Threshold Fully Homomorphic Encryption Enables Faster Iris Biometric Matching in Research Prototype

Global: Threshold Fully Homomorphic Encryption Enables Faster Iris Biometric Matching in Research Prototype

In January 2026, researchers released a preprint on arXiv describing a new cryptographic approach for large‑scale iris verification that aims to preserve user privacy while maintaining rapid matching times. The work evaluates a Threshold Fully Homomorphic Encryption (ThFHE) system designed for the World ID project, which seeks to authenticate individuals worldwide using iris scans.

Background and Motivation

Iris recognition is prized for its high accuracy in biometric databases that can contain millions of entries. However, storing raw iris codes raises significant privacy concerns, prompting the exploration of privacy‑enhancing computation methods that keep biometric data confidential throughout the matching process.

Prior Multiparty Computation Approach

Earlier research by Bloemen et al. employed a 2‑out‑of‑3 secret‑sharing multiparty computation (SS‑MPC) protocol. Their implementation could compare 32 users against a database of 2^22 iris codes in roughly 2 seconds using 24 H100 GPUs, more than 40 communication rounds, and 81 GB of data transferred per party, assuming a network speed exceeding 3 Tb/s. Security relied on the assumption that no more than one server would be compromised.

Introducing Threshold FHE

The new study investigates the use of Threshold Fully Homomorphic Encryption as an alternative. According to the authors, ThFHE offers several advantages: it eliminates the need for a trusted setup, allows the encrypted database and queries to be publicly visible, distributes the secret among many parties, and can incorporate active security without markedly affecting performance.

Performance Benchmarks

In the proof‑of‑concept implementation, the computation phase matched 32 eyes against a database of 7·2^{14} iris codes in approximately 1.8 seconds on eight RTX‑5090 GPUs. A smaller test involving four eyes completed in about 0.33 seconds. The authors note that the protocol adds only two to three communication rounds, depending on deployment choices.

Technical Innovations

The researchers leveraged recent advances in FHE‑based linear algebra, optimizing operations with int8 precision on GPUs. They also introduced a technique that reduces the number of ciphertexts processed early in the pipeline, thereby decreasing computational overhead.

Implications for Privacy and Deployment

By enabling fast, privacy‑preserving iris matching without a trusted third party, the ThFHE solution could broaden the feasibility of large‑scale biometric systems such as World ID. Nevertheless, the authors caution that real‑world deployment would still need to address network latency, hardware costs, and regulatory considerations.

This report is based on information from arXiv, licensed under Academic Preprint / Open Access. Based on the abstract of the research paper. Full text available via ArXiv.