Back to collectionResearch
Brave Leo Adds Verifiable Private Inference with NEAR AI
Brave shipped end-to-end encrypted GLM 5.3 Flash chats backed by attested NEAR AI hardware enclaves across its release channels.
Published
Brave and NEAR AI announced on October 15 that Leo chats using GLM 5.3 Flash now run through end-to-end encrypted private inference. Brave says prompts are encrypted in the browser, relayed without readable content, and decrypted only inside an attested trusted execution environment operated by NEAR AI; Brave’s public verifier checks the environment before the browser receives its encryption key. Brave reports the feature is available in Release, Beta and Nightly. The design reduces trust in service operators, but still depends on the stated hardware, attestation and public-code verification chain.