European Union approach

The policy discussion on the protection of AI‑generated outputs by copyright gained momentum especially following the European Commission's launch of a call for evidence on the review of European Union (EU) copyright rules in May 2026. The initiative sought to gather feedback on possible targeted measures to modernise the Union copyright framework, as well as on the practical impact and effectiveness of the rules introduced by Directive EU 2019/790 on copyright in the digital single market (the CDSM Directive). The consultation indicated that the Commission is exploring whether targeted measures are needed to regulate potential gaps in the intersection of copyright and AI, including the future of AI‑generated outputs. In September 2026, the Commission launched a second, targeted consultation on a potential initiative aimed at a better copyright environment, inviting all relevant stakeholders to share their views on the use of copyright‑protected content in the context of AI. The second consultation reaffirmed the Commission's intention to assess whether a need exists for potential future measures in the area of copyright and generative AI.

The CDSM Directive predates AI developments, and, therefore, does not provide for a specific regime on the topic. As Regulation (EU) 2024/1689 on artificial intelligence (the AI Act) was modelled on product safety law, it remains silent on whether AI‑generated outputs should benefit from copyright protection, despite the fact it was adopted at a time when AI models' capabilities and the impact of their outputs were already well‑known. However, it prescribes that providers of general‑purpose AI models should put in place a copyright policy allowing identification and compliance with the reservation of rights under Article 4(3) of the CDSMD, as well as the drawing up and publication of a detailed summary of the content used for the models' training, pursuant to Article 53(1)(c) and (d) of the AI Act.

Despite the lack of specific regulations in EU law on whether AI‑generated content is protected by copyright or related rights, such protection is generally understood¹ to be available only for works created by humans. The Court of Justice of the European Union (CJEU) has consistently held in its case law which works can enjoy copyright protection. Over time, it has confirmed the anthropocentric approach governing copyright law by highlighting that human authorship lies at the heart of copyright protection. An original work reflects the author's own intellectual creation, whereby an author has made free and creative choices.²

Therefore, if the production of content has been mainly determined by technical considerations, leaving no room for creativity and artistic freedom, then the outcome will most likely not meet the originality threshold required to enjoy copyright protection.³

Conversely, if a human has used AI as a tool for the production of content (widely known as AI‑assisted works), similar to how cameras for taking photos are used, having made substantive creative choices, the final output may be subject to copyright protection. The human's overall involvement may be assessed by considering the degree of creative control exercised, whether the output reflects their personal creative choices, and whether the contribution consists of selecting, arranging, or substantially transforming AI‑generated material.

Under the current understanding of EU copyright law, AI‑generated outputs produced without any human involvement or prompt‑based outputs with little to no human refinement and modification are unlikely to fall within the scope of the copyright protection in the EU, as noted in the European Parliament's 2025 study on generative AI and copyright. More elaborate prompting, selection, modification or other human intervention may, depending on the circumstances, help demonstrate the author's free and creative choices.The need to ensure the protection of human creativity and exclude content fully generated by AI from copyright protection has been repeatedly highlighted by the European Parliament in its resolutions.

The Commission had also highlighted this in a 2020 report, Trends and developments in AI – Challenges to the intellectual property rights framework, where it explicitly noted that 'the requirement of human intellectual effort excludes from copyright protection outputs that are produced without any human intervention' and that 'a production that is wholly generated by an AI system without any human intellectual effort is excluded from copyright protection'. However, as underlined in the same report, this requirement does not exclude creations by human authors made with the aid of machines, 'provided that the human contribution to the output meets the legal standard of originality/creativity'.

Assessing whether an AI‑generated output involves substantial human intervention and creativity to qualify as original work and, therefore, meet the criteria for copyright protection is neither a simple nor a straightforward task. What has crystallised so far, according to the 2025 Parliament's study, is that a mere prompt containing a general instruction, which leads to an accepted output with no further modifications, will not be subject to copyright protection. By contrast, where the user has taken actions that substantially shaped the outcome, through edits, specific instructions and modifications, the result might qualify for protection by copyright or related rights. Ultimately, it is up to the courts and to policymakers to determine which degree of human involvement suffices for the final output to enjoy copyright protection. At present, this assessment requires a case‑by‑case examination. The existing EU copyright framework and the CJEU case law are at present considered adequate to provide the necessary indicators for this assessment.

In addition to copyright protection, it has been left open whether a sui generis right could also be explored as a mechanism to protect AI‑generated outputs, as currently happens in Ukraine with regard to non‑original content generated by computer programs.

Approaches outside the EU

AI‑generated works' presence on marketplaces

Apart from considerations regarding output protection, there is growing evidence⁴ that the large‑scale availability of AI‑generated productions has substantially increased competition in the cultural and creative field, as more AI‑generated productions enter the market and operate as a substitute for the original works. In essence, AI‑generated content increasingly competes with human‑created works for consumers' and users' attention, visibility, demand and revenues.

As reported, an increasing number of AI‑generated books – often without any disclosure of their nature – have flooded the online marketplaces because of their extremely low production costs. A study examining over 14 000 self‑published books sold on Amazon between 2023 and 2026 notes that books containing substantial amounts of AI‑generated text, defined by the study as more than 25 % of the book, make up a large share of the catalogues. A similar trend can be observed in academic and other written publications. Within the first 12 months following the launch of ChatGPT, the share of primarily AI‑generated articles increased dramatically, reaching 36 % and 48 % by 12 and 24 months, respectively, as reported.

