News · 2026-09-02

Canada's music rights society sued Suno and put 150 outputs in the filing

SOCAN, the organization that collects performance royalties for more than 200,000 Canadian songwriters, composers, and publishers, sued Suno on September 2, 2026. The claim is not primarily about what Suno trained on. It is about what Suno streams: SOCAN alleges the platform generates and publicly plays outputs that replicate songs in its repertoire, and the filing lists 150 specific publicly available Suno tracks as a sample of what it found.

Key facts

• Filed September 2, 2026 by SOCAN, announced from Toronto.
• The claim lists a sample of 150 publicly available Suno outputs SOCAN says copy works in its repertoire.
• SOCAN represents over 200,000 songwriter, composer, and publisher members.
• Primary source: SOCAN's press release.

The legal theory here is worth separating from the pile of AI copyright cases it will get filed alongside. Most of those cases ask whether training a model on copyrighted material is lawful, a question that turns on fair use in the United States and fair dealing in Canada, and that courts have been chewing on for three years without a clean answer. SOCAN is asking a narrower question with a much older body of law behind it: when a platform publicly plays a piece of music, does it need a licence?

Performing rights are the least glamorous and most settled corner of music copyright. Every radio station, bar, streaming service, and shopping mall in Canada pays SOCAN for the right to play music in public, and SOCAN distributes that money to the people who wrote it. The organization has been doing this for over a century, and the French-language version of its release carries the same allegations. Its argument against Suno is that if the platform generates a track that reproduces a song in its repertoire and then streams that track to listeners, the streaming is a public performance, and no licence was obtained for it.

That framing sidesteps the hardest question in AI copyright. You do not need a court to rule on whether training is fair use to rule on whether streaming a copy is infringement. It is the difference between arguing about how a photocopier works and arguing about what came out of it.

"SOCAN has a responsibility to act when the rights of music creators and publishers are put at risk," said Jennifer Brown, SOCAN's chief executive. "The evidence shows that the Suno platform has generated and streamed outputs that copy works in our repertoire, and that cannot go unchallenged." Andrea Kokonis, the organization's chief legal officer, put the objective more precisely: "This case is fundamentally about ensuring that long-standing copyright principles continue to apply in the AI era."

The 150-output sample is the strategically important detail. Music copyright cases have historically foundered on proof, because showing that a new song copies an old one requires expert musicological analysis and a court willing to draw a line between influence and reproduction. SOCAN is not asking a court to assess a vibe. It has 150 specific artifacts, publicly available, that it says are identical or similar to identified works. The organization also says it expects additional unauthorized outputs to come to light as litigation proceeds, which reads as an invitation for its members to keep sending examples.

This is not the first evidentiary problem Suno has had with outputs specifically. A Munich court found earlier this year that Suno had memorised six songs, a ruling about the model's behaviour rather than its training data. Memorisation is the technical phenomenon underneath both cases: a generative model trained on enough copies of a popular song can reproduce recognizable pieces of it on demand, not because it stored a file but because the pattern is heavily overrepresented in what it learned. It is the same mechanism that lets a language model recite a famous poem, and it is one of the better-documented failure modes in the field.

Why this matters beyond music: output-side claims are a route around the training-data question that every AI company has been defending against, and they generalize. If a platform can be held liable for publicly distributing outputs that reproduce protected works, then the relevant compliance question shifts from what did you train on to what are you shipping, and that is a question with existing technical answers, including output filtering and similarity detection. It is a substantially worse outcome for AI companies than a training-data ruling, because it applies continuously rather than once.

The honest caveat is that this is a filed claim, not a finding. SOCAN's characterization of Suno's business, including its assertion that Suno trained on virtually all readily accessible music on the internet without licences, is an allegation that has not been tested in this proceeding. Suno has not responded publicly to the filing. Canadian fair dealing is also narrower than American fair use in some respects and broader in others, so the outcome will not map cleanly onto the United States cases running in parallel. What is settled is that the plaintiff here is not a startup or a class of individual artists. It is the institution that has licensed public performance in Canada since 1925, and it brought 150 exhibits.

Primary source, verified: read the paper →

Ground Truth. (2026, September 2). Canada's music rights society sued Suno and put 150 outputs in the filing. Ground Truth. https://groundtruth.day/news/canadas-music-rights-society-sued-suno-over-150-outputs.html

@misc{groundtruth:canadas-music-rights-society-sued-suno-over-150-outputs,
  title  = {Canada's music rights society sued Suno and put 150 outputs in the filing},
  author = {{Ground Truth}},
  year   = {2026},
  month  = {sep},
  url    = {https://groundtruth.day/news/canadas-music-rights-society-sued-suno-over-150-outputs.html}
}

Topics: copyright · legal · music · suno · generative-ai · canada