Brave launches end-to-end encrypted AI chat with GLM 5.3 Flash

Brave announced that Leo chats with the open-weight GLM 5.3 Flash model are now end-to-end encrypted and run inside attested, NVIDIA-backed Trusted Execution Environments (TEEs) operated by NEAR AI. The browser encrypts each prompt on-device using Oblivious HTTP (OHTTP), so only the attested TEE can decrypt it — neither Brave nor NEAR AI can read chats, and history stays on-device.

• Before any prompt leaves the browser, the TEE must present hardware-signed attestation verifying the hardware, model, and auditable code — cryptographic proof rather than a privacy promise.
• The encrypted prompt travels through Brave's open-source AI Gateway, which also hides the user's IP address from NEAR AI.
• The feature is available in Brave Release, Beta, and Nightly, building on Brave's earlier "Verifiably Private" work from November 2025.
• Brave's senior privacy researcher Ali Shahin Shamsabadi framed the launch as shipping "an answer in production" to proving an AI system is private — describing Leo chats as traveling "like a sealed letter" that only a verified NVIDIA TEE run by NEAR AI can open.
• NEAR Foundation AI director Cameron Dennis celebrated that NEAR AI is now available to Brave's 128M+ monthly active users; Brave also retweeted coverage of the launch from Thurrott alongside messages from NEAR Protocol and co-founder Illia Polosukhin.