Türkçe
English
РусскийDeepfakes and face swaps
The attacker tries to pass verification with AI-generated imagery or footage morphed into someone else’s face.
An AI model performs passive liveness analysis and detects deepfake screen attacks.
Attack Types
What Fraudsters Try, and How Prove ID Stops Them
Explore the most common attacks on remote identity verification and the Prove ID controls that counter each one.
Replayed screen recording
A video or photo played on another screen is shown to the camera.
Screen attacks are detected in both the face and ID steps, and the frame is checked for multiple faces.
Pre-recorded video
A recorded or generated video is fed into the verification flow.
Facial-gesture, finger and ID-tilt challenges change randomly every session; a prepared video cannot know the right move in advance.
Fake or altered ID
An ID with a replaced photo, a digitally edited ID or a photocopy is used.
Holograms, guilloche, rainbow print and hidden images are checked; digital manipulation, pasted photos and photocopies are detected.
Cloned ID data
Documents that look flawless but whose chip data does not match are used.
Chip data is read via NFC and goes through active and passive authentication and a certificate validity check; the face is matched against the chip photo.
Repeat fraudsters
Previously caught individuals try again with new applications.
Face, national ID number and device blacklists are maintained; suspicious transactions are flagged in the dashboard.
Layered Defence
One Check Is Not Enough, Five Layers Work Together
Bypassing a single check is not enough for an attacker; Prove ID collects independent signals at every layer and combines the decision on the server side.
Resistance to Injection
Every Session Is Unique, a Prepared Video Won’t Work
Verification challenges are chosen at random when the session starts. A pre-recorded, generated or injected video cannot know the move being requested at that moment.
- A pool of eye, mouth and head movements
- Placing a finger on randomly chosen parts of the ID
- Hologram checks by tilting the ID horizontally and vertically
- Variable security questions and the ID serial number
- 1, 3 and 5 second verification options
This session’s challenges
#S-4821
- Blink your eyes
- Smile
- Turn your head left
- Turn your head right
- Raise your head
- Place your finger on the top-right corner of the ID
- Place your finger on the bottom-left corner of the ID
- Tilt the ID horizontally
- Tilt the ID vertically
- Say your ID serial number
Video Call & AI Agent
Protection Continues Throughout the Call
In video calls guided by the digital assistant, voice, behaviour and process signals are monitored end to end.
Voice consistency
The system checks that the same voice continues throughout; a change of person mid-call is noticed.
Emotion analysis
Signs of threat or coercion are detected from facial expressions.
Verbal consent
Agreement and consent texts are confirmed verbally by the user and recorded.
End-to-end recording
A screen recording of the whole process and the verification media are stored in the dashboard for audit.
Independent Proof
Measured Performance, Not Claims
Our face verification model has been evaluated in tests by the US National Institute of Standards and Technology (NIST).
Face Verification
%99,13success rate in banking
- FAR
- %0,0001
- FRR
- %0,87
- F1
- 0,9956
Liveness Check
%95,9success rate in banking
- FAR
- %0,72
- FRR
- %4,1
- F1
- 0,9755
NIST%90+
The only Turkish company with a 90%+ score at NIST
Techsign is listed with a published report card in the NIST FRTE face recognition evaluation.
View the NIST report card0
Distinct identity checks
0
Completed verifications
0
Enterprise customers
0
Independent defence layers
Frequently Asked Questions
Your Questions, Answered
How does Prove ID detect deepfake attacks?
Prove ID analyses the image with an AI-based passive liveness model without requiring any extra movement, and detects deepfake and screen attacks in both the face and ID steps. In addition, active challenges chosen at random for each session are requested, and the result is re-verified on the server side.
Can verification be passed with a pre-recorded or AI-generated video?
Verification challenges are chosen at random when the session starts: eye, mouth and head movements, placing a finger on a random corner of the ID, tilting the ID horizontally or vertically, and reading out the ID serial number. Because a prepared video cannot know the move requested at that moment, the flow is resistant to such attacks.
How are fake or tampered ID cards caught?
Security features such as holograms, guilloche, rainbow print and hidden images are checked; digital manipulation, pasted photos and photocopies are detected. For NFC-enabled IDs, chip data goes through active and passive authentication and certificate validity checks, and the face is compared with the chip photo.
What happens if previously caught fraudsters apply again?
Prove ID maintains face, national ID number and device blacklists. When a blacklisted person or device starts a new verification, the transaction is blocked or flagged for review in the admin dashboard.
Has Prove ID’s face verification been independently tested?
Yes. Techsign’s face recognition model is listed in the US National Institute of Standards and Technology (NIST) FRTE evaluation, and Techsign is the only Turkish company with a score above 90% at NIST. The false acceptance rate (FAR) in face verification is 0.0001%.
Can Prove ID be used for KYC and KYB processes?
Yes. Prove ID provides remote KYC (eKYC) for individual customers: ID reading, NFC chip verification, face matching, liveness detection and video call verification. For corporate customers it supports a KYB business verification flow and verification of documents such as signature circulars and the Turkish Trade Registry Gazette. More than 60 institutions, led by banks, fintechs, payment, investment and crypto-asset companies, use Prove ID.
Stop fraud at the source
See with our team how Prove ID works against deepfake and fake ID attacks in your own flow.