The Leo AI assistant in the Brave browser now uses hardware enclaves to ensure that user prompts remain private from the company and its infrastructure partners. By integrating private AI inference from NEAR AI Cloud, Brave now offers end to end encrypted chat using the GLM 5.3 Flash model. The system encrypts prompts on the user’s device and decrypts them only within an attested Trusted Execution Environment (TEE), meaning neither Brave, NEAR AI, nor the cloud host can read the conversation.
Brave describes the feature as industry leading privacy that is verifiable by the user. Before the browser transmits any data, it checks the status of the secure hardware enclave to confirm the environment is protected. This architecture aims to remove the need for users to trust the AI service provider with their plaintext data.
