What is an AI detector?

A short explanation of what AI detection is, how it works, and where the boundaries are.

What an AI detector is

A model that takes a piece of content (audio, image, video, or text) and returns a probability that it was produced by a generative AI rather than a human. Different content types use different detection approaches. We focus on audio, specifically voice.

How AI Voice Detector works

The detector ingests an audio file, extracts an acoustic fingerprint (spectrogram features, prosodic patterns, temporal statistics), and feeds it to a classifier trained on millions of pairs of real and synthesized voices. The classifier returns a probability between 0% and 100%.

We retrain the classifier monthly to keep up with new synthesis systems. The current methodology is v3.2, released April 2026.

Why you need an AI detector for voice

Three reasons. First, current synthesis quality has crossed the threshold where the human ear is unreliable. Second, the cost of a mistake (publishing a fake quote, wiring money on a fake call) is high. Third, you need a verdict you can cite, not just a hunch.

Acoustic vs linguistic detection

Some detectors look at what the voice is saying (linguistic patterns: word choice, sentence structure). Others look at how it sounds (acoustic patterns: spectral artifacts, prosodic regularity). Acoustic detection is harder to evade because attackers control the linguistic content but cannot fully control the acoustic fingerprint. Our approach is primarily acoustic.

Limits

No detector is perfect. We are wrong roughly 1% of the time on clean audio, and more often on heavily compressed audio. When we cannot tell, we report low confidence rather than guess. Use the verdict as evidence in a chain, not as a final word.

An AI detector measures the audio, not the speaker. The verdict is about whether the recording was generated by a model. That is the only question we answer.