Onboarding session KYC-2026-55120Selfie check
A selfie of a woman taken for an identity check
selfie_capture.jpg, 02/04/2026, 14:37Capture data: none
AI tracesManipulation traces
Onboarding session KYC-2026-55120Selfie check
selfie_capture.jpg, 02/04/2026, 14:37Capture data: none
What we readWhat the eye sees
Logged to the sessionAnalysis kept, reason written
ConclusionSuspect
Justifications
AI tracesHigh
Manipulation traces, face area
No capture data
No generator signature found

A face on screen,
and a few seconds to trust it.

An identity check runs a selfie and a document match in the time it takes a customer to hold up a phone, and the face being matched can now be produced by a model even faster than that.

What it does

Checks the selfies, liveness videos and ID images that pass through your verification chain, and tells your decision engine whether AI generated or edited them.

Why it matters

A face produced by a model now looks like any other selfie, and once it gets through onboarding it opens an account that every fraud that follows can use.

Who it’s for

KYC and identity verification vendors, who can run it under their own brand, and the fraud and security teams at the banks, fintechs and platforms they serve.

The problem

Liveness checks were designed for masks, prints and screens

Face match and liveness were built to catch someone holding a photo, a mask or a phone up to the camera, and the generated face of today often never passes in front of a camera at all: it is injected straight into the stream through a virtual camera or an emulator, and it moves, blinks and matches the document it was made with.

The face swap

A real person in front of the camera with another face laid over theirs, frame by frame. The liveness gestures are genuine, which is why the session looks so natural.

The face that belongs to nobody

A selfie and an ID photo generated together so that they match each other perfectly, for an identity that has never existed anywhere else.

Selfie of a man holding an identity card marked specimenWhat your eyes see
What we see

The face and the card photo carry the traces of the same model.

The series

Hundreds of onboardings spread over weeks and countries, with the same model producing every face. Each session passes on its own, and the link appears once you can read how the images were made.

In figures

Deepfakes have become an everyday attack on identity checks

1 in 5

biometric fraud attempts is now a deepfake, and deepfaked selfies rose by 58% in 2025.

Source: Entrust, 2026 Identity Fraud Report
+96%

deepfake fraud attempts in France in 2025, and 94% more in the United Kingdom.

Source: Sumsub, Identity Fraud Report 2025-2026
+2,665%

virtual camera attacks on remote identity checks, which feed a generated face straight into the session.

Source: iProov, Threat Intelligence Report 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.

Session started

App, web flow or branch tablet, whatever your capture already is.

Your existing checks

Document authenticity, face match and liveness. Unchanged.

Label4.ai analysis

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

Decision

Approve, ask for a new capture, or send to manual review, with the reason written down.

A result anyone on the team can read

The reviewer sees the conclusion first, then three sentences written in the words of the job, like these.

  • This selfie carries traces of a generative model, and they are strongest on the face.
  • The face area was edited separately from the rest of the picture, which is typical of a face swap.
  • No generator signature was found, so the conclusion rests on forensic analysis alone.

Next to your liveness and injection checks, never instead of them

An identity check already has two lines of defence against a fake face, and Label4.ai adds a third one that looks at the content itself, so a generated face is caught whichever way it reached your flow.

Presentation attacks

A printed photo, a mask or a second screen held up to the camera.

Liveness detectionAlready in your chain, it checks that a real person is in front of the lens.Evaluated under ISO/IEC 30107-3

Injection attacks

The camera is bypassed and a stream is fed in through a virtual camera, an emulator or a modified app.

Device and channel checksAlready in your chain or your vendor's, they check that the stream really comes from the device's camera.Evaluated under CEN/TS 18099

Generated or edited content

A face swap, a face that belongs to nobody, an ID photo produced to match it.

Label4.aiReads the traces the generation process leaves in the image, the video or the voice, whether the content came through the camera or around it.

We do not rely on blinking, lip sync or skin texture, because each new generation of models corrects them. What we read is left by the way the content was produced, and it is still there when the face looks perfect.

Calls, voice messages and video meetings are covered by Deepfake & Impersonation, with the same analysis applied to the voices and faces that reach your teams.

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 people in front of the camera are who they say they are, so before the contract we run your own content 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 onboardings from suspicious into provable.

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

See the technology in detail

Your content

See what your own sessions reveal

You send a batch of captures 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 capture, 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.