On 9 September 2026, Suno began rolling out three music-generation models: v6, v6-wild and v6-mini. The company says they were developed with Warner Music Group, BMG and Believe, and describes future experiences in which individual artists can choose to participate and be paid as the next phase. Reuters confirmed that the models can generate new music inspired by licensed music from participating artists. The Verge further reported that v6 was retrained on a new collection of material that includes licensed content from the partners and user-supplied data. The published material does not provide a complete training inventory, the itemised scope of each licence, or confirmation that every artist separately agreed to every possible mode of generation.
The surrounding dispute did not disappear when recording and publishing licences were obtained. On 1 September, Jason Isbell and other musicians filed a proposed class action in a US federal court, alleging that Suno had used their names, images, voices and musical identities without permission. That case concerns rights of publicity and related identity interests, which are distinct from the earlier claims by record companies concerning copyrights in recordings and compositions. Suno denies the allegations and says its product blocks prompts that directly request imitation of particular artists or songs. The court has not ruled on the merits. These points must therefore remain claims and defences by the parties, not facts establishing liability.
Together, these developments raise one question: once an AI company has obtained a licence to use works from rights holders, has it also obtained authorisation to turn a creator’s voice, style or public identity into a condition of generation? My judgement is no. Permission to use a work, artist participation and authorisation of identity are three relations that can overlap but cannot substitute for one another. Establishing one of them can improve the legitimacy of the model’s training or outputs without completing the other two.
The first distinction is between a work and a person. Rights to recordings, musical compositions and particular acts of reproduction, adaptation or distribution may be held by different parties and licensed for specific uses. A creator’s name, vocal characteristics, public image and professional reputation concern how that person is identified, represented and understood by others. A recording certainly carries a performer’s voice and style, but permission to use that recording within an agreed scope does not by itself entail permission to make a system callable as that person. Conversely, an artist’s agreement to a particular personalised experience would not automatically authorise the use of all their works for every form of training and regeneration.
Style needs separate treatment as well. It is not an object with naturally precise boundaries and exclusive ownership. Musical traditions, techniques and genres are usually formed by many people, while creation depends on quotation, variation and mutual influence. Similarity of sound alone therefore does not establish that identity has been appropriated. The question shifts from abstract style to representation when a combination of name prompts, vocal resemblance, visual presentation, marketing context and attribution leads ordinary listeners to understand an output as the expression of a particular artist. At that point, the issue is whether the system is helping a user create or borrowing an identifiable person to give the result a source, credibility or commercial appeal.
The Stanford Encyclopedia of Philosophy’s account of intellectual property notes that copyright is primarily organised around concrete expressions and divisible rights of use, while European moral-rights traditions and personality-based arguments address attribution, integrity, personality and reputation. A philosophical distinction should not be treated as a statement of current law, and it cannot settle the pending case. Its conceptual value here is that economic rights and personality interests protect different objects. A licensing transaction can answer who is entitled to permit a use of a work without necessarily answering in whose identity a generative system appears to speak, to whom listeners attribute the output, or who bears the resulting reputational consequences.
Suno’s own presentation of v6 supports this distinction. The company says that industry partners have already helped develop the models, while describing individual artists’ voluntary participation and remuneration as a future experience. If catalogue licences fully contained every form of identity authorisation, there would be no reason to make artists’ choice to participate a separate step. This is only a limited inference from the public structure. It suggests that Suno also distinguishes catalogue-level licensing from individual participation, but it does not establish that any specific right is missing from the current agreements, whose full terms are not public.
In Sustenesis Theory, Difference is not merely ordinary difference. It is distinguishability and the starting point from which relations form. The work, the rights holder, the creator’s identity, the model output and the user must remain distinguishable here. Once they are compressed into the claim that “the content is licensed”, the relation requiring judgement disappears. Constraint is not merely an external restriction either; it is a condition that makes some modes of formation possible and others impossible. Contractual scope, artist opt-in, prompt blocking, rules for vocal resemblance, labelling, remuneration and withdrawal determine how a system can call upon works and identities.
Sustained Coherence is the structural consistency maintained through constraints, feedback, correction and effective operation. In AI music, authorisation is not a one-off label fixed at the moment of signing. The relevant question is whether training material, prompt handling, generated outputs, identity labelling, revenue allocation and complaint correction continue to correspond. A model might use licensed works during training yet create an identity implication at output time that was never agreed. It might also avoid copying any specific work while repeatedly drawing commercial value from a person’s public identity through voice and marketing. Licensing becomes a maintainable authorisation relation only when these stages can be examined separately, rather than being summarised by the fact that a partnership exists.
The significant change in v6 is that at least some musical material is now placed within an explicit structure of industry licensing and collaboration, accompanied by a public commitment to develop artist opt-in mechanisms. This is more open to examination than a training relation with unclear sources and permissions. The present evidence does not support the conclusion that licensing works has resolved the question of creators’ consent. Nor does it support the opposite conclusion that stylistic resemblance by itself constitutes an infringement of identity. What needs to be established is not a general answer to whether material was “authorised”, but which works and training or output uses are covered, who can consent on an artist’s behalf, what triggers identity-based generation, whether artists can refuse, withdraw and correct, and whether those conditions remain effective across model versions.
References
https://suno.com/blog/introducing-v6
https://www.reuters.com/legal/litigation/suno-releases-new-ai-music-models-partnership-with-warner-music-bmg-2026-09-09/
https://www.theverge.com/ai-artificial-intelligence/991977/suno-releases-its-first-ai-music-model-made-with-record-industry-help
https://www.reuters.com/legal/government/jason-isbell-other-musicians-sue-suno-over-ai-training-2026-09-01/
https://plato.stanford.edu/entries/intellectual-property/
Discover more from Geoffrey Chen
Subscribe to get the latest posts sent to your email.