Like many AI companies, music generation startups Udio and Suno appear to have relied on unauthorized fragments of copyrighted works to train their models. That’s what they and investors recognize, as well as the new lawsuits filed against them by music companies. If these lawsuits were brought before a jury, the trial could be both a damaging expose and a very useful precedent for equally delicate AI companies facing certain legal perils.
The lawsuits filed by the Recording Industry Association of America (RIAA) put us all in the uncomfortable position of supporting the RIAA, which for decades has been the bogeyman of digital media. I have received unpleasant messages from them myself! The matter is simply that clear.
The gist of both lawsuits, extremely similar in content, is that Suno and Udio (strictly speaking, Uncharted Labs doing business as Udio) indiscriminately plundered more or less the entire history of music recorded to form datasets, which they then used to train music-generating AI.
And here, let’s quickly note that these AIs don’t “generate” so much as match the user’s prompt to patterns in their training data and then attempt to complete that pattern. In a way, these models are just making covers or mashups of the songs they’ve ingested.
The fact that Suno and Udio ingested this copyrighted data seems, for all intents and purposes (including legal), very likely. Company executives and investors have been recklessly tight-lipped about copyright challenges in this area.
They admitted that the only way to create a good music generation model is to ingest a large amount of high-quality music. This is simply a necessary step for creating machine learning models of this type.
Then they said they did it without permission from the music labels. Investor Antonio Rodriguez of Matrix Partners told Rolling Stone just a few months ago:
Honestly, if we had had deals with labels at the start of this business, I probably wouldn’t have invested in it. I think they needed to make this product without constraints.
The companies told the RIAA’s lawyers that they believed the media they ingested fell within the scope of the fair use doctrine, which basically only applies to unauthorized use of a work. Fair use is a complex and fuzzy concept in its idea and implementation, but the companies’ use seems to stray somewhat outside the intended safe harbor, such as a seventh-grader using a Pearl Jam song in the background of his video about global warming.
To be frank, it seems like these companies’ golden goose is cooked. They might have hoped to take a leaf out of OpenAI’s book, using evasive language and misdirection to block their less well-heeled critics, like authors and journalists. (If by the time the AI companies’ shenanigans are exposed and they are the only distribution option, it doesn’t matter.)
But that’s harder to do when you have a smoking gun in your hand. And unfortunately for Udio and Suno, the RIAA claims in its lawsuit that it has a few thousand pieces of irrefutable evidence, and that the songs it has are clearly regurgitated by the music models. Its claim: Whether it’s Jackson 5 or Maroon 5, the “generated” songs are slightly truncated versions of the originals—something that would be impossible if the original wasn’t included in the training data.
The nature of LLMs – particularly their tendency to hallucinate and lose plot as they write – prevents the regurgitation of, say, entire books. This likely resulted in a lawsuit from the authors against OpenAI, since the latter can plausibly claim that the snippets its model cites were taken from reviews, front pages available online, etc. (The final goal post move is that they did you used copyrighted works at first but have since stopped, which is funny because it’s like saying you only squeezed the orange once but have since stopped.)
what do you can’t to do is plausibly claim that your music generator only heard a few bars of “Great Balls of Fire” and somehow managed to spit out the rest word for word and chord for chord. Any judge or jury would laugh in your face, and with luck, a court artist will have the opportunity to illustrate that.
This is not only intuitively obvious, but also legally consequential, since the recreation of entire works (scrambled, but obviously based on the originals) opens up a new avenue for relief. If the RIAA can convince the judge that Udio and Suno are causing real and substantial harm to the businesses of copyright holders and artists, it can ask the court to stop the entire activities of the AI companies at the outset of the trial, with an injunction.
The first few paragraphs of your LLM book? That’s an intellectual question to be discussed at length. A one-dollar “Call Me Maybe” generated on demand? Stop it. I’m not saying it’s good, but it’s likely.
The predictable response from businesses has been that the system is not destined to reproduce copyrighted works: a desperate and naked attempt to offload responsibility onto users under the safe harbor of Section 230. In other words, in the same way, Instagram does not is not responsible if you use a copyrighted song to save your Reel. Here, the argument seems unlikely to gain traction, in part because of the aforementioned admissions that the company itself ignored copyright in the first place.
What will be the consequence of these proceedings? As with all things AI, it is quite impossible to tell in advance, as there are few applicable, established precedents or doctrines.
My prediction is that companies will be forced to expose their data and training methods, as these items are of obvious evidentiary interest. And if this evidence shows that they are indeed abusing the copyrighted material, we will see an attempt to settle or avoid a lawsuit, and/or a speedy judgment against Udio and Suno. It’s likely that at least one of the two will attempt to continue, using legal (or at least legal-adjacent) music sources, but the resulting model (by their own training data standards) would almost certainly result in a big step forward. quality would drop and users would flee.
Investors? Ideally, they will lose their shirt, having bet on something that was in all likelihood illegal and certainly unethical, and not just in the eyes of the nebulous authors’ associations, but according to the legal minds of the infamous and ruthlessly contentious RIAA.
The consequences could be far-reaching: If investors in a new generative media startup suddenly see a hundred million dollars vaporized due to the fundamental nature of generative media, a different level of diligence will suddenly seem appropriate.
Companies can learn from lawsuit or settlement documents what can be said – or perhaps more importantly, what should not be said – to avoid liability and leave copyright owners in limbo .
Although this particular lawsuit seems almost a foregone conclusion, it will not be a playbook for suing or extracting settlements from other generative AI companies, but an object lesson in hubris.
It’s nice to have one every once in a while, even if the teacher is the RIAA.