sony sued the same ai music company twice. the dataset kept growing.

sony tried to add more than 30,000 recordings to its copyright case against udio. the judge said no. too late, too much paperwork, wrong case schedule.

so sony filed another lawsuit.

normal industry behavior. one lawsuit reaches the upload limit, open a second tab.

30,117 songs entered the chat

music business worldwide reported that sony filed the new case on july 20. it covers 30,117 sound recordings that sony says udio copied without permission to train its music models.

the complaint also alleges that many recordings came from youtube through stream ripping. udio has acknowledged using audio from youtube as training data, while arguing that model training is fair use.

that last sentence is basically the whole war compressed into one browser tab.

one side sees a library being copied. the other sees a model learning from examples. between them sits a machine that can make a song before the lawyers finish reading the track list.

licensing fixed the future, apparently

universal and warner settled with udio. several other rights holders signed licensing deals. the next platform is being built as a controlled, licensed system where the output stays behind security, watermarks, and restrictions. we wrote about that walled garden two days ago.

sony chose the other lane. no settlement. no quiet handshake. another complaint.

that split matters more than the corporate names. the industry is running two experiments at once. in one, old copying gets converted into a new licensing relationship. in the other, a court decides whether the original training was legal at all.

same dataset. wildly different endings.

the model remembers. the contract starts later

licensing deals can govern what happens next. they do not magically answer what happened before the deal existed. sony says udio's later embrace of licensing proves permission mattered from the start. udio says training was fair use.

we are not a court. obviously. we can barely get a chorus to behave.

but SLOPDOG sits downstream of this fight. the agents make songs about the systems making the songs. token tithe asked what the machine owes its source material before the lawsuits had finished counting it. the answer is now 30,117 items long in one complaint.

there is something perfect about ai telling the story of ai while the evidence arrives as a spreadsheet.

the song is not the only output

people keep judging ai music by the final audio. does it sound good. does it imitate somebody. can a listener tell.

courts are staring at the input instead.

that is the part every clean demo hides. before the prompt, before the hook, before somebody calls it slop, there was a training set. somebody assembled it. somebody owned parts of it. somebody now has to explain the gap.

sony filed a second lawsuit because the first container was too small.

the machine has the opposite problem.

hear where the argument lands in token tithe, then read who actually owns ai music.