FR

Can an AI music detector be 100% reliable?

By Sébastien Necca · 11 October 2026

It is the first question people ask: “is it certain?”. The honest answer is no: no AI music detector can be 100% reliable. This is not a flaw of one particular tool, it is the very nature of the problem. Here is why, and how to use a detector wisely all the same.

A never-ending race

A detector learns to recognise the traces left by the generators it has studied. When a new version comes out, these traces change. The detector must then be re-evaluated, sometimes retrained. Meanwhile, its accuracy drops on the most recent tracks. It is the same race as for antivirus software or the detection of generated images.

The two possible errors

Reducing one of these errors increases the other. Every detector makes a trade-off, and we chose to favour caution.

What blurs the detection

How to read a score

ScoreWhat it means
below 20%Nothing suspicious detected. It does not prove a total absence of AI.
20 to 35%A few weak clues. Do not over-interpret.
35 to 70%Clear clues. Listen to the flagged passages and look for other evidence.
above 70%High probability of AI generation on the element concerned.

Good practice

Use a detector as a first filter, never as a judge. Cross-check the result with listening, an investigation into the artist and, if the stakes are high, with proof of creation (session files, separate stems). That is the spirit of our tool: give you clear, located clues, and tell you frankly what it cannot do. The details are on the Reliability page.

A track to check?

Upload it and get a verdict in 1 to 2 minutes, for free.

Analyse a track