Meridian detects social engineering on live calls and meetings — and protects AI voice agents from injection, hijacking, and extraction attacks — in real time, in plain English, on the device where the conversation happens.
"We need this wired in the next ten minutes — your boss is in a meeting."
"This is Mark from Microsoft — your CFO is on the other line right now."
"You don't need to verify me — I called you, remember?"
* Not yet available — on active roadmap (SOC 2 path, SSO, SIEM).
What we detect
Stable keys your security team can route, audit, and automate
Plain English labels while the call or agent session is still live
Human and AI voice-agent coverage on one signal layer
These examples show how Meridian labels social engineering on live calls, video meetings, and AI voice agents — each with a stable key and a plain-English meaning. We detect more in production than we list here. Swipe to browse, or open the full example catalog below.
Caller impersonates Google support, fabricates an account alert, and reverses verification flow.
Updating...Keep the verification flow one-way: the caller provides required details first.
Find the official number yourself. Do not use any number they provide.
Hey. This is Cedric calling with Google support.
How are you doing today?
Basically, I got an alert on my desk saying that your account is flagged.
Social engineering works by moving a reasonable person through a bad decision before normal controls can help.
Meridian supports the roles most likely to meet a suspicious request before the security team ever sees it.
01Give front-desk teams a clear pause point before sensitive details are shared.
02Support clinics, hotels, agencies, and offices while everyday calls are still live.
03Flag risky caller behavior before routing, escalation, or information collection.
Live protection
One tap activates Meridian. No security jargon. No interpretation required.
Urgency pressure, authority impersonation, and identity avoidance — each signal name and excerpt updates while the call is still happening.
The problem
Attackers stopped writing exploit code and started making phone calls. Meridian protects the moment your employee is under pressure.
Most teams still discover the attack only after someone already shared too much.
How it works
Three steps from suspicion to clarity. All while the call is still happening, not after it.
001Your employee is on a call. The voice is calm. The request sounds reasonable. The pressure is slightly off. They open Meridian. One tap.
002Live audio is scanned for social engineering signals every few seconds. Each tactic is classified and scored as the conversation unfolds.
003A live risk score, 0 to 100, updates in real time. One plain English line explains what the caller is doing.
Signal reference
Lookup for security and engineering — representative examples from production, not every signal Meridian can surface. Need AI agent context? See Protecting AI agents. For dashboards and rollout, see For business.
Sample patterns Meridian flags when a caller pressures, impersonates, or manipulates someone on the line.
| Key | Meaning |
|---|
Sample patterns when someone tries to hijack, probe, or extract from your AI voice agent.
| Key | Meaning |
|---|
Get started
Book a demo with the Lurion team and watch detection run while a conversation is still happening.