Sony Just Refiled Against Udio With 30,117 Songs. Statutory Damages Turn That Into a $4.5 Billion Number.
TL;DR
On July 20, Sony Music filed a second copyright lawsuit against AI music startup Udio, this time asserting 30,117 sound recordings it says Udio copied without permission to train its generative music models. The complaint landed in the Southern District of New York, weeks after a judge refused to let Sony add those same recordings to its existing case.
The story is not really about music. It is about arithmetic. U.S. copyright law lets a rights holder claim statutory damages of up to $150,000 per work for willful infringement, and 30,117 works times $150,000 is roughly $4.5 billion. You do not need a damages expert to reach that number. You need a calculator.
What actually happened
Back in June 2024, the major labels, coordinated by the RIAA, sued both Udio and its rival Suno for training on copyrighted recordings. The Udio case went to SDNY before Judge Alvin K. Hellerstein; the Suno case went to Massachusetts. That original Udio suit is narrow: it turns on 333 specific recordings.
Then discovery happened. Once Sony got to inspect Udio's training data, it ran audio fingerprinting across the corpus and, per its filings, matched hundreds of thousands of its own recordings. It tried to fold a first tranche, those 30,117, into the running case. In late June, Hellerstein said no: the works could be litigated, just not bolted onto that docket. So Sony did the obvious thing and opened a second one.
"Udio's belated embrace of licensing only underscores the unlawfulness of its decision to copy," Sony's new complaint argues, per Music Business Worldwide.
The math is the weapon
Here is the part every builder should internalize, because it is not intuitive. In most lawsuits you have to prove harm: lost sales, a dollar figure you can defend. Copyright is different. For registered works you can elect statutory damages, a flat per-work amount a jury can set as high as $150,000 each when infringement is willful, with zero obligation to show you lost a cent.
So the count of works, not the harm, sets the exposure. Think of it as a parking fine that does not care whether you actually blocked anyone: each ticket is a fixed maximum, and Sony is alleging you left 30,117 cars in the lot. That is why a training-data case scales into the billions while a normal contract dispute does not.
Why Sony refiled instead of walking away
Two reasons. First, damages. The original 333-recording case caps out around $50 million at the willful maximum, real money but survivable. The 30,117-recording suit caps out near $4.5 billion, which is not. Second, leverage. Sony is, per Music Business Worldwide, the last major label still fighting Udio in court rather than licensing to it, after Universal and Warner reportedly struck deals. A $4.5 billion ceiling is a negotiating position with a very loud voice.
Books today, a soundtrack tomorrow
The timing is not subtle. The same day this surfaced, a federal judge gave Anthropic's $1.5 billion settlement with book authors its final approval, pricing pirated training books at about $3,000 each. The mechanism that produced that number, per-work damages multiplied across a huge corpus, is exactly the one Sony just pointed at Udio. Books got a price this week. Music is next in line, and the opening bid is an order of magnitude higher.
What this means if you train on scraped data
You are probably not training a frontier music model. It does not matter. The precedent being set is about provenance, and it reaches any model trained on content you do not have rights to:
- Your corpus is your liability, itemized. Exposure scales with the number of registered works you ingested, not with how much money you made or whether the output competes with the original.
- Discovery reads your training set. Sony did not guess. It fingerprinted Udio's actual data. If you cannot explain where each shard came from, opposing counsel will.
- Licensing after the fact can be used against you. Sony is literally quoting Udio's later licensing deals as evidence the earlier copying was unlawful.
- Fair use is a defense, not a force field. It might win. But you argue it after you have already been sued for a number with ten digits in it.
Key Takeaways
- Sony Music filed a second suit against Udio on July 20 in SDNY, asserting 30,117 sound recordings, after a judge barred it from adding them to the original 333-recording case.
- Statutory damages of up to $150,000 per willful work put the theoretical ceiling near $4.5 billion, versus roughly $50 million in the original case.
- The recordings were identified by audio fingerprinting Udio's training data during discovery, described as only a small portion of the total matches.
- It lands the same day Anthropic's $1.5 billion author settlement got final approval, extending the "price of unlicensed training data" fight from books to music.
- For any builder: legal exposure scales with the count of works in your training set, and your training set is discoverable.
Sources: Music Business Worldwide, RightsTech Project, Billboard, Cornell LII, 17 U.S.C. 504