ElevenLabs has added synthetic-voice detection to ElevenAgents, giving businesses a signal that a caller may be an AI before allowing sensitive actions. The check runs on the live audio stream during the first few seconds and returns a human-or-generated determination.
Customers can use that result in routing rules. A business might send verified people to its most capable agent or a human representative, while directing automated callers to a bounded workflow that authenticates before discussing an account. Requests such as password resets or bank-detail changes can be blocked when a synthetic voice triggers the policy.
The detector does not rely only on ElevenLabs’ watermark, because calls may come from other voice generators. The company says it also examines acoustic traces that listeners may not notice. It has not published an error rate in the announcement, so the classification should be treated as one security signal rather than proof of identity.
ElevenLabs is also joining the Meta- and Sierra-led Personal Agent Protocol group, which is working on how agents identify themselves and state whom they represent. Detection is currently available through ElevenLabs’ engineering-supported enterprise program, with broader enterprise configuration planned later this month.