US labels body the RIAA is suing two of the most prominent AI music startups, Suno and Udio, accusing both of copyright infringement “on a massive scale”.
All three major labels are listed as plaintiffs in the lawsuits – Sony Music Entertainnment, UMG Recordings and Warner Records specifically.
The body is seeking declaratory judgments that Suno and Udio infringed the labels’ sound recordings; injunctions barring them from doing that in the future; and statutory damages of up to $150k per work infringed, plus legal costs.
The RIAA has filed the lawsuits on behalf of the labels, but today’s announcement came with supportive quotes from independent labels body A2IM, publishing body the NMPA, campaigning organisation the Artist Rights Alliance (ARA), rights organisation SoundExchange, labour union SAG-AFTRA and Songwriters of North America.
Suno and Udio lawsuits: the details
The lawsuits have been filed under existing copyright laws, and thus focus on the technical aspects of how labels believe the two companies have infringed.
“Building and operating [these services] requires at the outset copying and ingesting massive amounts of data to ‘train’ a software ‘model’ to generate outputs,” is how one section of the filings laud this out.
“This process involved copying decades worth of the world’s most popular sound recordings and then ingesting those copies [to] generate outputs that imitate the qualities of genuine human sound recordings.”
In another relevant excerpt, the RIAA claims that Suno and Udio are “not exempt from the copyright laws that protect human authorship. Like any other market participant, [they] cannot reproduce copyrighted works for a commercial purpose without permission.”
The lawsuits also focus on both startups’ anticipated defences: that any use of copyrighted music to train their models is ‘fair use’.
Indeed, in its ‘Fact Sheet’ sent to journalists ahead of the announcement this afternoon, the RIAA said that both Suno and Udio had raised fair-use defences “in pre-litigation correspondence” with the labels body.
The RIAA sees this as a gotcha. “‘Fair use’ only comes up as a defense to unlicensed copying (though one that does not apply in this case). If Suno and Udio did not copy recorded music, why would they need to invoke this defense?” it asks.
“Both companies refused to come clean about their copying of sound recording while simultaneously claiming that if they did copy, it was fair use.”
“These are straightforward cases of copyright infringement involving unlicensed copying of sound recordings on a massive scale,” added RIAA chief legal officer Ken Doroshow. “Suno and Udio are attempting to hide the full scope of their infringement rather than putting their services on a sound and lawful footing.”
Who are Suno and Udio?
At their core, Suno and Udio are both ‘text-to-music’ AI models. Users type in prompts describing what they want the systems to make, hit the ‘Create’ button, and then see what comes out.
However, both have more advanced features including typing in your own lyrics rather than have the systems generate the words. Earlier this month, both also added the ability to upload audio files to use as seeds for the AIs’ output.
Both companies warned users not to do this with copyrighted music, and both said that tracks created using uploaded audio would remain private to the user, rather than published on their services for others to hear.
Suno emerged from stealth mode in December 2023, announcing a deal with Microsoft that saw its model integrated into the tech giant’s Copilot AI platform. The startup raised a massive $125m funding round in May 2024 and said that it had already been used to create music by more than 10 million people.
Lightspeed Venture Partners, one of the big VC firms that invested in Suno, claimed that the company’s latest model could “generate full length songs worthy of top 40 radio airplay in mere seconds”, and promised “a product roadmap that will reinvent the way we all experience music”.
Udio launched in April 2024, founded by former researchers from Google’s AI division DeepMind. It also had some starry Silicon Valley investors in the shape of VC firm Andreessen Horowitz, Instagram co-founder Mike Krieger and the boss of Google’s Gemini AI initiative Oriol Vinyals.
However, Udio also had music-industry investors including distributor UnitedMasters and artists Will.I.Am and Common. Music Ally interviewed CEO David Ding and COO Andrew Sanchez as it launched, and they said that the company was working on business models that would benefit human musicians.
“We’re in conversations with many different folks, and we’re building out new patterns for how artists are going to make money off of this in future,” said Sanchez. “We think there’s going to be a whole new model around this, and we think it can be an equitable and really fantastic one.”
Why are labels suing Suno and Udio?
The music industry has spent the past year and a half setting out its views on how AI music models and companies should operate, and how they should be regulated. In a nutshell, here’s what the industry wants: permission, payment, transparency and protection for musicians.
They want AI companies to ask permission before training their models on copyrighted music; they want them to pay to do it through licensing deals; and they want them to keep and share good records of their training materials so that rightsholders can check in on them. And they want laws to make it illegal to ‘clone’ a musician’s voice or likeness without their permission.
There have been various announcements from rightsholders and industry bodies along these lines, but you can find a neat summary in the industry-backed Human Artistry Campaign, which launched in March 2023 with 40 organisations signed up, and a set of regulatory principles they wanted governments to adopt.
What the music industry is pushing back against is the suggestion that AI companies shouldn’t have to seek permission, offer payment or be transparent about training their models on copyrighted music.
This is a bigger debate that covers all types of generative AI (or GenAI) technologies from ChatGPT-style chatbots through to photo, video and music creation models. The fair use argument is summarised here in various submissions by tech companies to a US Copyright Office study last year.
