ElevenLabs adds synthetic voice detection to ElevenAgents, judging in the first seconds whether a caller is human or an AI agent
Synopsis
ElevenLabs has added synthetic voice detection to ElevenAgents, which analyzes a caller's speech in the first few seconds of a call and returns a human-or-AI-generated determination, after which the business's own rules decide what happens. Previously businesses had no way at call start to separate personal AI agents, business agents and cloned voices posing as customers. Detection runs on the live audio stream without depending on a watermark, and is available to enterprise customers supported by the Forward Deployed Engineering team.
Interpretation
A growing share of the calls reaching businesses and governments are not from people: personal AI agents running errands, business agents confirming orders or checking claims, and occasionally a cloned voice posing as a customer. ElevenLabs released synthetic voice detection in ElevenAgents so businesses can tell who is on the line at the start of a call and respond accordingly. Before this, businesses lacked a way at the opening of a call to distinguish a human caller from an AI agent, and so could not route calls by caller type. The detection analyzes the caller's speech in the first few seconds and returns a human or AI-generated determination, after which rules set by the business decide what happens.
Detection runs inside ElevenAgents on the live audio stream, checking for traces a human listener would not notice but software can. All audio generated with ElevenLabs carries a watermark, but audio from other sources often does not, so detection does not depend on one being present. Rules a business can set include serving verified human customers first with its most capable AI agent or a human representative, sending machine callers to an agent with its own instructions and limits that authenticates first and answers within scope, and stopping a synthetic voice that requests a sensitive action before it gets information or takes action.
Perspective
Synthetic voice detection is currently available to enterprise customers supported by ElevenLabs' Forward Deployed Engineering team, and later this month it will come to a broader group of enterprise customers as a configurable option inside ElevenAgents. ElevenLabs is also joining the Personal Agent Protocol working group led by Meta and Sierra as a design partner, and that protocol is still being developed.
The announcement does not give the detection's accuracy or its error behavior. The Personal Agent Protocol is still being developed, and ElevenLabs says it will add support for agents that declare themselves as such standards take shape and will keep adding signals.