Remember when the internet was convinced that AI-generated images could be spotted by simply looking for six-fingered hands? It was a brief, triumphant moment for the human eye before the models learned how to count. Now we are seeing a repeat of that same hubris in the audio space. Deezer is attempting a similar feat with sound. According to TechCrunch, the streaming service has launched a tool that scans playlists across competing platforms like Spotify and Apple Music to flag AI-generated tracks. It is a strange strategic play. Usually, smaller players try to steal users by offering better discovery algorithms or lower prices, not by acting as the digital purity police for the rest of the industry (which is a bold claim for a company that consistently struggles for market share against the giants).
There is also the immediate problem of real-world friction. Scanning external playlists isn’t as simple as hitting a “detect” button; it requires dealing with fragile, restrictive APIs and the constant threat of being throttled by the very platforms Deezer is trying to “clean up.” If Spotify decides they don’t like Deezer playing detective on their proprietary playlists, they can simply cut off the pipe or change the data structure overnight. The technical overhead of maintaining cross-platform compatibility for a tool that essentially tells you “this song might be fake” seems like a poor use of engineering hours. Who is actually going to use this? The average listener isn’t spending their afternoon importing a Spotify playlist into Deezer just to see which lo-fi beats were made by a GPU.
Let’s be real: audio detection is a losing game. The gap between synthetic audio and organic recording is closing faster than the gap between a junior dev and a senior who knows how to use a cursor. If you can train a model to generate a convincing vocal, you can train a second model to remove the very artifacts the detector is looking for. It is a classic cat-and-mouse game, but the mouse has a jetpack. Trying to build a definitive “AI or Human” filter is like trying to spot a fake Rolex by looking at it through a window from across the street—you might catch the obvious fakes, but the high-end stuff is invisible to the naked eye. It is essentially trying to fight a tide with a plastic bucket.
Do we really care if a “Rainy Night in Tokyo” lo-fi beat was crafted by a human in a bedroom or a GPU in a warehouse? For most listeners, the utility is the same. The only people who care are the copyright lawyers and the purists who think art requires human suffering. Since we already accept autotune, quantized drums, and heavily processed vocals, the line between “assisted” and “generated” is a smudge. We’ve spent twenty years normalizing digital perfection; now we’re surprised when a machine can mimic it perfectly. (Or maybe we’re just terrified that the “soul” we attributed to music was actually just a specific set of frequencies and timing errors).
This move by Deezer also ignores the economic reality of the streaming business. Labels are already experimenting with AI to fill gaps in their catalogs and reduce royalty payouts to humans. By positioning themselves as the “anti-AI” platform, Deezer is picking a fight with the very industry that feeds them. It is a gamble that assumes users value “humanity” over “vibe.” But if the vibe is correct, the origin is irrelevant. We see this in the food industry all the time—people love the taste of a burger regardless of whether it was flipped by a person or a machine, as long as it tastes like a burger.
By Q4, these detection tools will be entirely obsolete. The next wave of multimodal audio models will produce waveforms that are mathematically indistinguishable from organic recordings, making this entire exercise a waste of compute. The “fingerprints” Deezer is looking for are already evaporating. Once the generation becomes indistinguishable at the sample level, the only way to prove a song is human is to film the artist in a room with a guitar and a microphone, and that is a logistics problem, not a software one. We are moving toward a world where “proof of personhood” requires physical evidence, not digital analysis.
A futile exercise in vanity.
