Deep Fake Detection

Baaj Group LLC

The Problem

Fraudsters are increasingly using deepfakes to compromise customer onboarding and eKYC systems. Techniques such as face morphing, video/image injection, image swapping, and synthetic images using AI are undetectable to the human eye. These techniques pose a threat to traditional eKYC onboarding technologies such as automated document authentication and traditional biometric matching systems.

AI as a Solution

AI can be used to fight modern deepfake attacks. A robust AI solution incorporates:
  • Custom AI models to detect anomalies in facial photographs.
  • Computer vision used to train thousands of fraud samples to detect anomalies and inconsistencies.
  • LLMs for behavioral validation using dynamic questioning and behavioral analysis in a manual review process and/or real-time communications.

Real World Performance

Our deepfake detection has been deployed commercially in a national deployment of over 3 million users. Key metrics from this deployment show:

  • Over 8000 fraud attempts detected and neutralized with ~$10Million of losses prevented
  • Response time: < 2 seconds (includes image transmission time)
  • FAR: ~0.03%*
  • FRR: ~0.15%*
  • 3-Try FRR: ~0.015%

*Note:  FAR and FRR may vary due to differences in user population, geographic location, application, and other factors.

Please contact us to see how we can use our Deepfake detection models and our review processes to enhance your fraud detection