Video meeting, treasury approvalParticipant 2 of 3
A video call with a finance executive
Live stream, 14/05/2026, 14:02Source: virtual camera
AI tracesManipulation traces
Video meeting, treasury approvalParticipant 2 of 3
Live stream, 14/05/2026, 14:02Source: virtual camera
What we readWhat the eye sees
Logged to the meetingAnalysis kept, reason written
ConclusionSuspect
Justifications
AI tracesHigh
Manipulation traces, face area
Stream from a virtual camera
No generator signature found

A familiar voice on the line,
and a transfer to approve.

A call from your CFO, a video meeting with a long-standing supplier: the voice and the face are familiar, and you are asked to act quickly. Yet both can now be cloned from a few public recordings.

What it does

Checks the voices in your calls and voice messages and the faces in your video meetings, and tells your team whether AI generated or edited them, while there is still time to act.

Why it matters

A cloned voice or a generated face now asks for the transfer, the password reset or the change of bank details, and the people it imitates are exactly those your team is used to trusting.

Who it’s for

Finance and treasury teams, executive assistants, contact centres, and the security teams who protect them.

The problem

A known voice used to be proof enough

Payment procedures, call-backs and dual approvals all end with someone who recognises a voice or a face and acts on it, and that recognition is exactly what a generated voice or face is built to pass.

The CEO’s voice

A cloned voice calls the finance team with an urgent, confidential transfer, with the tone, the phrases and the pressure of the real executive.

Waveform of a voice message with two speakersWhat you hear
What we detect

One voice message, a genuine speaker and a cloned one.

The meeting where every other face was fake

An employee joins a video call with the CFO and several colleagues, all of them generated, and is asked to make a series of transfers. Every face on the screen is familiar, and none of them is real.

The call to the contact centre

A cloned customer voice gets past voice authentication or convinces an agent to reset a password, change a phone number or move money, one call at a time.

In figures

Impersonation by deepfake is already counted in millions

$25m

wired by Arup after a video call where every other face was fake.

Source: Hong Kong police via Business Insider, May 2024
€95m

wired by a private bank’s chairman after a call with a cloned voice.

Source: Forbes Italia, Sept 2026
+1,300%

deepfake attempts against contact centres in 2024, as cloned voices reached the phone line.

Source: Pindrop, 2025

Our answer

Where Label4.ai fits in your process

Label4.ai adds one question to the checks you already run, and leaves the rest of your process as it is.

Call or meeting starts

Phone line, voice message, video meeting or contact centre, whatever your channels already are.

Your existing checks

Call-back on a known number, dual approval, caller ID. Unchanged.

Label4.ai analysis

Is this voice what it claims to be? Checked automatically, in seconds, with its reasons.

Decision

Go ahead, call back on a known number, or escalate to security, with the reason written down.

A result anyone on the team can read

The person on the call sees the conclusion first, then three sentences written in plain words, like these.

  • The voice on this call carries the signature of a speech synthesis model from its first sentence.
  • The face in the video was generated and laid over another person, frame by frame.
  • No generator signature was found, so the conclusion rests on the traces in the signal alone.

Next to your call-back procedure, never instead of it

Impersonation is usually stopped by people and procedures, and Label4.ai adds a third line that looks at the voice and the face themselves, so a cloned executive is caught even when the procedure is skipped under pressure.

Social engineering

An urgent request, a confidential deal, a supplier who has changed bank.

ProceduresCall-back on a known number, dual approval and payment limits, which work when someone has the time to apply them.Already in your finance process

Spoofed channels

A caller ID that matches, an invitation from a familiar account, a message from a known number.

Channel checksCaller ID, meeting platform accounts and contact centre authentication check where the call comes from, not who is speaking.Already in your systems

Generated or edited content

A cloned voice, a swapped face, a generated participant in a meeting.

Label4.aiReads the traces the generation process leaves in the voice or the video, whoever the caller claims to be.

We do not rely on accents, hesitations or lip sync, because each new generation of models corrects them. What we read is left by the way the voice or the video was produced, and it is still there when the imitation sounds perfect.

Account opening is covered by Identity Fraud Detection, with the same analysis applied to the selfies and ID images that open an account.

How we do it

What sits behind each conclusion

  • An analysis shaped to your process Each project is set up for the way you work, with the checks that matter for your content, your threshold and the action we recommend. Every analysis looks first for a signature left by the generator, then for the traces a model leaves and for signs of editing.
  • Detection that keeps the process running Most calls are exactly what they seem, so before the contract we run your own recordings and agree with you on the false alarm rate that suits your process. That is why we quote no generic detection rate: we set up a project and tune it to give you the best detection rate on your own use case.
  • A conclusion that holds up The same content, analysed twice, gives the same report twice, which turns a series of calls from suspicious into provable.

It runs in our cloud or on your own servers, through the Anomaly Engine console or the API.

As an option, with our watermarking expertise, your executives’ voices and videos carry an invisible mark, so anyone who receives them can check that they really come from you. Anything without that proof is analysed as usual. About our watermarking

See the technology in detail

Your content

See what your own calls reveal

You send a batch of recordings whose outcome you already know, including the attacks you caught and the ones you found out about too late. We come back with a result for each recording, your false alarm and missed fraud rates, and a plain statement of what we did not see.

We reply to every request within two working days, under NDA if you prefer.