Canada Wants $20,000 a Song From Suno. The Money Isn't the Point.

TL;DR

On September 2, SOCAN, the Canadian collecting society for more than 185,000 songwriters, composers and publishers, sued Suno in Federal Court. The claim documents 150 publicly available Suno outputs drawn from 137 songs in SOCAN's repertoire, and asks for CA$20,000 per work, CA$10 million in punitive damages, and a disgorgement of Suno's profits. That money is noise against a company that raised US$400 million in June at a US$5.4 billion valuation. The part that matters is the remedy SOCAN asks for last: a wide injunction covering every present and future work it controls.

What is actually in the filing

SOCAN alleges that Suno trained on effectively all readily accessible online music without permission, and that the platform now generates and streams outputs that reproduce works in SOCAN's repertoire. The claim frames this as infringement of performing rights and of the making-available right, not merely as an argument about training data.

SOCAN published its evidence rather than describing it. Its lawsuit page hosts five side-by-side comparisons: Tom Cochrane's "Life Is a Highway," Avril Lavigne's "Sk8er Boi," Alexisonfire's "Passing Out in America," K. Maro's "Femme Like U," and Daniel Balavoine's "S.O.S. d'un terrien en detresse." The full claim also names Joni Mitchell's "Both Sides Now," Celine Dion's "A New Day Has Come," and The Tragically Hip's "Bobcaygeon." The 150 cited outputs are explicitly described as a sample.

SOCAN CEO Jennifer Brown put the memorization problem in the plainest terms available to a non-lawyer, telling the Canadian Press that you will hear an AI output and already know all the words to it. SOCAN also confirms it has issued no licence to Suno and received no royalty payments from it.

Where liability attaches, and why that changed

The useful thing about this case is not the dollar figure. It is that the collecting societies have stopped arguing only about the input.

This matters because the input-side defence is the one Suno has spent two years building. If the claim is that the model hands back a recognisable "Sk8er Boi" and then streams it to Canadians, an argument about the legality of a 2023 scrape does not resolve it.

Canada does not have fair use

This is the detail most builders get wrong. The United States has fair use: an open-ended standard where a judge weighs four factors and can find almost any purpose transformative if the argument lands. Canada has fair dealing, which is a closed list of enumerated purposes such as research, private study, criticism, and news reporting.

Fair use is a discretionary door policy. Fair dealing is a guest list. If your purpose is not on the list, you never get to make the four-factor argument at all. Entertainment lawyer Matthew Gorman, quoted by the Canadian Press, noted that whether AI training fits inside any of those enumerated purposes is still legally untested in Canada.

So the American fair-use playbook does not travel north. It also did not travel east, which is the second thing that should worry anyone shipping a music model.

Three societies, three countries, one defendant

The Munich I Regional Court decided GEMA v Suno (case 42 O 763/25) on July 31, 2026, covering six musical works. It held that the works were reproduced twice: once inside the model parameters, and again in the generated outputs. That first finding is the one with teeth. If a memorised work inside the weights is itself a copy, then the model file is the infringing artifact, and deleting outputs does not cure it. The judgment is not final and remains appealable.

Denmark's Koda, which represents over 52,000 members, filed in Copenhagen City Court on November 4, 2025. That case is still running.

Note who is bringing these. Not labels, not individual artists. Collecting societies, which hold aggregated performing rights across an entire national repertoire and can therefore sue over a catalogue rather than a track list. That is a structurally different opponent from a single publisher, and it is why the requested remedy looks the way it does.

The money is a rounding error. Read the last paragraph of the prayer for relief.

Canada's statutory ceiling for commercial infringement is CA$20,000 per work, against US$150,000 for wilful infringement in the United States. Multiply the ceiling by 137 cited works and you get CA$2.74 million, roughly US$2 million. Add the CA$10 million punitive claim and the entire monetary ask is under CA$13 million. Suno raised more than thirty times that in a single June round.

So SOCAN is not really asking to be paid. It is asking for a switch. Alongside a permanent injunction on the named works, the claim seeks a wide injunction covering all present and future works in SOCAN's repertoire, plus an accounting of Suno's profits. A wide injunction is the remedy that does not scale with the size of the sample, which is exactly why you file with 150 examples instead of 150,000.

Suno did not respond to the Canadian Press by deadline on the day of filing. It has told Music Ally the claims are without merit, pointing to guardrails such as blocking artist-name prompts and screening uploaded files. Those guardrails are input filters. The claim is about output similarity, which is a different failure mode: nobody typed "Sk8er Boi."

What to do if you build on a music model

• Territory is now a product decision. A model that is defensible in a US fair-use posture can still be enjoined in Canada, Germany, or Denmark, where fair use does not exist. If you ship to those markets, your legal exposure is not your vendor's US case.
• Output similarity is the surface to test. Prompt filters do not catch regurgitation. If you are shipping generated music commercially, run outputs through similarity detection before release, not prompts through a blocklist.
• Licensed catalogues are not a nice-to-have. The Munich reasoning implies that a memorised work is a copy sitting in the weights. That is not something a downstream user can filter away.
• Injunctive risk beats damages risk. A CA$13 million judgment is survivable. A wide injunction against a national repertoire is a market exit.

What is not decided

All of it, technically. SOCAN filed a claim; nothing has been proven. The Munich judgment is a first-instance ruling under appeal and binds nobody in Canada. The Koda case has not been decided. Suno has not filed a defence in Canada. Every number above is what SOCAN is asking for, not what any court has ordered, and the CA$2.74 million figure assumes the statutory maximum on every one of the 137 works, which is not how these usually land.

What has changed is the shape of the argument. Three national collecting societies, in three legal systems with no fair use between them, are all now pointing at the output rather than the scrape. That is a harder thing to engineer around than a training-data disclosure.

Key Takeaways

• SOCAN sued Suno in Canada's Federal Court on September 2, 2026, citing 150 outputs drawn from 137 songs in its repertoire.
• The ask is CA$20,000 per work (CA$2.74 million total), CA$10 million punitive, Suno's profits, and a wide injunction over all present and future SOCAN works.
• Canada has fair dealing, a closed list of enumerated purposes, not open-ended US fair use. The American defence does not transfer.
• Munich's July 31, 2026 GEMA ruling held that works memorised inside model weights are themselves reproductions, and that the outputs infringe separately. It is under appeal.
• Suno is now defending against collecting societies in Germany, Denmark, and Canada, having raised US$400 million in June 2026 at a US$5.4 billion valuation.
• Suno's cited defences are input-side filters. The claims are output-side, which those filters do not address.

Sources: SOCAN lawsuit page, BNN Bloomberg / Canadian Press, Music Business Worldwide, Stereogum, Music Ally, Vossius on GEMA v Suno, Koda, MBW on Suno's Series D