Techniques to prove ML model inference was executed correctly without revealing model weights or inputs, enabling private scoring and verifiable on-chain AI.