This document discusses zero-knowledge proofs and their application to machine learning models through a technique called Zero-Knowledge Machine Learning (ZKML). ZKML allows machine learning models to generate cryptographic proofs that verify computations on private inputs without revealing the inputs or model. This could enable applications like a machine learning marketplace with verifiable predictions on sensitive data or training models privately. Some challenges to ZKML include the expensive computation of zero-knowledge proofs and representing models with arithmetic circuits. The author's company, Attic 42, is working on tools to address these challenges and enable practical applications of zero-knowledge techniques like ZKML.