Traxsource will label tracks human-made or AI-assisted, and remove fully-AI music

Traxsource is about to put a label on every track it sells: human-made or AI-assisted. The dance-music download store starts rolling out the system on July 1, 2026, and will keep removing fully AI-generated music outright.

How Traxsource’s human-made and AI-assisted labels work

Traxsource has partnered with two AI detection companies, SH Labs and SoundPatrol, to classify the music submitted to it. The system flags fully AI-generated tracks for removal and uses the detection data to sort everything else as either AI-assisted or human-made. There’s a rebuttal process for content providers who think a track was misjudged.

The move follows the store’s February 2026 position statement, which accepted AI “as a production tool” inside “a larger human-driven creative process,” while drawing a hard line: entirely AI-generated music “does not belong on Traxsource.”

On paper, this is the spectrum approach I’ve argued for. A blanket “AI” sticker tells a listener nothing. Splitting human-made from AI-assisted, the way the SIQA charts already do, is the right direction.

Where the Traxsource classification gets blurry

The problem is that detection reads the finished file, not the process that made it. And the file doesn’t carry the workflow.

Take a track built in something like Suno Studio. You generate stems individually, you write the lyrics yourself, you sing the topline in your own voice. How does a detector know the voice is yours and not a clone? It can’t. Run that same vocal through a voice-cloning service and it would likely get flagged for removal, even if every creative decision was yours. The detector can’t see who wrote the lyrics, who arranged the stems, or how many hours went in.

That’s the gap. There’s a real difference between a 10-second text prompt and a track where a human made every meaningful choice but used AI tools to render it, the same way someone can spend hundreds of hours directing an AI-generated music video. Both can read as “AI” to a model that only hears the output.

What honest AI labeling would have to capture

The lines here stay genuinely blurry, and I won’t pretend this is easy. Labeling is better than a blanket ban, and Traxsource deserves credit for trying to classify rather than just delete.

But a label that judges the output and calls it a verdict on the process is going to misfire on exactly the cases that matter most: the human who used AI tools deliberately. Until the classification is matched to what actually happened in the creation workflow, not just what the audio sounds like, it’s a flawed system wearing a fair one’s clothes.