Could the most contentious lawsuits in AI music end up laying the foundation for its most lucrative partnerships? That is the question facing the industry as major labels shift from litigation to licensing, forging deals with platforms like Suno, Udio, and newcomer Klay that could redefine how generative music is created, distributed, and monetized.
Just a year ago, Universal Music Group (UMG), Sony Music Entertainment (SME), and Warner Music Group (WMG) were united in accusing Suno and Udio of training their AI models on “reams of copyrighted recordings” without permission. The suits alleged direct copying of sound recordings, citing outputs that replicated elements from Chuck Berry’s “Johnny B. Goode” and Mariah Carey’s “All I Want for Christmas Is You.” Damages were sought at up to $150,000 per infringed work, plus penalties for alleged encryption circumvention. Suno countered that “no one owns musical styles” and claimed its technology was “transformative,” designed to generate new outputs rather than regurgitate content.
Now, the narrative has shifted. WMG has settled with both Suno and Udio, replacing court battles with licensing agreements. UMG has also reached a deal with Udio, and all three majors have signed with Klay, a Los Angeles-based AI music startup that trains its models exclusively on licensed recordings. Klay’s platform will allow subscribers to remix existing tracks in different styles a process it calls “active listening” while an attribution system identifies source recordings and ensures per-stream payments to rights holders. This architecture contrasts sharply with Suno’s text‑to‑music generation, which has been used by nearly 100 million users in the past two years.
From a technical perspective, these partnerships hinge on the dataset licensing frameworks that have been absent in most early AI music ventures. Klay’s model, trained only on authorized content, sidesteps the fair use debate entirely. By embedding attribution metadata in outputs, it enables automated royalty routing a capability that could be extended to other AI platforms. Suno, meanwhile, is preparing higher‑quality models trained on licensed music, though questions remain over whether artists will be allowed to opt out of training datasets. WMG CEO Robert Kyncl underscored the principle: “AI becomes pro‑artist when it adheres to our principles: committing to licensed models, reflecting the value of music on and off platform, and providing artists and songwriters with an opt‑in for the use of their name, image, likeness, voice and compositions in new AI songs.”
Licensing deals are also forcing AI companies to integrate content‑control mechanisms. Udio’s settlement requires disabling downloads of generated music, while Suno’s agreement permits downloads for paid users but imposes monthly caps. The mentioned actions are consistent with the larger trends in the music industry to filter and monitor for unlawful uses of copyrighted content. Deezer has its own artificial intelligence technology that can track down and take off entirely AI created songs from their lists. In contrast, Spotify uses a system where copyright owners voluntarily disclose their content, which is at risk of being abused, yet is less likely to identify incorrectly.
Behind the business headlines lies a deeper engineering challenge: building guardrails that prevent models from reproducing copyrighted works verbatim while still enabling stylistic emulation. Output‑based guardrails, such as those Anthropic applied to block lyric reproduction, can be bypassed through descriptive prompts. Input‑based controls ensuring only licensed or public domain material enters the training set offer stronger legal protection but require transparent dataset documentation. Companies like Audible Magic are partnering with AI music platforms to detect copyrighted audio in prompts before generation, adding another layer of compliance.
The monetization models emerging from these deals reflect lessons from the streaming era. Labels are pushing for per‑play compensation, mirroring Spotify’s royalty structure, rather than lump‑sum licensing. In addition to aligning the interests of AI platform providers with those of the rights holder, this method also allows for scalable revenue generation to occur as more and more user generated AI music exists. For artists, this will ultimately be determined by the level of transparency provided by AI platforms with respect to when their works are used, the frequency of use, and the manner in which revenue is determined. As Irving Azoff of the Music Artists Coalition cautioned, “We have to make sure [technology] doesn’t come at the expense of the people who actually create the music—artists and songwriters.”
The shift from “build first, clear later” to licensed, rights‑centric AI models signals a maturation of the sector. Additionally, this shift in thinking represents a change in strategy for translating AI into something major label to continue to use as an asset. Instead of being a hindrance or an obstacle to getting paid, now they will begin to implement (1) consent, (2) attribution and (3) remuneration into their technical and contractual agreements. It remains to be seen whether these frameworks can grow and whether they can deliver the same leverage to independent artists that major labels would deliver. However, the technical and contractual framework being developed will be critical in determining the way that AI Music is paid for and who gets paid what in years ahead.
