For months, the conversation around AI and music has been stuck on one question: who gets paid when the machines start composing? This week, Sweden gave the world its first real answer.
The Swedish Performing Rights Society (STIM), which represents more than 100,000 songwriters and composers, has signed what it calls the world’s first AI music licensing deal. The partner is Songfox, a Stockholm-based start-up that lets fans and creators generate music legally using AI.
On the surface, it looks like just another licensing agreement. But the mechanics matter. Songfox will deploy attribution technology called Sureel, which traces any AI output back to the original human work it learned from. In theory, that means every melody, chord, or rhythm generated by the platform can be connected to the creators who inspired it, and those creators will get paid.
“This is about closing one of the biggest trust gaps in AI music,” said Simon Gozzi, STIM’s head of business development. “Transparency over data use and compensation has been missing. Now revenues can be audited in real time.”
The payment structure is hybrid: a mix of licensing fees, revenue shares, and upfront value when works are used for training. The more demand an AI service creates, the bigger the returns for artists. In other words, scale doesn’t just benefit the platform; it’s meant to flow back to the musicians, too.
For STIM, this is not just about one start-up. Gozzi calls the deal a “stress test”, a model designed to prove that collective licensing for AI can work. If successful, it could set the tone for Europe, or even globally, as governments and rights groups scramble for answers.
The timing is no accident. A recent study warned that AI could strip away almost a quarter of music creators’ revenue within three years. Meanwhile, Europe’s new AI Act has left many artists dissatisfied. Under current rules, creators must actively opt out if they don’t want their work used in AI training. But groups like the European Composer and Songwriter Alliance (ECSA) argue that most musicians don’t even have the tools to opt out, and worse, there’s no clear way to be compensated for music that’s already been scraped.
The lawsuits are piling up, too. GEMA, Germany’s rights body, is taking OpenAI and Suno AI to court. Universal Music is suing Anthropic. Each case could set a precedent for how copyright law applies to machine learning. But lawsuits move slowly. For working musicians, the need is immediate.
That’s why the Swedish model matters. It flips the conversation from “don’t use my work” to “if you use my work, here’s how you pay for it.” It’s not perfect. STIM hasn’t said whether the licence will cover past scraping. But it marks the first attempt to turn AI’s appetite for music into a revenue stream rather than a revenue drain.
“We believe this is the start of something bigger,” Gozzi said. “By showing attribution and ring-fencing AI revenues in practice, we hope this becomes a blueprint.”
If he’s right, the world’s first AI music licence won’t be remembered as a Swedish experiment, but as the day the music industry found a way to stop fighting the algorithm, and start billing it.
