ElevenLabs launches a tool to catch its own AI voices
ElevenLabs just rolled out the AI Speech Classifier, a web tool that lets anyone check if an audio clip was generated with its voice tech. Users can upload an audio or video file without logging in, and it checks the first minute to return a probability score plus closest voice matches. The catch: it doesn't reliably identify audio made with its Eleven v3 model yet.
AI-generated voices have gotten so convincing that it is genuinely hard to tell who is speaking anymore. Now, one of the biggest names behind that hyper-realistic audio is handing everyone a detector.
ElevenLabs just launched its AI Speech Classifier, a dedicated tool designed to verify whether an audio clip was generated using its own voice technology.
Drop in a file, no sign-up needed
The tool is intentionally frictionless to try. You do not need an account or even a login to use it. You simply upload an audio sample or a video clip directly into the classifier.
The system analyzes only the first minute of your uploaded sample. Within seconds, it returns a calculated probability indicating how likely it is that the audio originated from ElevenLabs AI.
Interestingly, if the tool detects that the voice was indeed AI-generated, it can cross-reference its library of over 10,000 voices and point you toward the closest matching voice model.
There is an explicit caveat, though. ElevenLabs noted that the classifier does not reliably classify audio generated with its Eleven v3 model, so older clips might still slip through the cracks undetected.
Why it matters
We think this is a necessary self-policing move from ElevenLabs. As deepfake audio risks mount, the creators of frontier voice models face immense pressure from regulators and the public to prevent misuse. Putting out a free, friction-free detector helps blunt that criticism.
For researchers, journalists, and moderators, it is a practical utility. When a fishy voice clip surfaces online, checking if it came from ElevenLabs takes less than a minute without hitting a paywall or sign-up form.
However, the approach exposes the fundamental limit of vendor-specific detectors. It only spots ElevenLabs audio, leaving deepfakes from every other model unchecked, and it even struggles with its own v3 generation.
It is a smart stopgap, but it underscores that single-company classification tools cannot solve the broader AI authentication challenge alone.
Sources (1)
- beta.elevenlabs.io(opens in a new tab)Originalbeta.elevenlabs.io