The view from labels – as expressed in today’s lawsuits – is that this argument is based on a fundamental misunderstanding of fair use, and that Suno and Udio’s services fail to meet any of the four factors that must be considered in the application of that law.
While the music industry has focused its energies publicly on pressing governments and regulators to back its views with new laws and rules, it was always clear that if rightsholders believed that musical AI models were infringing existing copyright laws, they’d have to also file some lawsuits.
We’ve seen one big lawsuit already, although it focused on a different strain of GenAI. Music publishers sued a prominent AI company called Anthropic, which makes the Claude chatbot – a rival to ChatGPT.
That lawsuit was about lyrics, with the publishers claiming that Claude had not just been trained on them, but could be prompted to regurgitate them – thus producing copyright-infringing outputs. If you have yet to experience the delights of ‘Goodbye, Yellow Brick Butt’, well, now’s the time.
In the UK, labels body the BPI also threatened to sue a startup called Jammable (formerly Voicify) earlier this year, over its service enabling people to “create AI covers” with famous singers’ voices. But today’s RIAA vs Suno and Udio lawsuits are a big step up in terms of profile and likely impact.
But why sue Suno and Udio specifically?
Good question. First, it’s because they’re currently the two buzziest AI music models, with the most momentum in terms of users and media coverage.
Second, they have financial backing – to the tune of $125m in Suno’s case – with big Silicon Valley investors involved. Third, both offer paid subscription models, so they’re making money from their (in the eyes of the RIAA) infringement.
Fourth, both have been accused publicly of training their models on copyrighted music. AI-music veteran Ed Newton-Rex published two long tweetstorms in recent months – the Suno one is here and the Udio one here – outlining his (successful) efforts to get these models to generate music in the style of famous artists. These receipts are key to the RIAA’s case.
When Music Ally interviewed Udio, we asked about training materials. “I can only get high-quality outputs [if] we have to train on a large amount of publicly-available and high-quality music,” said Ding.
“We think that we can also show that when you are creating transformative novel outputs, there’s going to be a net good to the industry,” added Sanchez – ‘transformative novel outputs’ being a strong pointer to a likely fair-use defence.
But fifth – and we think this is a very important point – Suno and Udio are both really good. By which we mean the quality of their output is noticeably better than rivals who do not train on copyrighted music.
That’s a big danger for the music industry. If the best models are the ones that train without permission, and if they run away with the market, that’s bad news for rightsholders. The startups they see as ethical will be left straggling behind.
If you follow the industry’s arguments through to their logical conclusion, they don’t really have a choice: they have to sue now.
Suno and Udio: responses
Suno has responded to the lawsuits with a statement from CEO Mikey Shulman, sent to Music Ally by the company’s representative.
“Suno’s mission is to make it possible for everyone to make music. Our technology is transformative; it is designed to generate completely new outputs, not to memorize and regurgitate pre-existing content. That is why we don’t allow user prompts that reference specific artists,” he said.
“We would have been happy to explain this to the corporate record labels that filed this lawsuit (and in fact, we tried to do so), but instead of entertaining a good faith discussion, they’ve reverted to their old lawyer-led playbook,” continued Shulman. “Suno is built for new music, new uses, and new musicians. We prize originality.”
We have also reached out to Udio for a statement, and will update this story accordingly.
So, what happens next?
To state the obvious: these lawsuits will either end up in court, or they will end up in settlements and licensing deals.
If it’s the former and the RIAA wins, it sets a precedent that training an AI model on copyrighted materials is not fair use, and so is copyright infringement if deals aren’t in place. If Suno and Udio win, it sets the opposite precedent.
If it ends with settlements, markers may be laid down for the kind of deals that can be struck between rightsholders and AI music companies. But rest assured, the lobbying of politicians and regulators will continue regardless.
We’ll finish with a few stray thoughts on today’s news though:
- One theory we’ve heard doing the rounds in the music world focuses on both companies’ investors, particularly the big VC companies. It suggests that they would actively relish legal battles to establish the fair use question – something that may affect other GenAI startups they have invested in. Whether you buy the ‘Suno and Udio as Trojan horses for a bigger fair use battle’ theory or not, the stakes are clearly high for the Lightspeeds and a16zs of the world.
- This is all quite awkward for Udio’s music-industry investors. A penny for the thoughts of UnitedMasters, Will.I.Am and Common at this point…
- In the music industry’s view, training a musical AI without permission, payment or transparency is wrong, and so these companies need licensing deals. But the obvious follow-on question is this: what do those deals look like? How are they structured: flat fees to cover the inputs (the music used to train) but also royalties in some form for the outputs (the music created with them)? And how will that money be fairly shared with artists and songwriters? Answering these questions is important – not least to ensure the startups who do want to license aren’t left behind.
- What was that line about history not repeating itself, but rhyming? The fact that the RIAA has filed today’s lawsuits can’t help but be a reminder of the heyday (or nadir) of filesharing in the late 90s and early 2000s, when the RIAA was prominent in lawsuits both against P2P services and the people using them to download music. The main difference is that today’s lawsuits are only focused on the services – Suno and Udio – not the people using them to create music.
