Nillion's cryptography team is cooking hard! Our Fission paper just got accepted at the 2025 IEEE Conference on Communications and Network Security, @ComSoc. _____ Private LLM inference has been stuck in a bad tradeoff: You can either use cryptography like MPC or FHE and stay private but slow, or go fast using enclaves, but risk leaking data. Nillion's Fission solves that tradeoff. It’s a new way to run large language models securely and efficiently by combining MPC with a network of fast, secure evaluators (which can run in secure enclaves). _____ Fission runs sensitive parts of the model under MPC. Then it splits the intermediate outputs, shuffles them across multiple evaluator parties and has them compute the next layers. Once done, the outputs re-enter MPC for the next round. MPC Evaluate → Shuffle → Split → Evaluate → Reshare → Repeat. This lets us keep the private info protected, while offloading compute heavy chunks to trusted hardware, without ever giving one party too much. Fission is a step toward making private LLMs practical, without sacrificing speed or security. Dive into the paper here: