AI music is no longer just a debate about the future of music. It is now a legal battle over who owns the past, and who gets paid for it.
On September 2, 2026, Canada’s largest music rights organisation, SOCAN, filed a copyright lawsuit against AI music company Suno, alleging that Suno’s platform has generated and publicly streamed AI outputs that reproduce songs represented by SOCAN without permission or compensation.
The case is significant for a reason that goes beyond another lawsuit against an AI company.
Earlier legal actions against Suno have largely focused on how copyrighted music was allegedly used to train AI models. SOCAN’s case goes further by placing the spotlight on what those models are producing and making available to users.
SOCAN says it has identified 150 publicly available Suno outputs that are identical or substantially similar to songs in its repertoire.
And this comes just weeks after a German court delivered another major setback for Suno.
On July 31, 2026, the Munich Regional Court largely ruled in favour of GEMA, finding that Suno had infringed copyright in connection with six protected musical works, including songs such as Atemlos durch die Nacht, Rasputin, Big in Japan, Forever Young and Daddy Cool. The court addressed both the reproduction of works for AI training and their reproduction within the AI model, as well as certain uses involving generated outputs.
Together, these developments point toward a much bigger question:
Can an AI music company build a commercial music-generation system using copyrighted music without licensing it—and then generate outputs that compete with or reproduce that music?
The answer is increasingly being tested in courts around the world.
What exactly is SOCAN accusing Suno of doing?
SOCAN’s lawsuit is different from simply arguing that AI models learn from copyrighted music.
According to SOCAN, Suno has made available and streamed AI-generated outputs that reproduce songs represented by the organisation.
SOCAN’s legal action identifies a sample of 150 publicly available outputs. The organisation says these outputs represent only a portion of the activity it believes is infringing.
SOCAN has also published several examples comparing original songs with Suno-generated outputs, including:
* Tom Cochrane — Life is a Highway
* Avril Lavigne — Sk8er Boi
* K. Maro — Femme Like U
* Alexisonfire — Passing Out in America
* Daniel Balavoine — S.O.S. d’un terrien en détresse
The organisation also points to Joni Mitchell’s Both Sides Now in its statement of claim, alleging that a Suno output was virtually identical in melody, harmony and lyrics after a user supplied only the lyrics.
In the case of Sk8er Boi, SOCAN alleges that the generated track retained substantial musical similarities while changing some lyrics into Korean.
These examples matter because they move the discussion away from an abstract question“Can AI learn music?”toward a much more concrete one:
What happens when an AI system produces something that is recognisably the same protected musical work?
SOCAN is seeking potentially millions of dollars
The financial stakes are substantial.
SOCAN says it is seeking damages and profits resulting from the alleged infringement, with the final amount to be determined at trial.
Alternatively, it seeks statutory damages of up to CAD $20,000 per infringed song for commercial infringement.
Based on the 150 cited outputs, which SOCAN says relate to 137 songs represented by it, the organisation calculates that statutory damages could reach approximately CAD $2.74 million.
SOCAN is separately seeking CAD $10 million in punitive and exemplary damages, alleging that Suno’s infringement was wilful and knowing.
Importantly, these are claims made by SOCAN, not damages already awarded by a court.
The case is still ongoing.
SOCAN is also asking for injunctions that would prevent Suno from infringing the works identified in the lawsuit and, more broadly, works represented by SOCAN.
That could make the case considerably more important than a dispute over a fixed amount of money.
The bigger issue: training vs. output
One of the most important things to understand about the AI-music copyright debate is that there are actually several different legal questions.
1. Was copyrighted music used to train the model?
Generative AI systems require large datasets for training.
The U.S. Copyright Office’s 2025 report on generative AI training explains that AI development can involve multiple stages at which copyrighted works may be copied or otherwise implicated by copyright law. The Office specifically examined data collection, training, memorisation and outputs.
2. Is that training legally permissible?
This is where the debate around fair use becomes important in the United States.
And there is a common misconception here.
The U.S. Copyright Office has not said that AI companies automatically have the right to train on copyrighted music without permission.
Its Part 3 report instead analysed the existing fair-use framework and the factors that courts may consider.
The Office noted that unlicensed AI training can create significant market harm. In particular, where a model produces substantially similar outputs that directly substitute for works in the training data, it can potentially result in lost sales. The report also discusses broader market dilution and lost licensing opportunities.
The Office also recognised that voluntary licensing is already emerging in some sectors and could be feasible for certain types of works, training and models.
So the legal position is much more nuanced than:
“AI companies can train on anything because it is fair use.”
There is no blanket rule saying that.
Then comes the third question: What does the AI actually generate?
This is where the recent European and Canadian developments become particularly important.
The Munich Regional Court’s July 2026 judgment against Suno dealt with six musical works.
According to the official court statement, the court largely upheld GEMA’s claims for injunctive relief, information and damages. The case concerned allegations that the works had been reproduced during training and memorised within the AI model.
The court’s judgment also addressed reproduction within the model and certain reproduction and public-use issues involving generated music.
This creates a significant distinction.
An AI company might argue:
“The model doesn’t store songs like a traditional music library.”
But if a protected musical work can nevertheless be reproduced or recognisably recovered through the model, courts may have to examine whether the model’s internal representations and resulting outputs implicate copyright.
That is precisely the kind of question that is now being tested in court.
GEMA vs. Suno: a major warning sign for AI music companies
GEMA’s case is particularly important because it resulted in an actual court judgment rather than merely an allegation.
On July 31, 2026, the Munich Regional Court largely ruled for GEMA.
The official court announcement states that the judgment concerned six musical works and that GEMA had argued that these works were reproduced during training and memorised in the AI model. The court granted claims concerning injunctive relief, information and damages.
