Which identity verification providers can detect deepfake faces generated by AI during a selfie or video check?

Nowadays, traditional selfie checks that match a photo to an ID are no longer enough to stop bad actors who can now use virtual cameras or face swap apps to inject fake imagery into a session. 

As such, using specialized identity verification providers is often the best way to confirm that a person is genuine and present in real time – especially for regulated businesses.

In this article, we’ll look at three examples of such providers, including GBG, iProov and AU10TIX.

Comparing top identity verification providers that can detect deepfake faces

GBG

We’ll introduce ourselves first before moving on to two other providers worth considering. 

Our end-to-end identity verification platform, GBG Go, combines document verification, biometrics, liveness detection and cross-industry fraud intelligence within a single solution. This layered approach helps verify genuine customers while detecting increasingly sophisticated fraud techniques. 

Here are three reasons businesses ranging from financial institutions to marketplaces to gaming operators choose to work with us:

Deepfake injection attack detection and biometric verification

As fraudsters use AI-generated imagery more and more often to bypass onboarding controls, biometric verification now has to go beyond just selfie matching. 

That’s why our biometric security capabilities are designed to identify physical and digital manipulation attempts in real time. 

Our platform monitors the relationship between hardware and software during the onboarding process to detect signs of emulators, virtual cameras and other non-standard system modifications commonly used in fraud attacks. It also provides protection against face swap applications and delivers 100% detection of deepfake injection attacks when using our unified capture tools.

To verify that an identity belongs to the person presenting it, our facial matching technology compares a customer's selfie against the portrait image on their identity document. Our system then analyzes 68 biometric facial landmarks to confirm identity ownership in seconds, which helps you recognize genuine users.

Certified passive liveness testing designed to stop presentation attacks

Traditional liveness checks often require users to blink, smile or perform other actions to prove they are physically present. These active challenges can introduce friction into the onboarding journey and create opportunities for customer error.

Our passive liveness testing delivers protection against presentation attacks while requiring only a single high-resolution selfie. Certified to the highest ISO/IEC standards for presentation attack detection, our technology analyzes subtle characteristics that are difficult for AI-generated content and spoofing tools to replicate.

Our system also evaluates indicators such as skin texture and elasticity, along with natural light reflections from the human eye, to determine whether a real person is present. 

And because customers aren't required to perform specific movements or actions, you can maintain a smooth onboarding experience while reducing abandonment rates.

Network intelligence that identifies fraud patterns across applications

A biometric check can confirm whether a face matches an identity document, but it can't always reveal the broader patterns associated with organized fraud. To uncover these threats, we combine biometric verification with intelligence from GBG Trust, our cross-industry network built from millions of identity records and fraud insights.

By analyzing activity across applications and industries, GBG Trust can identify signals that could otherwise remain hidden. This includes detecting reused documents and biometric identities, identifying high-velocity application activity within short time periods, and recognizing fraud patterns that span multiple organizations and sectors.

These additional layers of intelligence help identify coordinated attacks that may involve deepfakes, synthetic identities or account farming schemes. As David Thomas, Global Head of Product, Documents and Biometrics at GBG, explains:

"Velocity is much more than just, ‘have we seen this person before?’ It's about whether we've seen that face paired with a document carrying different information, or whether the same document is showing up repeatedly with different faces attached to it. That's a telltale sign of fraud. The combination of biometric matching and velocity is what makes the difference."

iProov

iProov specializes in biometric identity assurance and is known for its high-level security certifications. The company focuses on ensuring that an individual is a real person, authenticating in real time, rather than a recording or a deepfake.

Here is an overview of how iProov defends against deepfakes and AI-generated fraud:

• Express Liveness: This near-instant face capture analyzes imagery for 2D and 3D masks, photos, and videos presented to the camera. It also monitors for system-level spoofs like man-in-the-middle attacks.
• Dynamic Liveness: This patented technology uses a unique color code flash on the device screen. The reflection confirms the individual is present and authenticating in real time, providing advanced protection against AI-generated attacks.
• Active threat management: The iProov Security Operations Center monitors biometric transactions globally to detect emerging attack vectors and deploy algorithmic updates automatically.

AU10TIX

AU10TIX provides a multi-layered defense model that focuses on balancing speed with sophisticated fraud detection. It can be effective at identifying fakes that traditional verification systems often miss. It achieves this with: 

• Comprehensive biometric data: The platform supports ID-to-selfie, selfie-to-selfie and even selfie-to-token comparisons alongside liveness detection.
• Advanced neural networks: It uses algorithms that analyze network traffic and detect suspicious behavioral patterns to identify fraud.
• Consortium validation: Its deepfake detection prevents fraud through validation across a network of 60+ trusted users to identify large-scale forgery attempts.

What should a deepfake detection solution look for?

Fraudsters now use generative AI to manipulate biometric sessions. A strong solution should detect: 

Presentation attacks and spoofing attempts

This involves presenting a fake physical or digital image to the camera. A solution must be able to detect:

• High-resolution printed photos
• Screen replays of pre-recorded video
• 3D masks and lifelike dolls
• Natural human indicators via ISO-certified liveness testing

Deepfake injection attacks

Injection attacks present a higher level of risk because they bypass the physical camera sensor entirely to feed AI-generated faces into the verification stream.

Effective defense relies on identifying face swap applications in real time while monitoring for virtual cameras and emulators. Your solution must perform consistent device and environment integrity monitoring to ensure the software hasn’t been modified or tampered with.

Identity fraud patterns beyond the selfie

Sometimes a deepfake is so good it passes a single biometric check. This is why fraud signals outside the session are important. A solution should be able to:

• Track reused biometric identities across different accounts
• Spot the same ID document being used with different faces
• Use velocity and network-based detection to see large-scale attacks

Why we are a strong solution for deepfake detection

A reliable defense against deepfakes requires more than a camera check. We combine biometric verification with system-level integrity monitoring and our extensive GBG Trust network. 

This helps ensure that even if an AI-generated face looks perfect, our system will flag the injection attempt or the high-velocity data pattern. 

For companies that need application fraud protection without adding friction for genuine users, this combination of certified passive liveness and network intelligence provides the necessary security.

FAQs: Deepfake detection

What is the difference between a presentation attack and an injection attack?

A presentation attack occurs when a fraudster holds a physical object (like a photo or mask) up to a camera. An injection attack involves bypassing the camera and feeding a digital stream (like a deepfake video) directly into the verification system.

How does passive liveness detection help stop deepfakes?

Passive liveness detection analyzes a single image or video stream for micro-signals that AI cannot yet reproduce, such as skin elasticity and specific reflections in the eyes. It is more secure and has less friction than active checks that ask users to move or speak.

Are traditional face matching systems effective against AI deepfakes?

Traditional systems often focus on matching facial features between an ID and a selfie. While they can confirm two images look like the same person, they often cannot distinguish between a real face and a high-quality AI-generated deepfake without added liveness detection.

Sources: 

• https://www.iproov.com/liveness-detection
• https://www.au10tix.com/solutions/biometric-verification/
• https://www.au10tix.com/solutions/deepfake-detection/

*Disclaimer: Information relating to third-party products and companies referenced in this article is based on publicly available sources and official publications at the time of writing. While reasonable efforts have been made to ensure accuracy, product features, positioning and company information may change and GBG does not guarantee that all information remains current or complete.

Nothing in this article constitutes an endorsement, recommendation or ranking of any third-party provider. Readers should consult each provider’s official website and conduct their own assessment before making any purchasing decisions.

To request an update or correction, please contact communications@gbg.com