, see P. Bernt Hugenholtz and João Pedro Quintais (2021), Copyright and Artificial Creation: Does EU Copyright law Protect AI-Assisted Output?, IIC – International Review of Intellectual Property and Competition Law, 52, pp. 1190–1216, at pp. 1201 ff.
6. For any work, including potentially AI-generated output, to qualify for copyright protection, it must be original, i.e. the “author’s own intellectual creation” and “reflecting his personality in that subject matter. See e.g. CJEU Case C-469/17 – Funke Medien, para. 20 and CJEU Case C-833/18 – Brompton Bicycle, para 26.
7. CJEU Case C-145/10 – Painer, para. 90.  For an in-depth discussion, see P. Bernt Hugenholtz (2024), Copyright and the Expression Engine: Idea and Expression in AI-Assisted Creations.
8. Johannes Fritz (2024), The Notion of ‘Authorship’ under EU Law — Who Can Be an Author and What Makes One an Author? An Analysis of the Legislative Framework and Case Law, Journal of Intellectual Property Law & Practice, 19(7), pp. 552–556.
9. Carys J. Craig and Ian R. Kerr (2021), The Death of the AI Author, Ottawa Law Review, 52(1), pp. 31-86.
10. Jozefien Vanherpe (2025), Artificial Intelligence and Intellectual Property Law, in: Nathalie A. Smuha (ed.), The Cambridge Handbook of the Law, Ethics and Policy of Artificial Intelligence, Cambridge University Press, pp. 211-227, at p. 217.
11. For an in-depth discussion, see European Commission: Directorate-General for Communications Networks, Content and Technology et al. (2020), Trends and developments in artificial intelligence – Challenges to the intellectual property rights framework – Final report, pp. 89-94.
12. See COMMUNIA policy recommendation #4: Legislative power to reduce the Public Domain should be constrained.
13. COMMUNIA policy recommendation #2: Full copyright protection should only be granted to works that have been registered by their authors.