GitHub project demos cat classification under fully homomorphic encryption

jedisct1 · x · 2026-07-28

A GitHub project demonstrates privacy-preserving image classification under fully homomorphic encryption (FHE).

The system runs a small CNN that decides cat / not-cat on a 32×32 image while the inference server sees only ciphertext. The server never gets the plaintext image, the result, or the secret key.

It was built for the IETF Hackathon and implements the design in an Internet-Draft on ciphertext-based AI inference for MCP. The repo ships as ready-to-run Windows and Linux tools, with no Python source build required.

Original post →

More from Research

Research channel →