Music streaming debates series part 2: streaming and GenAI discussions in canon

Part 1 of this series gave a general overview of the copyright-related discussions regarding streaming services from the last year. In Part 2, we will gain a clearer picture of the expected challenges for fair remuneration and control over one’s artistry created by new GenAI music services. Also, the implications for “good old” streaming services will be examined. Some concrete legal solutions will be proposed, while also highlighting uncertainties that remain.

Discussions in canon

A new wave of questions has arrived, revolving around GenAI music services. But many debates as to how fair remuneration should look on (by now) traditional streaming services have not yet been resolved either. The question of what should be deemed “sufficient” remuneration for music creators is central to a Dutch ruling on streaming royalty rates from 2025. Or rulings, rather, as three musicians separately orchestrated very similar litigation against their mutual record label, Universal Music Group (“UMG”). The artists claimed that UMG’s rates for digital exploitation of music recordings, streams and downloads, are too low. As argued, UMG bases these rates on old, pre-streaming era contracts, when there was more justification for costs being deducted from the artists’ share because of the labels’ task of physical distribution (vinyl, CDs etc.). Among other things, the artists sought recalculation by UMG and for the record company to pay at least 50% (instead of 20%) of the income from digital exploitation to the artists.

The outcome in all three cases is essentially the same: a deal is a deal. According to the court, the artists did not sufficiently substantiate that 50/50 is a standard revenue split for digital exploitation. Additionally, a national fair remuneration rule relied upon by the artists was only introduced in Dutch legislation after the contracts had been concluded between the parties. Therefore, the court did not apply the rule in this case. The court explicitly limited itself to the individual cases and the contracts involved therein, without ruling in principle on how record companies in general should deal with the remuneration of artists from streaming revenues.

The same question now comes up for GenAI tools and services: should this different way of music distribution and consumption lead to different approaches to remuneration? A recent study finds that (German) people are less willing to pay for AI generated music compared to human-made music (Friedrichsen et al. 2026). Moreover, electronic music listeners in particular were largely indifferent to whether the music is AI or human made. So, GenAI music offers a combination of lower market entry barriers (already evidenced by surges in uploads of GenAI music on various platforms), and listeners willing to listen but less so to pay. The serious risks of dilution of royalty pools are obvious.

Legal resolutions

Let us now assess some regulatory interventions that have potential to improve fair remuneration despite market saturation setbacks. For one, introduction of EU-wide residual remuneration rights has been proposed to fix substandard payment of creators in the streaming era, and more recently in light of GenAI too (Senftleben & Izyumenko 2025; Strowel 2025). The idea is that on top of exclusive rights, an additional non-waivable remuneration right is granted to those who personally invest in the creation of valuable content, i.e. authors and performers. Although intended to create new income streams, from an economic perspective such an intervention could be dismissed as distribution of scarcity. That scarcity is likely to remain. Listeners do not seem willing to pay more, and perhaps more importantly the allocation to music creators of what is earned by large rightholders and platforms remains skewed. What this residual remuneration might look like will depend in large part on the forthcoming Streamz and Others CJEU ruling (see here).

For original works used in GenAI output, there is also a question concerning how to apply the remuneration right when no traditional “play” takes place. With the double revenue model put forward in some of the GenAI-label deals, should it be linked to both the training fee and the output fee? That is, if the nascent technologies for attribution – tracing AI outputs back to the specific training inputs that shaped them – turn out to be sufficiently workable (Kuhn et al. 2025).

Other regulatory interventions that could counteract the dilution of the royalty pool are conceivable. We could take the example of the EU regulation for broadcasting and streaming services, the Audiovisual Media Services Directive (AVMSD). According to Article 13, Member States have to ensure at least a 30% share of European works on audiovisual streaming services in their jurisdiction, as well as the prominence of those works. Analogously, one could think of a minimum percentage and prominence of human-made works on (traditional) music streaming services.

Some services are already trying to distinguish themselves with more outspoken AI policies on their own initiative. Deezer focuses on transparency, tagging fully AI generated songs and excluding them from algorithmic and editorial recommendations. In doing so, they are already ahead of the AI Act’s transparency and labelling requirements for GenAI content that enter into force in August 2026. Music platforms like Bandcamp and cooperative Subvert are more prohibitive, banning (fully) AI generated music altogether. Some commentators have speculated about increasing divisions between traditional streaming services on the one hand, and GenAI focused services on the other (Cooper 2025; Rosenblatt 2025).

Deepfake takes

The newly licensed versions of the most popular GenAI music services Suno and Udio will include (major) label artists’ deepfakes on a consensual basis. It has convincingly been argued on this blog that a copyright or related rights approach to deepfake regulation is not the way to go. Still, there are plenty of uncertainties with the consent direction chosen. Will deepfakes be desired as a compensatory source of income for labels when artists have disappointing figures for their release, and are thus unable to recoup their advances? Will aspiring new signees be nudged towards consent? The bargaining power of individual artists has always been unequal vis-a-vis (large) record companies, including risks of blacklisting. Additionally, (budding) artists’ knowledge of legal matters can result in them unknowingly signing away control. If consent for deepfake usage is tucked away in a general clause or in terms and conditions, should this be considered full consent in view of the far-reaching use deepfakes allow for? How will consent be handled postmortem?

Moreover, will (major) labels together with GenAI services create the general frameworks for what is and is not considered an acceptable deepfake? The rate at which deepfakes can come into being through GenAI music tools is enormous. Yet which deepfakes an artist would find in accordance with their artistic and personal integrity seems very case dependent. Will smaller artists be able to enforce precise preferences? Collective bargaining about deepfakes in the German and American film industry has resulted in some baseline arrangements (see here). So far, we have not witnessed a similar central mobilisation for collective agreements in the music industry.

Scarcity vs. abundance

The private ordering by major labels described above may increase even further in scale. Since the end of 2024, UMG, the largest major label with an estimated market share of at least 31.5%, has been aiming to acquire Downtown Music Holdings, the largest independent music distributor. In 2025, the European Commission opened an investigation in response, to determine whether the acquisition is competition law-proof. The Commission’s preliminary concerns include that the deal may reduce competition in the market for artist and label services by removing an important competitive force. Multiple independent labels and interest groups have urged the Commission to block the acquisition.

UMG stated that "this deal is about offering independent music entrepreneurs access to world-class tools and support to help them succeed". It is so broad a statement that trying to interpret it might be in vain. Still, one explanation could make sense. Namely, that “world-class tools” – at least in part – refers to the UMG’s AI streaming service with Udio and AI-tool partnership with Spotify, both announced around the same time as the statement. For UMG, acquisition of the huge indie distributor would mean more music (data) to train data hungry models on and more musicians to deepfake on the new services and tools. The newest of which will spring from NVIDIA’s Music Flamingo model through the recent deal between UMG and the AI giant (sharply unpacked at The IPKat).

The licensing deals of recent months will provide more income – to some actors – compared to when virtually all of the internet’s music was scraped and used for GenAI tools without compensation. How substantive the announced compensation will really be remains unclear as of yet. Keller’s recent comment with regard to the handling of access to GenAI works on the internet in general is significant here. Artificial scarcity may produce deals in the short term. But it risks accelerating concentration in the long run, leaving smaller publishers and the public interest “structurally disadvantaged in an internet rebuilt around controlled access rather than shared abundance” (Keller 2025).

The road ahead for the music sector is unclear. Embracing the abundance of musical works can clearly lead to an oversaturated market with a diluted royalty pool. Conversely, artificially creating scarcity can facilitate a few dominant players taking an unequal benefit of the communal value of music.