GEMA described the decision as a landmark for international copyright law, arguing that AI companies must obtain licences and remunerate creators when protected works are used.
The judgment is significant but it is also important not to overstate it.
It is a first-instance decision and is not the same thing as a worldwide ruling that all AI training is illegal.
Different countries have different copyright laws, and other courts can reach different conclusions.
But the ruling demonstrates that AI training is no longer being treated as a purely theoretical copyright issue.
Suno’s position has also been changing
Suno has historically defended the idea that AI systems learn patterns from music rather than simply copying individual songs.
In a 2024 statement responding to the RIAA lawsuit, Suno argued that its models learned from music available on the open internet and compared AI learning to how humans learn musical styles and patterns. Suno said that learning from existing material is not inherently infringement and argued that its system was designed to generate original music.
But Suno’s strategy and business relationships have evolved significantly since then.
In August 2026, Suno announced a global partnership with BMG. According to BMG, the agreement covers recorded music and publishing repertoire and forms part of Suno’s upcoming music model developed in partnership with the music industry.
Crucially, BMG says participating artists and songwriters will have their rights protected and be compensated, while the agreement also settles prior use of BMG recordings and publishing works.
That represents a very different model from simply sourcing huge amounts of music and arguing about whether permission was necessary afterward.
It points toward a future in which licensed AI training becomes a commercial product in its own right.
The industry is moving toward licensed AI
This shift is happening beyond BMG.
IFPI’s 2026 Global Music Report says global recorded music revenue reached US$31.7 billion in 2025, growing 6.4% year over year.
Paid subscription streaming alone generated 52.4% of global recorded music revenue, with 837 million paid subscription accounts worldwide.
At the same time, IFPI says record companies are actively engaging with AI developers to develop licensed business models that create revenue opportunities for artists.
The objective is increasingly becoming:
AI + licensed music + human creativity = a sustainable music ecosystem.
IFPI has also introduced principles distinguishing human-led creativity from purely synthetic or unauthorised AI-generated recordings, while supporting transparency around AI use in music.
In other words, the music industry’s position is not simply “AI is bad.”
It is increasingly:
AI can innovate but rights need to be respected.
The economic question may be even bigger than the legal one
Copyright disputes often sound technical until you look at the money involved.
CISAC’s global study on AI and creative industries estimates that generative AI music outputs could represent a €16 billion annual market by 2028.
But under current conditions, the study estimates that 24% of music creators’ revenues could be at risk by 2028, representing approximately €4 billion in annual revenue at risk.
That doesn’t mean every AI-generated song directly takes money away from a specific musician.
The issue is substitution.
If AI-generated tracks increasingly occupy playlists, libraries, advertising, background music, social media and other commercial uses, they can compete with human-created music for the same listening time and commercial opportunities.
And that is why the question of licensing matters.
If AI companies generate billions of dollars from music-related products while the musicians whose work helped make those systems possible receive nothing, the economic model becomes difficult to justify.
SOCAN’s case introduces another important principle: Authorization, Remuneration and Transparency
SOCAN has framed its position around three principles:
Authorization.
Remuneration.
Transparency.
The organisation argues that copyrighted music should not be used by generative AI systems without authorization, creators should be compensated, and AI companies should be transparent about how protected works are being used.
These principles are increasingly echoed across the wider copyright debate.
CISAC similarly identifies authorization, remuneration and transparency as three core principles for AI policy.
The importance of transparency is obvious.
If an AI company does not disclose what music went into training a model, rights holders may not even know that their work has been used.
And without knowing what was used, licensing becomes extremely difficult.
What does this mean for musicians?
For musicians, producers and songwriters, the AI debate is not just about whether AI can make a good song.
It is about who controls the underlying rights and who participates in the economic value created by technology.
There are two very different futures.
The first future
AI companies scrape huge quantities of music.
Models are trained without permission.
AI outputs compete with human-made music.
Rights holders discover similarities after the fact.
And musicians are forced to fight through individual lawsuits to protect their work.
The second future
AI companies work directly with musicians, songwriters, publishers and labels.
Training data is licensed.
Rights are documented.
Creators are compensated.
AI-generated music is labelled transparently.
And musicians can choose whether and how their work participates in the AI ecosystem.
The recent BMG-Suno partnership demonstrates that the second model is commercially possible.
The future of music may not be AI vs. humans
That is ultimately the most important takeaway.
The music industry has adapted to technological disruption many times before from physical formats to digital downloads, streaming, social platforms and now generative AI.
The question is not whether technology will change music.
It will.
The real question is whether the people whose music powers the ecosystem will participate in the value created by that change.
The SOCAN lawsuit, the GEMA judgment and the growing number of licensing partnerships suggest that this question is moving from boardrooms and policy discussions into courtrooms.
And the legal landscape is still being written.
For AI music companies, the message is becoming clearer: building a powerful model may require more than technical innovation.
It may require rights infrastructure.
For musicians, the message is equally important: your catalogue, compositions, recordings, lyrics, melodies and creative identity are not simply raw material for the next technology company.
They are the foundation of the music economy.
And as AI becomes a bigger part of music creation, discovery and consumption, the industry will increasingly have to answer one fundamental question:
If AI is going to make money from music, who should get paid for the music that made AI possible?
What this means for Beat22
At Beat22, the future of music starts with real, original music made by human musicians.
As AI changes how music is created and discovered, one thing becomes even more valuable: knowing where the music came from, who made it and what rights come with it.
That is why building better discovery and licensing infrastructure matters.
The future does not have to be AI versus musicians.
It can be technology that helps musicians get discovered, helps artists find the right music faster, and helps original music remain valuable in an increasingly automated industry. Explore original human made beats on Beat22.com.
