Full Report
ESET’s Jake Moore used smart glasses, deepfakes and face swaps to ‘hack’ widely-used facial recognition systems – and he'll demo it all at RSAC 2026
Analysis Summary
# Tool/Technique: Biometric Identity Subversion (AI-Enhanced Facial Recognition Bypass)
## Overview
This research demonstrates the vulnerability of facial recognition systems and electronic Know Your Customer (eKYC) platforms. It involves using consumer-grade hardware and AI-based software to deanonymize individuals in public, bypass financial identity verification, and evade law enforcement watchlists through real-time digital overlays.
## Technical Details
- **Type**: Tool / Technique (Adversarial AI)
- **Platform**: Facial Recognition Systems, eKYC Platforms, CCTV Infrastructure, Smart Glasses (Hardware)
- **Capabilities**: Real-time deanonymization, synthetic identity creation, and real-time face-swapping.
- **First Seen**: March 2026 (Demonstrated at RSA Conference)
## MITRE ATT&CK Mapping
- **[TA0001 - Initial Access]**
- **[T1566 - Phishing]**: Using deanonymized data gathered via smart glasses to craft highly targeted social engineering attacks.
- **[TA0006 - Credential Access]**
- **[T1556.006 - Modify Authentication Process: Multi-Factor Authentication]**: Bypassing biometric MFA/KYC using synthetic imagery.
- **[TA0005 - Defense Evasion]**
- **[T1036 - Masquerading]**: Using face-swap technology to appear as a different individual to surveillance systems.
## Functionality
### Core Capabilities
- **Real-time Identity Scraping**: Utilizing modified off-the-shelf smart glasses to capture facial data and cross-reference it with publicly available online sources and social media profiles instantly.
- **Synthetic eKYC Bypass**: Using AI-generated imagery and freely available software to create fictitious personas that are accepted by financial institutions as genuine "live" applicants.
- **Real-time Face Swapping**: Overlaying a digital likeness (e.g., a celebrity/different person) onto a live camera feed to prevent matches against watchlists (e.g., Police/CCTV systems).
### Advanced Features
- **Public Data Correlation**: The ability to link a physical face in a crowd to a specific digital identity and social media footprint within seconds.
- **Liveness Detection Evasion**: Techniques used to fool the "liveness" tests often employed by banking apps to ensure a user is physically present.
## Indicators of Compromise
*Note: As this refers to a physical/adversarial AI research demonstration, traditional file-based IOCs are not applicable. Behavioral indicators are listed below:*
- **Behavioral Indicators**:
- Rapid, automated queries to public social media profiles originating from hardware with camera access.
- Inconsistencies in biometric depth perception or "digital artifacts" around the edges of a face in high-definition video feeds.
- Unexpected account creation patterns utilizing high-quality synthetic images that lack historical digital footprints.
## Associated Threat Actors
- **Identity Thieves/Fraudsters**: Exploiting eKYC for fraudulent bank account creation.
- **State-Sponsored Actors**: For deanonymizing targets in public spaces or evading international border/surveillance watchlists.
- **Private Investigators/Stalkers**: Using smart glasses for unauthorized surveillance.
## Detection Methods
- **Behavioral Detection**: Monitoring for non-human patterns in identity verification (e.g., perfect pixel consistency in synthetic images).
- **Deepfake Detection Algorithms**: Implementing specialized models designed to detect temporal inconsistencies in video streams (glitches during face-swapping).
- **Physical Hardware Audits**: Detection of unauthorized modified wearable devices in sensitive environments.
## Mitigation Strategies
- **Multi-Modal Biometrics**: Do not rely solely on facial recognition; integrate voice, gait, or behavioral biometrics.
- **Hardware-level Liveness Checks**: Utilizing infrared and depth-sensing cameras (LiDAR) that are harder to fool with 2D screens or digital overlays.
- **Adversarial Testing**: Regularly subjecting facial recognition systems to "red team" attacks using deepfake and face-swap tools.
- **Privacy Hardening**: Users reducing the amount of publicly searchable high-resolution facial imagery on social media.
## Related Tools/Techniques
- **Deepfake Software**: Tools like DeepFaceLive or SwapFace.
- **OSINT Tools**: Facial recognition search engines (e.g., PimEyes).
- **Generative AI**: StyleGAN or similar models used for creating synthetic personas.