Sony Filed the 30,117 Tracks Again.

Sony Music's original case against Udio was built, like most AI copyright claims, on inference from what the model produced. Then discovery let its lawyers look at the training data itself, and audio fingerprinting produced a specific list: 30,117 Sony-controlled recordings. On 29 June the court refused to let those claims be added to the existing action. On 20 July, Sony filed them as a separate lawsuit.

Why there are two cases

Sony moved in May to amend its 2024 complaint, having inspected training-data material in discovery. On 29 June the court denied leave — in substance, that the additional claims did not have to be brought in that lawsuit rather than that they were bad claims.

Sony treated the ruling as a routing decision rather than a defeat and opened a second action in the Southern District of New York on 20 July. The tracks barred from one pleading returned in another.

What changed evidentially

The significance of the new complaint is the type of evidence behind it. Most training claims argue backwards from outputs: the model can produce something close enough to our catalogue that it must have seen it. Defendants answer that similarity proves style, not copying.

Sony says audio fingerprinting matched more than thirty thousand of its recordings to material in Udio's training data. That is a direct assertion about inputs rather than an inference from outputs, and it is the kind of claim that either survives technical scrutiny or collapses under it.

It remains an allegation. Nothing about it has been admitted, proved or tested, and the fingerprinting methodology has not yet faced an adversarial challenge.

What is being claimed

The complaint pleads three routes: infringement of post-1972 sound recordings, infringement of pre-1972 recordings protected under the Music Modernization Act, and circumvention of technological measures under the DMCA.

The third is now the industry's favoured second front. It attacks how the data was obtained rather than what was done with it — a route that does not depend on winning the fair-use argument about training at all.

Reported exposure runs to roughly $4.5 billion, on the statutory maximum of $150,000 per recording plus up to $2,500 for each act of circumvention. Statutory maxima are ceilings, not forecasts, and courts award far less; the figure indicates the shape of the threat rather than a likely outcome.

Sony is now alone

The strategic landscape moved while the amendment was being argued. Universal and Warner have both settled with Udio and licensed their catalogues to it. Sony is the only major label still litigating.

That changes how the case reads. It is no longer a united industry front against a generative music company; it is one label continuing to litigate what its two largest competitors chose to convert into a commercial relationship.

It also raises a question the settlements do not answer publicly: what session musicians and featured performers receive when a catalogue they appear on is licensed for AI training. The American Federation of Musicians has separately sued Universal and Warner over precisely that, arguing the recording agreement's terms are engaged when recordings are put to a new use.

What to watch next

The next reliable documents are Udio's answer or motion, any request to consolidate the two Sony actions, discovery rulings about training-data disclosure, and eventually a summary-judgment record.

Udio has previously sought confidentiality over training-data information on competitive-harm grounds. How much of that record becomes public is, for anyone trying to understand what these models were built from, at least as consequential as who wins.