Sierra launches fleming-1 to identify AI callers on phone calls

Sierra has launched fleming-1, a model designed to detect when an AI agent is calling a business by phone. It analyzes a caller’s speech while the call is underway, scoring audio for signs that it was generated by AI. Sierra describes the result as a signal for companies, rather than a decision about whether a call should be accepted.

The launch addresses a change Sierra says it has seen in its own work: agents built on its technology called other agents in healthcare last year. Sierra also points to the growth of personal agents such as Instinct and Muse, saying more calls to companies could come from AI acting for consumers. The company announced Personal Agent Protocol, an open standard it is developing with industry partners, the day before its fleming-1 announcement. The protocol is intended to let personal agents identify themselves and the people they represent, but Sierra says some agents may not use it.

Sierra says it is becoming harder to distinguish synthetic voices from human ones over phone lines. Background noise, unclear audio and poor connections can hide clues, while AI voices continue to improve. fleming-1 looks for signs of generated speech beyond what a human listener might notice, with the aim of identifying likely synthetic voices during the call.

The model flags calls it judges likely to involve AI and is designed to err on the side of caution to avoid mistakenly labeling real people. Sierra says each business can choose how to respond to a flag. For example, a bank could add a verification step when it detects an agent caller. A company handling many calls could instead begin by tracking how often such calls occur. For companies receiving voice traffic, the model provides a real-time indication that can inform their handling decisions without prescribing a single response.

Sierra says fleming-1 works with any voice agent built on its platform, and that businesses need to turn it on. The announcement positions fleming-1 as a fallback for calls where an agent does not identify itself through Personal Agent Protocol. For AI builders, that means the product is aimed at identifying likely agent callers during a phone conversation, while the business retains control over what happens next.