Liberate Research
November 20, 2026

The Future of AI in Insurance Isn’t the Phone Call

Liberate Research Team

Personal AI agents like Meta Muse, ChatGPT dots and Claude Duet already call businesses for their users to compare prices, request quotes and check on claims. Insurers are on that call list, and most can't tell when the caller isn't a person.

That makes the AI agent a new kind of customer, and a new way to win business. In a Liberate poll of 1,000 US policyholders this October, 54% said they would consider switching to an insurer that works with their AI. Carriers that can recognize these agents, answer them in seconds and hold them to the same rules as people will win that business. Carriers that can't will either tie up licensed staff talking to machines or turn away real customers.

That's why we built Argus 1, a governed suite of models and products for a world where AI already talks to AI. It starts on the phone today and is built for the direct agent-to-agent connections coming next.

The caller has changed

Consumer AI assistants can already place a phone call and hold a natural conversation. Ask one to shop for home insurance and it won't stop at one carrier. It can call fifty at once and ask each of them for a quote.

Every one of those calls lands in a queue staffed by people who were trained to talk to people. The CSR runs the standard script, delivers the standard pitch and answers questions that could have been pulled from a structured database in a fraction of a second. Meanwhile, the policyholder with a burst pipe is waiting on hold.

Carriers have two obvious responses, and neither one scales:

  1. Hire more CSRs to absorb volume that keeps growing.
  2. Let wait times grow and push customers with real needs to the back of the line.

There's a third problem underneath both. Without detection, AI traffic shows up in your reports as ordinary calls, and you can't plan staffing around a share of volume you can't see.

The new caller also behaves differently. An assistant calls at 2 a.m. as readily as at 2 p.m. It has no brand loyalty, and it won't sit through a pitch before deciding. But it does have a legitimate intent, such as a quote, a claim status or a policy question. That intent deserves to be fulfilled, and there's still business to win on the other end of it.

Introducing Argus 1

Argus 1 is a suite of models and products that lets carriers recognize the AI agents contacting them, serve their requests and hold them to the same rules as people.

The first product in the suite is AI Intercept. It answers the first question every call has to answer: who, or what, am I talking to?

AI Intercept listens to the opening seconds of a live call and looks at three kinds of signal:

  • How the caller sounds: the acoustic features of the voice itself.
  • How they compose what they say: the linguistic makeup of each utterance.
  • The rhythm of the exchange: cadence and turn-taking, which differ between people and machines.

Those signals combine into a single score, and the call is routed on it. A person gets a natural, human-centered conversation with your team when they need it. An AI assistant goes down a structured lane built for machines, where its request is fulfilled in a few turns without tying up a CSR. 

Detection is where Argus 1 starts. The rest of the suite governs what happens once you know an AI is on the line: how agents exchange information, across which channels and under what rules.

  • Structured data, not small talk. Two agents don't need to trade sentences the way people do. Argus 1 lets them exchange structured data across voice, digital and email.
  • Visibility. Carriers can see how much of their traffic actually comes from AI assistants and size their teams around the real number.
  • One rulebook. Everything runs as a single consolidated suite, governed by the same authentication, security, trust and authorization policies you already apply to human callers.

The hard parts

Two questions come up every time we show this to a carrier.

What if a person is treated like a machine? That's the mistake that matters most, and we designed AI Intercept around it. Sending an AI assistant down the human path costs a little efficiency. Sending a person down the machine path costs trust, and that customer may be calling during the worst week they'll ever have with you. So AI Intercept doesn't route on a coin flip. Only callers it is very confident are AI go to the machine lane. Callers it is very confident are human go to the human experience, and when the score falls in between, the call gets its own path instead of a guess. We benchmark these thresholds at scale against our own data before they reach a carrier's line.

Doesn't this open the door to bad bots? No. AI Intercept only decides which experience a caller gets; access is a separate question. An AI assistant asking for policy or claim data still has to clear the same security and authentication checks a person would. Every guardrail you run today stays in place.

Voice is the bridge

Today, when an AI assistant calls a carrier's AI, the exchange runs through a strange relay. One system produces language and turns it into speech. The speech crosses a phone line. The other side transcribes it back into text, works out what was meant and then acts on it. Two machines are pretending to be people so they can talk over a channel built for people.

Voice was built for humans, and it won't be how AI talks to AI for long. The future is the consumer's assistant connecting directly to the carrier's AI over a secure, governed platform and completing the task without the round trip through speech. The industry has already started forming standards for agent-to-agent interaction.

Until those standards are widespread, the phone line is where these conversations will happen. Detecting AI on voice is the bridge to that future, which is why AI Intercept is the first product in Argus 1.

What the industry has to get right

Getting there safely depends on three things.

Disclosure. AI agents should identify themselves. Liberate's agents already identify themselves as AI agents acting on behalf of the carrier, and the same should hold for consumer assistants. Regulators are moving in this direction: in February 2024, the FCC ruled that AI-generated voices count as "artificial" voices under the Telephone Consumer Protection Act, and states are following with their own disclosure rules.

Trust and authorization. An assistant calling for a policyholder has to prove it's authorized to act for them, with real consent, through a mechanism the carrier can verify.

Accountability. Both agents need complete records of the data they exchanged and the actions they took. When there's a conflict or an error, you should be able to see exactly how the interaction unfolded and where it went wrong.

Argus 1 is built around all three: governed exchanges, the same authorization policies as human interactions and a full audit trail on every transaction.

Why we're building it

At Liberate, we build AI agents that hold long, regulated conversations with policyholders and write the results back into the systems carriers run on, like Guidewire and Duck Creek. Building those agents taught us how differently AI behaves from people. Watching AI talk to AI, on both sides of the call, made the next step obvious.

The carriers that move first won't be guessing at their volume or staffing for callers who never needed a person. They'll fulfill every request, from every kind of caller, at any hour, and keep their people for the conversations that need people.

AI should handle AI, so people can handle people.

Want to see how much of your inbound volume is already coming from AI? Talk to our team.

‍

Key takeaways
  • AI agents are already calling insurers. They compare prices, request quotes and check on claims for their users, and most carriers can't tell them from people.
  • The AI agent is a new customer channel. 54% of policyholders would consider switching to an insurer that works with their AI. The carriers that answer these agents well will win that business.
  • Today's options don't work. Putting a licensed CSR on the line with a machine wastes people. Hanging up loses customers.
  • Argus 1 is a governed suite for AI-to-AI. AI Intercept, its first product, tells a person from an AI in the first seconds of a call and routes each to the right experience.
  • The rules don't change. AI callers clear the same authentication and security checks as people, and every exchange is auditable.
  • Voice is the bridge. Next, agents will connect to insurers directly, and Argus 1 is built for both.

‍

About Liberate

Liberate builds insurance-native Al agents that handle claims, servicing, and sales, trusted by carriers representing $100B+ in premiums on network.
By clicking “Accept”, you agree to the storing of cookies on your device to enhance site navigation, analyze site usage, and assist in our marketing efforts. View our Privacy Policy for more information.
Button Text