Good morning producers 👋🏼

Today is a very special day, because it marks the 1-year anniversary that I (John) have been co-founder of time off audio! Woo! 🍾 Well… maybe it's just special for me. But as I've been researching and writing this newsletter month after month, I've come to really appreciate the network of audio engineers and producers who are not only interested in our plugins, but in being part of the greater conversation in the music industry.

This month we've only got 1 big story that we're going to unpack. You're in for a treat.

Thanks for taking the time (off) ;)

Suno is caught with copyright infringement

When I pitched this story to Moe, he kind of looked at me confused like "...yeah? Obviously." But the interesting thing about Suno, and a lot of AI models for that matter, is that it's actually very difficult to prove what material it's being trained on.

However, on July 31st, 2026, GEMA did just that: they managed to get Suno to spit out near-exact replicas of several different songs. GEMA is the German equivalent of ASCAP in the US and SOCAN in Canada. They are the performance rights and collecting organization that protects music copyrights and collects licensing fees for performances and reproductions. How they did it is beyond me (believe me – I actually made an account to try to recreate their findings and got copyright flags every single time), but they did. The lawsuit was filed in January 2025, so perhaps the guardrails were not as prevalent back then, or perhaps they used some sort of "prompt injection" – using specific inputs to trick a language model into ignoring its original rules/safety guardrails.

According to the Munich Regional Court's own press release:

When generating the outputs in question, the plaintiff entered the original lyrics of the musical piece, the desired musical style, and the work title into the input form of the music generator. The prompts contained no specifications regarding melody, harmony, rhythm, or arrangement.

This is important because, at the end of the day, Suno operates on the same architecture as a large language model. LLMs are, in essence, prediction machines that output the most likely next word by calculating its probability based on patterns from its data set.

For example, if you type "Finish this sentence: Happy Birthday to ______", it will fill in the word "you", because across all its data points, "you" is consistently the most common word following that sentence.

Now, if you are a little more nuanced with the prompt and write "Finish this sentence: Mr. and Mrs. Dursley, of number four, Privet Drive, were proud to say that they were perfectly normal _______", the AI will respond with "thank you very much". Notice that I did not specify the number of words to finish with, and yet the AI will spit out the next 4 words completing the first sentence of the first Harry Potter book. I encourage you to try it out for yourself with different types of literature if you have not already. It really exemplifies how LLMs function.

Once you do that, it becomes increasingly obvious that the LLM is trained on that specific copyrighted material. If it wasn't, there's no way those specific set of words would be the most likely output across all models. What GEMA achieved was the exact same thing, except with Suno's generative models.

I highly recommend you go on their page and listen to the side-by-side comparisons of GEMA's AI outputs vs the originals. The one that is most blatant, in my opinion, is the song "Forever Young" by Alphaville. "Mambo Number 5" is also quite interesting for a different reason (the intro and inexplicable key change?). "Rasputin" also has its quirks – the chorus seems to almost mimic what a backing vocal would be rather than the original lead. The songs are not exact copies, but they're close to the point that it is undeniable that the models are being trained on it – the same way it is undeniable that models are training on Harry Potter.

So anyways, where does that leave us? GEMA won the settlement, but the judgment is not final, and Suno will likely appeal, at which point this would go to the Munich Court of Appeal. GEMA also won a case against OpenAI last year, and OpenAI did the same thing: appealed. And that decision is still pending. In the meantime, Suno still exists in Germany, but the court has issued an injunction, a ban on further training use, a disclosure of revenue, and damages (TBD).

GEMA has a very interesting approach to this whole AI thing. It seems like their primary goal is to get these various AI companies to use GEMA's proposed licensing model for generative artificial intelligence, the goal being that when models use copyrighted work, artists are compensated fairly.

This is very different from how most AI companies have been operating thus far. In order to train their models, these companies have been pleading "fair use". Fair use is a statutory exemption that lets you use copyrighted material without permission. It's usually used for criticism, commentary, news reporting, teaching, parody, and research. AI companies are leaning on that "research" category to defend their model's data training on copyrighted material.

Personally, I like GEMA's approach because it's grounded in the reality that this technology is here to stay and is a "new form of music utilization." In tandem, it seeks to ensure artists are compensated accordingly for the work that these models are training on.

Generative AI presents new opportunities for revenue generation. The foundation of corresponding business models is built on artistic, creative, and journalistic content. A fair remuneration model starts where value is created.

You can check out their licensing model for Gen AI here.

I know that Gen AI is a very hot topic in our industry, but for those of you who actually read this newsletter to the end, I'm curious to know: if royalties are paid out to artists who opt in to a licensing model, how does that affect your position of Gen AI in our industry? Please reply to this newsletter and let me know your thoughts.

Have a great month of September!

Honourable mentions

• Suno signs new deal with BMG
• Dr. Dre says he is "embracing AI in music"
• Sony Music and Warner Chappell sue Anthropic

Thanks for taking the time, and as always, feel free to reach out via email or Discord for any questions or feedback!