The music sector has seen a comparable development. An analysis of over 1 000 000 music tracks, conducted in July 2026, found that 23.2 % were fully AI‑generated. Data reported by a global music‑upload platform showed a notable increase in AI‑generated music, with AI‑generated tracks accounting for 50 % of all tracks uploaded to the platform in June 2026. Against this backdrop, the Australian Recording Industry Association decided to exclude entirely or predominantly AI‑generated music from Australia's official music charts, requiring all musical works included there to be substantially human‑created. A similar approach has been followed by SNEP, the French Union of Phonographic Publishing, which has announced that purely AI‑generated tracks will be excluded from its official rankings. SNEP further set three conditions that need to be fulfilled for a work to be included in its charts: substantial human intervention; clear indication of the use of AI; and the AI tool used for the production of the track(s) to respect the legal framework applicable in France.

Competition with human‑created works

As noted by the USCO in its 2025 report, the increasing presence of AI‑generated content in digital markets may affect the visibility and revenues of human‑created works, although the extent and distribution of these effects are still emerging. The economic study of copyright and AI, carried out at the request of the European Parliament's Committee on Legal Affairs (JURI), has examined, among other issues, the potential economic implications of generative AI for creators and rights‑holders, including matters relating to the traffic to original content and associated revenues.

A recent European Commission study has also examined challenges relating to the discoverability of European cultural content in the digital environment, with a particular focus on music and books. Its findings point to the growing role of AI‑generated content and algorithmic recommendation systems among the factors that may affect the visibility and discoverability of European cultural works online. The study suggests further examining the discoverability of cultural content in the online environment, including the effects of algorithmic bias, filter bubbles⁵ and the growing use of AI, with a view to informing possible policy responses.

As users may be drawn to AI‑generated content without being aware of its nature, the labelling of such content as AI‑produced may influence their choices regarding the type of content they consume online. The AI Act establishes specific rules in Article 50 on the labelling and marking of AI‑generated output to ensure that natural persons are informed when they interact with such content, with the aim of reducing phenomena of manipulation and deception. The provision, which started applying from 2 August 2026, is further specified by a code of practice and European Commission guidelines.

Impact on revenues and suggested options

Visibility leads to more revenues for the workers in the cultural and creative sector. In the EU, authors, artists and other rights‑holders should receive appropriate and proportionate remuneration for the use of their works, according to the CDSM Directive. Therefore, for remuneration to be due, there should be a copyrighted work that – as explained above – is linked to a human author. Autonomously AI‑generated productions, with no human involvement, cannot be subject to remuneration under Union copyright law.

Since purely AI‑generated productions generally do not benefit from copyright protection, they can neither be licensed nor generate remuneration through copyright law. For this reason, users may simply use unprotected AI‑generated content, rather than copyright‑protected content, in an effort to avoid payment for the use of copyright‑protected material.

However, the absence of copyright protection in the output does not necessarily mean that the output is unrelated to copyright‑protected works. AI‑generated outputs may reproduce, incorporate, or otherwise reflect elements of protected works that have been used as inputs in the AI system. This raises the question of whether, and under what conditions, the use and influence of such protected works in the generation of an output should give rise to remuneration for authors and rights‑holders.

In light of this, some European author societies have proposed the introduction of a remuneration system for authors and rights‑holders when their works have been used and have influenced, or been embedded in, the final AI‑generated output. However, it remains to be further specified which factors should be considered for such influence to be established (such as the resemblance of the output to the original works or other factors).

The German collecting society for music, GEMA, has presented its licensing model for AI that aims to cover both the use of music to train AI systems and the content those systems later generate based on the original works. As the organisation notes, such a system aims to provide fair, high‑level remuneration while staying flexible enough to adapt to fast changes in the market and technology.

Several US academics suggest that a copyright holder contractually agrees with the AI model owners that part of their work will be incorporated in the AI‑generated output, which will then be commercially exploited, making them participants in all the revenues generated by the AI model. In this scenario, the author or rights‑holder will not merely participate in the royalties collected by a concrete output influenced by their work but also in all the revenues the AI company has gained.

A different study recommends the launch of a 'data intermediary' who will negotiate with AI companies on behalf of the creators and distribute royalties to them. Under this model, payments will take place according to the AI company's revenues, proportional to each creator's contributions to social surplus.

Similarly, another report suggests a framework that distributes royalties to those copyright holders whose works were used to train AI systems, based on how likely the output is to meet users' needs. Royalties will then be distributed to copyright owners depending on their contributions. The criterion chosen in this proposal is highly subjective, as it is linked both to how accurately the output responds to the prompt and to the users' satisfaction, which may raise several issues on how this can be neutrally assessed.

It has also been suggested that, rather than using the AI training process as the basis for calculating remuneration owed to authors, the legislator should consider the introduction of an AI levy. Under this proposal, AI systems capable of generating content similar to protected works would be subject to a lump‑sum levy, which would then be distributed among authors and rights‑holders.