According to Reuters, music streaming service Deezer has licensed its AI-detection technology to France’s royalty agency, Sacem, in what it calls a landmark commercial deal. The company says it identified and removed up to 85% of fraudulent AI-generated music streams from its royalty pool in 2025, flagging over 13.4 million tracks. Deezer CEO Alexis Lanternier revealed the platform now gets about 60,000 fully AI-created tracks daily, which is roughly 39% of total uploads—a huge jump from just 10% in January of last year. The detection tool analyzes audio for patterns from AI generators like Suno and Udio, and has been trained on 94 million songs. Deezer has filed two patents for the tech and is now in talks with other European royalty societies.
The New Frontier of Streaming Fraud
Here’s the thing: this isn’t just about blocking bad songs. It’s about stopping a sophisticated new business model for fraud. Bad actors are flooding platforms with thousands of AI-generated tracks designed specifically to game algorithmic playlists and recommendation engines. The goal? To siphon off tiny royalty payments that add up to real money, directly stealing from human artists and songwriters. And with AI tracks making up nearly 40% of daily uploads on Deezer, the scale is already staggering. This is a systemic attack on the royalty pool itself, which Lanternier notes comprises 70% of subscriber revenue. So this fight is existential for the creative community.
Is Detection Enough?
But detection is a reactive game of whack-a-mole. That’s the criticism hinted at by Sweden’s royalty society, Stim. They told Reuters that tools like Deezer’s “alone cannot be an answer.” Their belief is that you have to prevent fraud at the source. Stim’s approach? Mandatory licensing and full transparency for the data used to train AI models. They actually launched a license in 2025 for AI companies to legally use copyrighted songs for training. It’s a fundamentally different philosophy: regulate the input rather than just policing the output. Which method wins out will shape the entire economics of AI music.
The Race to Sell the Solution
This Sacem deal is a huge proof of concept for Deezer. They’re not just using the tech internally anymore; they’re productizing it. They’re in talks with other collective societies in Europe and plan to pitch organizations in Los Angeles during Grammy Week. Think about it. Every label, every publisher, every streaming service is now terrified of this fraud vector. If Deezer can position itself as the trusted sheriff, this becomes a significant new revenue stream. But they won’t be alone for long. You can bet other tech and audio analysis firms are racing to build competing systems. The market for AI content authentication is about to explode.
A Fundamental Tension
This whole situation highlights the core tension in the AI era. On one side, you have tools that enable incredible creativity (and, unfortunately, fraud). On the other, you have the legacy systems of copyright and royalty distribution straining to adapt. Can technology and copyright really “go hand in hand” as Stim hopes? Or are we just putting band-aids on a broken system? Deezer’s detection tool is a necessary first line of defense. But the industry needs to figure out the rules of the road for AI training data, attribution, and compensation. Otherwise, we’re just building better mousetraps for an ever-growing number of mice.
