Sierra launches Fleming-1 to detect AI agents calling by phone – Unite.AI

Sierra launched Fleming-1 on October 7, 2026, a model designed to detect when the caller on a phone call is another AI agent. The company described the release as an update to the constellation of models behind the agents built on its platform.

Sierra says the question businesses face when the phone rings has shifted from who’s calling to what they’re calling. The company first encountered the issue in 2025, when Sierra-based agents began calling Sierra-based agents in the healthcare industry. As personal agents like Muse and Instinct take off, Sierra said, a large portion of the calls businesses receive could come from artificial intelligence acting on behalf of consumers.

How Fleming-1 detects agent callers

According to Sierra, fleming-1 analyzes a caller’s speech in real time, evaluating the audio for any signs that it was generated by artificial intelligence. The model goes beyond how a voice sounds to a human listener, evaluating whether the voice on the other end is synthetic while the call is still ongoing.

Sierra said distinguishing a synthetic voice from a human one on the phone is becoming increasingly difficult. Background noise, muffled audio, and a bad connection can obscure clues, while the AI ​​voices continue to improve.

Sierra flags calls that Fleming-1 identifies as likely AI. The company said the model is optimized to be conservative by default, so that real people aren’t inadvertently flagged and the company decides what happens next.

Caller ID comparison

Sierra framed the model through the history of Caller ID, which BellSouth launched in Memphis, Tennessee, in 1988. Caller ID never tells you whether to answer, and Sierra says fleming-1 works the same way: It’s up to companies to decide how they want to handle such calls.

The company said its starting principle is that companies should know what it is calling for. As an example of what companies could do with such information, Sierra said a bank might want to add a verification step when the caller is an agent, and a company with a high call volume could start by measuring how often this happens.

Sierra said some robocallers will be scammers testing account security on a large scale, while others will be customers who have given a task to an agent or people who rely on text-to-speech.

Sierra said caller ID has ended up on all phones because people find it useful, and he believes the same will be true for knowing when an agent is calling.

Relationship with personal agent protocol

The launch came one day after Sierra announced the Personal Agent Protocol on October 6, 2026, an open standard that Sierra and Meta are developing with industry partners Genesys, Instinct, Rocket, Shopify, Stripe and Walmart that defines how personal agents interact with businesses. When an agent uses the protocol, the company knows who they are an agent and who they are acting for. Sierra noted that some officers may not follow protocol.

Sierra described the best case as an agent saying who they are, what the protocol is for, with fleming-1 covering calls where that doesn’t happen. Sierra plans to publish the protocol’s v0.1 specification later in October 2026, host design workshops with stakeholders, and publish a reference implementation to help developers get started.

The voice and architecture of the constellations

fleming-1 applies to a voice channel announced by Sierra on October 9, 2024, when the company said its AI agents can answer the phone. Sierra said these agents integrate with existing call center platforms, can sit in front of or behind traditional IVR systems, and hand over to human teams with AI-generated summaries when a situation requires human intervention. Sierra’s Agent OS allows a company to create an agent once and deploy it across all channels, including chat and phone.

The model also joins a Sierra architecture detailed in a December 3, 2025 post: Sierra-based agents are assembled from more than 15 frontier, open-weight, and proprietary models, each selected for the job at hand. Sierra said it performs asset-specific assessments and invests in refined models where standard models fail to meet its constraints.

Sierra said fleming-1 works with any Sierra-based voice agent and only requires the company to turn it on.