On 21 May 2026, Universal Music Group and Sony Music Entertainment filed a motion in the U.S. District Court for the District of Massachusetts seeking to expand their copyright infringement case against Suno, Inc. to cover 61,026 sound recordings. The case, captioned UMG Recordings, Inc. v. Suno, Inc., 1:24-cv-11611, sits before Judge Dennis Saylor IV. A parallel matter against Uncharted Labs, the company behind Udio, is moving forward in the Southern District of New York under UMG Recordings, Inc. v. Uncharted Labs, Inc., 1:24-cv-04777, before Judge Alvin K. Hellerstein, with Sony separately moving to add more than 30,000 recordings to its complaint there. Discovery, the labels say, has shown that the two AI music generators were trained on millions of copyrighted master recordings, in part by stream-ripping audio from YouTube.
The cases will set the shape of the next decade of generative audio. They will also, almost certainly, leave African music out of the room.
What the lawsuits actually cover
The complaints filed by Universal, Sony, Warner and the Recording Industry Association of America in June 2024 do one specific thing. They assert ownership of master recordings registered with the United States Copyright Office, identify outputs from Suno and Udio that allegedly reproduce those masters, and seek statutory damages of up to 150,000 dollars per work for wilful infringement. The proposed second amended complaint against Suno raises the asserted catalogue to more than sixty thousand works. It also adds a claim under Section 1201 of the Digital Millennium Copyright Act, alleging that Suno circumvented technological protection measures on YouTube to acquire training audio.
The exhibits, the takedown lists and the discovery requests are all in English. The catalogue numbers reference U.S. Copyright Office registrations. The damages calculations assume statutory protection that only exists once a work is registered in Washington. The framework is built around major-label masters, and the major labels have the resources to enforce it.
Whose music is actually in the training set
Suno and Udio do not publish their training corpora. The strong inference from output behaviour, model size and the labels’ own audio-fingerprinting analysis is that the corpora were scraped from open audio surfaces, principally YouTube, and likely SoundCloud, Audiomack, Bandcamp, and the long tail of MP3 archives that the open web still indexes.
That description maps almost exactly onto the way African music has been published for the last fifteen years. Nigerian Afrobeats, Ghanaian hiplife, South African amapiano, Tanzanian bongo flava, Congolese rumba, Ivorian coupé-décalé, Egyptian mahraganat and Ethiopian eskista all moved from physical and broadcast distribution to YouTube and Audiomack uploads long before they had reliable presence on Spotify or Apple Music. The catalogues are there. They were scraped.
African music is also disproportionately useful to a generative audio model. The rhythmic vocabulary is denser than the four-on-the-floor backbone of most Western pop. The polymetric structures of mbira music, the call-and-response architectures of Yoruba juju, the log-drum patterns of amapiano and the syncopation of soukous all carry information that thins out generic AI output if removed. The vocal techniques, from Mbube harmony to North African melisma, give models texture they cannot get from major-label rock and pop. The languages, Yoruba, Twi, Wolof, isiZulu, Lingala, Amharic, Swahili and Arabic dialects across the Maghreb, give the models phonetic range that English-only training corpora lack. Removing African music from a model would make the model demonstrably worse at producing what is recognisably contemporary music.
The silence from African institutions
Six months into a litigation that will shape the global market for AI-generated music, no African collective management organisation has filed against Suno or Udio. The Nigerian Copyright Commission has not issued a public position on generative audio training. The Performing Rights Organisation of Nigeria, the Music Copyright Society of Kenya, the Composers, Authors and Publishers Association of Ghana, COSON, SAMRO and SACEM Cameroon have not formed a joint working group on training-data audits. There is no African equivalent of the European Commission’s Article 53 AI Act guidance, no equivalent of the U.S. Copyright Office’s three-part AI report.
The asymmetry is not accidental. It is structural. Litigation against frontier AI companies requires forensic discovery budgets that most African CMOs do not have. SAMRO is still working through the FY2026 distribution rules and a member-facing royalties drive. MCSK has cycled through governance crises. COSON spent years in court over its own operating licence. The capacity to commission audio-fingerprinting analysis at scale, file complex pleadings in U.S. federal court and sustain a multi-year discovery battle does not exist on the continent at the institutional level today.
The ethnomusicology grab, digitised
There is a longer pattern here. In the twentieth century, Western ethnomusicologists travelled to West African villages, recorded ceremonial drumming on reel-to-reel tape and deposited the recordings in university archives in Berlin, Paris and Bloomington, Indiana. Some of those recordings were later licensed for sample libraries that producers in London and Los Angeles built careers on. The artists named on the field notes received nothing.
The generative-AI training corpora are the same operation at a different scale. The recordings are no longer on reel-to-reel tape. They are on YouTube, ripped by an automated pipeline, fed into a model that can produce derivative output at the rate of a song every twenty seconds, sold as a subscription service, and protected by the legal argument that the underlying training was fair use. The artists are not named. The villages are not credited. The compensation arrangement is identical, which is to say there is no compensation arrangement.
What standing African institutions could actually do
The lawsuits running in Massachusetts and the Southern District of New York will not include African plaintiffs because there are no obvious ones. There are still meaningful actions available.
The Nigerian Copyright Commission could open a public consultation on generative AI training and copyright, on the model of the U.S. Copyright Office. It would not have extraterritorial enforcement reach, but it would establish a record. SAMRO, MCSK, COSON, COSGA and SACEM Cameroon could form a pan-African working group on AI training audits and ask member rightsholders to authorise collective representation in foreign litigation. The cost is low and the political weight of a coordinated African position would be substantial.
African distributors and aggregators have the cleanest leverage. Boomplay, Audiomack, Mdundo, Africori, Tribesound and the platforms operating downstream of them control the catalogue licences that feed every commercial DSP. Any of them could insert training-data audit clauses into their next round of licensing renewals. The legal architecture is the same one Warner used with Audiomack and the same one Universal used with TikTok in the May 2026 deal. The question is whether African catalogue holders have the leverage and the appetite to ask the same questions about training corpora that the majors are now asking.
InterSpace Distribution operates inside that catalogue layer. The position we have argued internally, and that we will argue publicly through this publication, is that any DSP or AI music platform requesting catalogue access from an African distributor should be required to disclose its training-data provenance and to commit to a remediation mechanism for unauthorised prior use. That is the minimum standard the majors are now demanding in U.S. court. African catalogue owners are entitled to ask for the same thing.
The Suno and Udio cases will resolve, one way or another, in the next eighteen months. The settlement structures, the licensing frameworks and the opt-out mechanisms that emerge from them will become the global default. African music will be inside the training data whether African institutions are inside the conversation or not. The window for the second condition is open now, and it is narrow.
