Homomorphic Encryption in iOS 18

This blog post explains homomorphic encryption for readers without a maths background, using it as a way to understand the search technology Apple uses in Photos on iOS 18 and macOS 15. It describes how a server can compute on encrypted values without reading them: the client encrypts a value, the server applies addition or multiplication to the ciphertext, and the client decrypts the result. The author notes that practical programs must be restructured around these operations, for example a private database lookup using a dot product, and that the approach runs many orders of magnitude slower than plaintext equivalents. The post then begins a high-level overview of Apple's Scalable Private Search with Wally paper, which uses somewhat homomorphic encryption to return encrypted similarity scores instead of full database entries.