Deepfake detection advancing with multi-signal approach

Fraud attempts carried out with deepfakes have become established as part of the online security landscape, with organizations turning to presentation attack detection and injection attack detection to fight spoofs against face or voice biometrics systems. They are also increasingly deploying multi-modal biometric and signals analysis tools.

Reality Defender held a webinar last week to present its vision for “Protecting Contact Centers from AI-Powered Voice Fraud.” Co-founder and CEO Ben Colman and Head of Product Diana Hsieh explained how the company applies real-time detection and multi-factor authentication.

Colman noted that voice is the modality most often spoofed with generative AI, and presented real-life examples of voice deepfakes. The company has found that out of 10,000 calls a day handled for a tier 1 bank customer, 0.17 percent are deepfakes.

Reality Defender VP of Human Engagement Gabe Regan writes in a company blog post that the move towards real-time, multi-modal deepfake detection represents a strategic shift in organizations’ synthetic media detection approaches.

Regan notes that Entrust’s recent Identity Fraud Report shows deepfake attacks on biometric systems are part of 40 percent of identity fraud attempts.

New multi-signal method for watch lists from Daon

Daon has been issued a patent for a way to build better deepfake voice watch lists.

The patent for “Methods and Systems for Enhancing Detection of Fraudulent Data” was issued by the U.S. Patent and Trademark Organization last week to Daon, listing CPO Ralph Rodriguez as its inventor.

Current voice watch lists are declining in value, according to the patent document, because they identify known threats to biometric systems, but are unable to adapt to evolving fraud methods. They also “are not typically known to adequately analyze behavioral patterns associated with voice interactions and thus do not capture clues that could indicate fraudulent activity.”

The system for enhancing the detection of fraudulent data described in the patent consists of a risk score mechanism that ingests behavioral, contextual and voice data. The data is analyzed for an overall assessment of risk level, with high-risk scores triggering an action like a review or transaction suspension. Like Reality Defender, the analysis is carried out in real-time to find anomalies associated with synthetic speech.

Daon plans to deploy the innovative technology within its xDeTECH solution for production deployments in 2025, Rodriguez tells Biometric Update in an email.

Deepfake detection among India AI projects

In India, MEITY (Ministry of Electronic & Information Technology) initiative IndiaAI has published an expression of interest for individuals and organizations to lead a set of development projects to increase the safety and fairness of AI.

IndiaAI seeks tools to watermark or label AI-generated content, perform AI risk assessments and management, stress testing and deepfake detection, along with frameworks for ethical AI.

The deepfake detection tools should perform real-time deepfake identification and mitigation.

Applications are open now through January 9, 2025, and submission guidelines are available here.

The rise in deepfake misinformation and fraud is recognized in a November 27 advisory from the Indian Computer Emergency Response Team (CERT-In). The Hindustan Times reports the advisory followed the release of several fake audio clips by President Modi’s Bharatiya Janata Party.

BioID deepfake detection awarded

BioID’s real-time deepfake detection technology for photos and videos picked up the 9th Munich Digital Innovation Award in November.

The award winner was determined by live audience voting during the event, which was held under the theme “Reality Check AI.”

“With deepfake and impersonation attacks on the rise, safeguarding digital identities is more critical than ever,” says BioID CEO Ho Chang.

The company launched a new version of its deepfake detection software, which analyzes facial features for signs of manipulation, earlier this year.

Article Topics

BioID  |  Daon  |  deepfake detection  |  deepfakes  |  digital identity  |  fraud prevention  |  India  |  Reality Defender  |  tender