{
  "id": 9564540,
  "title": "Combining Machine Learning and Homomorphic Encryption in the Apple Ecosystem",
  "url": "https://urgent.news/2026/09/24/combining-machine-learning-and-homomorphic-encryption-in-the-apple",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-24T13:27:42.000Z",
  "source": {
    "name": "Lobsters",
    "slug": "lobsters",
    "url": "https://machinelearning.apple.com/research/homomorphic-encryption"
  },
  "original_language": "en",
  "account": "Apple emphasizes user privacy as a fundamental right and strives to incorporate this principle into its products and services. One method used is to perform computations locally on user devices, reducing the amount of data shared with Apple or other entities. To offer machine learning (ML) powered experiences while maintaining privacy, Apple combines homomorphic encryption (HE) with other privacy-preserving technologies. HE is a cryptographic method that enables computation on encrypted data, ensuring that a server can operate on encrypted queries without ever decrypting the original request or having access to the decryption key.\n\nApple employs HE in conjunction with private information retrieval (PIR) and private nearest neighbor search (PNNS) to provide private server lookups and ML capabilities. With PIR, a client encrypts its keyword and sends it to the server, which conducts HE computation between the incoming ciphertext and its database. The server then sends the encrypted value back to the client, which decrypts it to learn the associated value without revealing the original query. In use cases where an approximate match is required, Apple's PNNS is used to encrypt a vector embedding and send it to the server for a nearest neighbor search. The server then decrypts and sends back the encrypted value, allowing the client to learn the nearest neighbor to its query embedding, also without exposing the client's private embedding or retrieved results.\n\nBy integrating these privacy-preserving techniques into the Apple ecosystem, the company ensures that user privacy is protected while enhancing on-device experiences with private server lookups. To support developers in adopting HE for their applications, Apple has open-sourced swift-homomorphic-encryption, an HE library. The implementation of HE in Apple's ecosystem focuses on being efficient at scale while maintaining high security, specifically utilizing the Brakerski-Fan-Vercauteren (BFV) HE scheme. The BFV scheme supports homomorphic operations suitable for ML workflows, such as dot products and cosine similarity, and has been designed to provide post-quantum 128-bit security, protecting against both classical and potential future quantum attacks.",
  "summary": null,
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}