In 1988, BellSouth launched Caller ID in Memphis, Tennessee. It answered the question everyone asks when the phone rings: I wonder who that could be?
38 years later, we’re facing a different question: what is calling? We first encountered this issue last year, when agents built on Sierra started calling agents built on Sierra in healthcare. But with personal agents like Muse and Instinct taking off, a large share of the calls companies receive could come from AI acting on behalf of consumers.
Yesterday, we announced Personal Agent Protocol, an open standard that we’re working to develop with industry partners. It gives personal agents a direct, authorized way to work with businesses. When an agent uses the protocol, the company knows it’s an agent and who it’s acting for. Of course, some agents may not respect the protocol.
So today we’re updating our constellation of models with the launch of fleming-1, which detects when the caller on a phone call is another agent.
Information, not judgement #
Caller ID never told you whether to pick up. fleming-1 works the same way: it’s for companies to decide how they want to handle these calls. Yes, some automated callers will be fraudsters testing account security at scale. Others will be customers who handed off a chore to an agent, or people who rely on text-to-speech.
We start from a simple principle: companies should know what is calling. A bank might want to add a verification step when the caller is an agent. A company with high call volume might start by measuring how often it happens.
How fleming-1 works #
It’s getting harder to tell a synthetic voice from a human one over the phone. Background noise, muffled audio and a bad connection can obscure the clues, while AI voices keep getting better.
fleming-1 analyzes callers’ speech in real time, scoring the audio for signs that it was generated by AI. By looking beyond how a voice sounds to a human listener, the model helps teams assess whether the voice on the other end is synthetic—while the call is still happening.
We flag the calls fleming-1 identifies as likely AI. The model is tuned to be conservative by default, so real people don’t inadvertently get flagged. Your business can decide what happens next.
The next caller ID #
Caller ID ended up on every phone because people found it useful. We think the same will be true of knowing when an agent is calling.
The best case is an agent that says who it is, and that’s what Personal Agent Protocol is for. For the calls where it doesn’t, there’s fleming-1.
fleming-1 works with any voice agent built on Sierra. You just need to turn it on.