Skip to main content
2025

Computer Vision Deepfake ID Detection

Computer Vision Deepfake ID Detection

Instinctools helped a Netherlands-based digital bank strengthen its KYC onboarding process by developing a custom computer vision layer for deepfake ID detection. Integrated alongside the bank’s existing SaaS identity verification platform, the solution improved fraud detection accuracy while maintaining a seamless onboarding experience for legitimate users.

Challenge

The bank’s third-party identity verification platform struggled to detect AI-generated and manipulated identity documents, while also producing high false-positive rates for real users. Vendor-controlled update cycles limited the bank’s ability to quickly adapt to emerging fraud techniques, creating operational, financial, and reputational risks at a scale of millions of annual onboarding checks.

Solution

Instinctools designed and deployed a YOLO-based computer vision pipeline trained on a fraud-realistic dataset containing synthetic IDs, manipulated documents, and hard-negative samples. The team integrated the CV system into the bank’s existing KYC workflow with tiered decision logic, real-time risk scoring, and rapid retraining capabilities to adapt quickly to new deepfake attack patterns.

Results

The custom computer vision layer increased synthetic document fraud detection by 30%, reduced false positives for legitimate users by 60%, and lowered manual review workload by 40%. The bank gained a scalable, bank-owned fraud detection capability capable of evolving alongside emerging AI-driven identity threats.

Project Details

Year

2025

Live ProjectView →