Private Inference (Confer Blog) (confer.to)

Confer has unveiled a groundbreaking approach to AI inference that emphasizes user privacy by employing confidential computing techniques. Unlike traditional AI services that handle user prompts in plaintext—which can lead to potential data breaches and exploitation—Confer encrypts user inputs from the device to a secure Trusted Execution Environment (TEE). This setup utilizes hardware-enforced isolation to ensure that even the server operators do not have access to the plaintext of prompts or responses, thereby significantly enhancing user data confidentiality. The implementation features remote attestation, which verifies the integrity of the code running inside the TEE, ensuring that it functions as intended. With a combination of cryptographic techniques, such as Noise Pipes for encrypted data transmission and ephemeral session keys for improved security, users can have confidence that their interactions are shielded from unauthorized access. This innovative model not only protects user data but also highlights a critical shift in AI service design towards prioritizing privacy and security, addressing longstanding concerns over data handling in machine learning applications.