From Voiceprint to Iris: How AI Glasses Enable Silent 'Glance-to-Pay' Authentication

Why Iris Recognition Replaces Voiceprint on AI Glasses

Current AI glasses rely on voiceprint recognition for payment verification, requiring users to speak commands aloud. This creates social discomfort in public settings like offices, subways, or meetings, causing privacy-conscious users to abandon glasses-based payments.

Iris vs. Voiceprint: Beyond Silent Verification

Iris biometrics offer uniqueness, stability, and non-replicability with a false recognition rate below one in a million. Unlike voice, iris features inherent anti-spoofing properties: 3D eyeball structure, dynamic iris texture changes, and pupillary light reflex make forgery far more difficult. However, deploying iris recognition on glasses poses distinct challenges: narrower interpupillary distance, tighter module volume constraints, variable wearing angles, and a complex end-to-end chain from glasses sensors to phone and cloud, each vulnerable to attack. Traditional phone-based iris solutions cannot be directly transferred.

Ant Group collaborated with multiple vendors to introduce an industry-first full-chain trusted security solution, achieving sub-million false acceptance rates for iris verification.

Technical Architecture: Three-Layer Security Stack

1. Hardware Layer: Miniaturized NIR Iris Module

A custom near-infrared (NIR) camera module fits inside the temple, capturing high-resolution iris images in a tiny form factor. IR LEDs provide precise illumination for clear texture even in low light. The module is imperceptible during wear.

2. Security Layer: TEE-Based Trusted Execution

Payment-grade identity verification demands full-lifecycle biometric security. Through hardware and chip vendor partnerships, a dedicated secure image capture chain isolates from the normal OS path. Leveraging a dual-security system, iris feature extraction, template matching, and policy execution all occur inside a Trusted Execution Environment (TEE).

3. Risk Control Layer: Continuous Multi-Factor Defense

A comprehensive risk control system provides continuous detection and adversarial resilience. It combines multi-modal verification (voiceprint/iris), implicit authentication, and device pairing relationships. Built on encrypted computing and a bidirectional key infrastructure, real-time risk assessment during payment prevents single-factor or single-chain breaches, forming a self-reinforcing, multi-source defense.

Algorithm Principles: From Capture to Recognition

Iris Recognition: Handling Off-Axis, Deformation, Occlusion, and Glare

For AI glasses scenarios, the solution introduces an innovative framework based on iris reconstruction and cross-scale semantic collaboration. It fuses real-time streaming multi-frame features and applies binocular pose compensation algorithms, enabling stable, unobtrusive iris verification in open environments and complex lighting.

Liveness Detection: Defending Against Presentation Attacks

The system confronts print, screen, and mask attacks. Under cross-device and cross-illumination test protocols, both Attack Presentation Classification Error Rate (APCER) and Bona Fide Presentation Classification Error Rate (BPCER) are below 1%. A defense-in-depth system covers 35 categories of 2D and 3D attack materials. Colored contact lens detection spans mainstream brands with a 99% interception rate, balancing high security with usability. The model fuses spatial texture, frequency-domain artifacts, multi-scale features, and local evidence consistency supervision for collaborative discrimination, deployed in a lightweight, production-ready form.

Future: Standardization and Cross-Platform Expansion

Starting with the Qianwen N1Pro, iris verification enters daily consumer use. Ant Group is partnering with multiple smart terminal brands and ODMs to deploy the solution, and jointly building an AI terminal identity authentication evaluation lab with the China Academy of Information and Communications Technology (CAICT) Tel Terminal Laboratory to standardize iris, voiceprint, and other verification technologies.

The ultimate vision for AI glasses is technology disappearing into natural interaction: users simply live, and devices understand intent and fulfill services without explicit operation. Iris payment exemplifies this—no deliberate verification action, just a natural glance, and the system instantly recognizes the user and completes payment silently. The shift from "speak to pay" to "glance to pay" changes not just a command word but the human-computer interaction paradigm.

Meanwhile, the Lingying AOS second-generation prototype has validated the RTOS iris solution on the Juxin ATW6095 and Xingchen SSC309QL chips, proving iris verification can adapt to lightweight platforms. Ecosystem partners are invited to collaborate on broader terminal deployment.