SolutionsContact centres
AI agents call until someone says yes.
Voice agents work a queue the way a script works a login form: short calls, small variations, looking for the agent or the IVR branch that gives something away. Each call gets its own verdict, so the pattern shows in your queue within minutes.
- At risk
- Account probing, IVR enumeration, agents talked into an exception
- Arrives on
- In-house and outsourced queues, and the IVR in front of them
- On a flag
- The call is tagged and routed to a specialist team
- First verdict
- About twelve seconds of the caller's speech
In your queue
The pattern shows in the queue, not in one call.
Every call gets its own verdict, so a burst of short synthetic calls stands out against normal traffic within minutes. A call too short to score is marked as that, not as clear.
twelve minutes on one queue3 of 8 flagged
| Answered | Line | Length | Verdict | Flag |
|---|---|---|---|---|
| 09:04 | Care 2 | 3:12 | Clear | |
| 09:13 | Care 2 | 0:41 | Flagged | at 12 s |
| 09:13 | Care 5 | 0:38 | Flagged | at 12 s |
| 09:14 | Care 1 | 4:02 | Clear | |
| 09:14 | Care 3 | 0:44 | Flagged | at 14 s |
| 09:15 | Care 2 | 0:09 | Inconclusive | too short |
| 09:16 | Care 4 | 2:27 | Clear | |
| 09:17 | Care 5 | 5:48 | Clear |
- 09:04Care 23:12Clear
- 09:13Care 20:41at 12 sFlagged
- 09:13Care 50:38at 12 sFlagged
- 09:14Care 14:02Clear
- 09:14Care 30:44at 14 sFlagged
- 09:15Care 20:09Inconclusive
- 09:16Care 42:27Clear
- 09:17Care 55:48Clear
an illustrative queue: short calls a minute apart, each flagged on its own, and one too short to score
Why now
62%
of organisations surveyed had a deepfake attack using social engineering or automated processes.
Source: Gartner survey of 302 security leaders · Sept 2025 ↗The call your agent hears
“Hi, I'd like to check the balance on the account ending 4471.”
AI voice agents call in at scale, looking for the agent who will say yes.
On a flag
A flag is advice. Your call flow decides.
Trueline never ends or blocks a call. It tells the systems your team already works in, with the reason, and the response stays yours.
01
Agent screen
The agent sees the flag while the call is live and follows the script you set for automated callers.
02
Fraud team
The morning's flags arrive as records that line up: the same lines, the same minutes, the same short calls.
03
Call flow
Your routing tags the call and sends it to a specialist team, or to a queue built for it.
What contact centres ask first.
Every answer here is what the system does today. Where the honest answer is a weakness, it is written as one.
Detection performance →01We run on someone else's platform. Can it still work?
What Trueline needs is a copy of the caller's audio. Twilio and Telnyx send it with one URL. Contact-centre platforms open a session and stream it, and the record that comes back is the same either way.
Nothing is installed on agent machines and your routing does not change.
How the audio reaches us →02What about very short calls?
A first verdict needs about twelve seconds of the caller's speech. A call that ends before that closes as inconclusive, with how much was heard recorded on it.
It is never closed as clear, so a run of nine-second calls stays visible instead of looking like a quiet morning.
03Most of our callers are on mobile phones. Does that matter?
Our numbers come from internet and carrier routes, which is where voice agent fraud arrives first and where Trueline is strongest.
A mobile leg is a harder path for any detector. The call is compressed by the handset and the mobile carrier before anyone else hears it, and that first hop takes away part of what makes a synthetic voice detectable.
It is where our measurement work is now, and we will not tell you it is solved. If your callers are mostly on mobiles, say so before a pilot rather than after, and the report will tell you where your own traffic stands.
How we measure it →Bring one queue. We will show you its calls.
A pilot runs on one number or one queue for a week, beside everything you already have, and ends in a written report of what was flagged and why.