Sumsub has announced an upgraded deepfake detection solution that uses machine learning to more quickly identify emerging AI-generated fraud threats.
Adaptive Deepfake Detector is continuously updated, rather than relying on scheduled model upgrades that can take weeks or months to deploy.
The announcement comes as fraudsters use more sophisticated deepfake images, audio, video, and injection techniques to circumvent online verification checks.
Multi-stage attacks increased by 180% in 2025, reaching 28% of all fraud detected on Sumsub’s platform worldwide.
Sumsub’s upgraded deepfake detection model analyzes documents, geolocation, IP addresses, device signals, facial biometrics, liveness checks, and verification patterns across multiple users to detect potentially fraudulent networks.
Nikita Marshalkin, head of machine learning at Sumsub, said:
“Modern deepfakes can no longer be detected by the human eye, and decisions should be made based on real-time multi-signal analysis.
That’s why we released an upgraded deepfake detector to provide our clients with more than just a tool, but an online learning system that combines advanced document checking, device intelligence, and fraudulent network analysis to complement deepfake detection capabilities. ”
The solution also checks for presentation attacks, injection attempts, third-party involvement, and low-quality validation inputs such as motion blur, glare, and unusual facial expressions.
Sumsub added that the model can adapt to new fraud patterns without manual retraining as new threat signals enter the system.
Featured image: Compiled by Fintech News Singapore Based on image by warecreativestudio via Magnific
