In just six weeks from June to July 2025, The Velvet Sundown released three albums of thirteen songs each. Virtually overnight, the songs racked up millions of streams. The trick? The “band” and its images, lyrics, and songs were all output by generative artificial intelligence (AI). From the characteristic yellow tint of the Dall-E-generated images, made-to-match with the vintage aesthetic, to the neo-kitsch ChatGPT-generated lyrics reminiscent of hippie age protest songs, and the SUNO-generated sonic artefacts masked by a rather bland and saturated ‘70s soundscape, it all looks coherent and skillfully done.
The Velvet Sundown is a good example to illustrate the many copyright-related issues with which SUISA is increasingly confronted in the era of generative AI.
Are the Velvet Sundown’s songs protected by copyright?
The Federal Copyright Act defines protected works as literary and artistic intellectual creations with individual character. Accordingly, to qualify for protection a work must be an expression of thought and an act of human will. It follows that the author can only be a natural person, i.e. a human being. Consequently, pure AI output, like the Velvet Sundown’s songs, does not enjoy copyright protection and is in the public domain. Put simply, the decisive factor is whether a composition is a human creation or the output of an algorithm.
A work cannot be protected by copyright if it is simply the result of a prompt – i.e. a short input instructing an AI model to generate content on the prompted basis. Especially if two identical prompts produce two completely different melodies – a clear indication that the melody was not created by humans but output by generative AI.
Copyrightability is more difficult to assess in cases where the human-made part of a song – as opposed to that generated by an algorithim – is large enough on the one hand, and is expressed in the generated output on the other. In some cases, a work may still be copyrightable even though it was produced with the help of AI. That applies, for example, if the music for a song is AI-generated while the lyrics are written by a human (or vice versa). Such hybrids are protected works if, standing alone, the part of the work created by a human being satisfies copyright protection requirements.
Could these songs be registered with SUISA?
There are two requirements enabling SUISA to charge a licence fee for the use of a musical work: the musical work must be protected by copyright, and the rights of use in that work must have been transferred to SUISA. By registering a work with SUISA, the necessary rights for licensing music usage are transferred to the latter. Music generated by AI poses a special challenge in terms of copyright protection.
Meanwhile, there are tools on the market purporting to be able to distinguish between songs generated by artificial intelligence and conventional music. Analysing the songs on the third Velvet Sundown album with such a tool (called an AI detector) it appears with a 98% probability that all thirteen tracks were generated by Suno, a popular music AI. The problem is that the analysis focuses on the production rather than the composition. We cannot simply assume that, in addition to the production, the lyrics or composition were also generated by Suno.
Because of the uncertain evidence, SUISA cannot determine with certainty whether a given work is protected by copyright. However, it contractually ensures that it does not licence music in the public domain: by signing SUISA’s General Terms and Conditions for Rights Administration, authors undertake not to register with SUISA any songs that were entirely generated by artificial intelligence.
To the extent the Velvet Sundown’s songs are neither works composed by a human nor hybrid works (see above), but purely AI-generated output, they cannot be registered with SUISA. As far as we can tell, no registration applications have been received by SUISA or any of its sister societies for these songs.
For more information about the registration of works produced with the help of AI, see the relevant FAQs on SUISA’s website.
Is it permitted to upload to Spotify etc. songs that are entirely generated by AI?
In principle, AI songs that are neither copyrighted nor registered with a collective management organisation can be uploaded to streaming services. Depending on the aggregator – in the case of the Velvet Sundown, this is DistroKid – the uploading of music generated entirely by AI may be tolerated or prohibited (e.g. iMusician). DistroKid references the AI provider’s general terms and conditions. And Suno’s Terms of Service stipulate that only users who have purchased a subscription may make commercial use of the songs. However, Suno reserves the right to use any output generated by users, including for commercial use by third parties. Moreover, Suno reserves the right to use any prompts, lyrics or audio files uploaded by users for its own purposes – not only for further training of its model but also to pass on to third parties.
Are AI generated songs labelled as such?
Conversely to Spotify or Tidal, Deezer, the French streaming service, indicates when a song is likely to have been output by AI; this indication appears on all three of the Velvet Sundown albums. According to Deezer, some 30 000 AI-generated songs are uploaded to the platform every day, accounting for a 28% share (status: September 2025). The AI notice informs listeners that AI music is on the platform; moreover, such songs no longer appear in Deezer’s editorial and automated recommendations. This is designed to reduce the visibility and reach of AI music, thereby generating less revenue for it. However, because Deezer cannot distinguish between composition and production, notices of this kind could also result in human works being declared ‘AI-generated content’ when implemented by an AI tool such as Suno.
Could the songs infringe third-party copyrights?
If an AI-generated song falls within the scope of protection of an existing song, this could constitute a copyright infringement. For this to be the case, the AI song would either have to be identical to a protected work, or the individual character of the original would have to be recognisable in the AI-generated song. Gema, SUISA’s German sister organisation, has managed to demonstrate that Suno is not only capable of generating new material, it can also plagiarise.
Plagiarism is not to be confused with mere imitations of stylistic elements by generative AI. For music, this means that composition techniques and styles can be imitated as long as no specific parts of a work are plagiarised. Chord progressions, i.e. the combination and temporal sequence of chords, are not protectable either.
Is it fair to use AI music?
For a music model like Suno to generate songs, it must first have been trained with large volumes of human-made songs; so far, this has been largely done without permission from the rightholders. There is considerable legal uncertainty surrounding the use of copyrighted works as training input for AI systems – the legal situation needs to be clarified in this regard.
A study conducted by CISAC, the international umbrella federation of collective management organisations, anticipates that, under the existing regulatory framework, AI-generated content will have a huge financial impact on the creative industry. While music creators’ revenues are expected to drop by over 20%, AI providers’ revenues increase by millions each year.
Notwithstanding, generative AI tools like Suno and Udio operate without permission from composers, lyricists, arrangers, performers, publishers, labels, etc., without disclosing their training data, and without paying any licence fees. Therefore, a fair legal framework for cultural creation is urgently required.
The current regulatory framework also spawns uncertainties regarding the use of AI-generated content outside the private sphere, ranging from possible copyright infringements in the form of plagiarism to violations of the (possibly diverging) contractual terms and conditions of AI providers, streaming platforms and aggregators.
