Papers

Research, technical reports, and whitepapers from the Nillion team.

Fission: Distributed Privacy-Preserving Large Language Model Inference

Mehmet Ugurbil, Dimitris Mouris, Manuel B. Santos, José Cabrero-Holgueras, Miguel de Vega, Shubho Sengupta

IEEE CNS 2025

A privacy-preserving framework that combines MPC with shuffled evaluator networks for LLM inference, achieving up to 8× faster inference and 8× reduced bandwidth compared to prior works while retaining high accuracy.

Conference2025

Wave Hello to Privacy: Efficient Mixed-Mode MPC using Wavelet Transforms

José Reis, Mehmet Ugurbil, Sameer Wagh, Ryan Henry, Miguel de Vega

PoPETS 2025

New protocols for secure MPC leveraging Discrete Wavelet Transforms for computing nonlinear functions over large domains, achieving extremely fast evaluation of activation functions with minimal communication.

Journal2025

The Nillion Standard: The Intelligent Internet

Nillion

Nillion's vision for the intelligent internet — introducing blind computation as the foundation for data sovereignty, modular privacy, and a new economic model built on privacy-preserving infrastructure.

Whitepaper2025

Nillion: Humanity's First Blind Computer

Tristan Litre, Miguel de Vega, David Butler

Overview of the Nillion Network architecture: the Coordination Layer, Petnet, nilVM, nilDB, nilAI, and the cryptographic technologies enabling private storage and computation on encrypted data.

Whitepaper2025

Securing an Agentic World

Nillion

How Nillion's blind computation infrastructure addresses the security challenges of autonomous AI agents, from secure key management to private inference and delegated authority.

Whitepaper2025

HealthBlocks: The Era of User-Owned Health Data

Nillion

How HealthBlocks leverages Nillion to give individuals ownership and control over their health data, enabling privacy-preserving analytics and preventative health insights.

Case Study2025

Nubila: Enabling Decentralized Weather Intelligence

Nillion

How Nubila uses Nillion's infrastructure to build a decentralized weather data network, improving forecast accuracy with privacy-preserving local sensor data aggregation.

Case Study2025

Pindora: The Emerging Sovereign Layer

Nillion

How Pindora is building a sovereign agent layer on Nillion, enabling AI agents to securely manage calendars, documents, and wallets with private delegation and blind verification.

Case Study2025

Curl: Private LLMs through Wavelet-Encoded Look-Up Tables

Manuel B. Santos, Dimitris Mouris, Mehmet Ugurbil, Stanislaw Jarecki, José Reis, Shubho Sengupta, Miguel de Vega

CAMLIS 2024

An MPC framework that evaluates non-linear functions as wavelet-compressed lookup tables, achieving up to 19× round and communication reduction for LLM inference on BERT, GPT-2, and GPT Neo.

Conference2024

Ripple: Accelerating Programmable Bootstraps for FHE with Wavelet Approximations

Charles Gouert, Mehmet Ugurbil, Dimitris Mouris, Miguel de Vega, Nektarios G. Tsoutsos

ISC 2024

A framework using discrete wavelet transforms to decrease the number of entries in homomorphic lookup tables while maintaining high accuracy, improving FHE runtime for logistic regression and cross-correlation benchmarks.

Conference2024

More Efficient Comparison Protocols for MPC

Wicher Malten, Mehmet Ugurbil, Miguel de Vega

IACR ePrint 2023/1934

Two new comparison protocols that reduce online multiplications by more than half for logarithmic-round protocols and by two-thirds for constant-round protocols, without increasing preprocessing or round complexity.

ePrint2023

Evaluation of Arithmetic Sum-of-Products Expressions in LSSS with a Non-Interactive Computation Phase

Miguel de Vega, Andrei Lapets, Stanislaw Jarecki, Wicher Malten, Mehmet Ugurbil, Wyatt Howe

IACR ePrint 2023/1740

An information-theoretically secure extension to linear secret sharing schemes that evaluates sum-of-products expressions with multiplications on private values, requiring no communication during the computation phase.

ePrint2023

Secure Truncation with Applications to LLM Quantization

Mehmet Ugurbil, Miguel de Vega, Wicher Malten, Manuel B. Santos

A new framework for secure truncation with perfect security and low memory footprint, reducing communication complexity by 61% and memory usage by 83% for 8-bit fields — especially relevant for quantized LLM inference.

Technical Report2023

Threshold ECDSA in the Preprocessing Setup

Manuel B. Santos, Miguel de Vega, Wicher Malten, Mehmet Ugurbil, Nathalie Regnault

A threshold ECDSA signature scheme in the client-server model combining optimal online round complexity from custodial settings with resilience against collusion, plus protection against user impersonation attacks.

Technical Report2023

Decentralized Multifactor Authentication

Nillion

Technical deep-dive into MPC protocols for exact and approximate matching in decentralized MFA, covering biometric authentication, email/phone verification, and wallet-based identity across distributed trust domains.

Technical Report2023