Brave adds end-to-end encryption for AI chats with its Leo assistant

[Natalie91]

Brave has given its Leo AI assistant an end to end encrypted mode for chats. The Register's Thomas Claburn reports that Leo now runs Z.ai's GLM 5.3 Flash model inside a trusted execution environment operated by NEAR AI. It builds on an experiment that began in November 2025

The setup is clever. A prompt is encrypted on the user's device and sent using Oblivious HTTP, so Brave's AI gateway forwards it without seeing the user's IP address. The browser checks the secure enclave's attestation, then encrypts the request with a key tied to that enclave. Only the enclave can decrypt the request, run the model and send back an answer that only the browser can read

The hardware behind it uses Intel TDX and Nvidia Confidential Compute. The result is that Brave cannot see what was asked, and NEAR AI cannot see who asked. Signed responses let users check that the model has not been swapped for a different one. Brave Search queries can also run from inside the secure environment

Chief executive Brendan Eich said too much AI runs on trust me bro promises. Brave still plans further improvements, including moving more verification into the browser, publishing a log of enclave keys and attesting the exact model weights loaded in memory

Privacy has been one of the biggest concerns about AI chatbots, because people share very personal information with them. Technical guarantees rather than policy promises are exactly the right approach. Hopefully other AI providers take note

Would encrypted AI chats make you more comfortable using AI assistants? And should this become the standard?

Want your AI chats to stay private? Brave TEEs up end-to-end encryptionwww.theregister.com