Most AI providers ask you to trust their privacy policy. Venice is trying something different: hardware-enforced encryption that makes it technically impossible for anyone - including Venice itself - to read your prompts.
The company launched two new privacy modes on March 18. The first, Trusted Execution Environment (TEE), runs AI inference inside secure hardware enclaves operated by partners NEAR AI Cloud and Phala Network. These enclaves produce cryptographic certificates that prove the model is running in genuine secure hardware, and users can verify those certificates independently. TEE mode supports the full Venice feature set including web search, file uploads, and memory.
The second mode, End-to-End Encrypted (E2EE), goes further. Your prompts are encrypted on your device before they leave your browser, stay encrypted through Venice's servers, and only get decrypted inside a verified hardware enclave. Each response is cryptographically signed. The tradeoff: E2EE disables web search, memory, and file uploads since those features require Venice's servers to see your data.
11 Models, All Open-Source
Both modes work with eleven models, all open-source or open-weight: Venice Uncensored 1.1, several GLM variants (4.7, 4.7 Flash, 5), Qwen models (2.5 7B through 3.5 122B), Gemma 3 27B, and two GPT OSS models (20B and 120B). No Claude, no GPT-4 - those are closed models that can't run in third-party enclaves.
This matters because Venice already positioned itself as the privacy-focused AI alternative, but its previous "Anonymous" and "Private" modes relied on policy commitments - essentially promises not to log or read your data. The new modes replace trust with math. A cryptographic attestation is harder to break than a terms-of-service update.
The practical question is performance. Running inference inside hardware enclaves adds overhead, and the available models top out at Qwen 3.5 122B - capable, but not frontier-class. For users whose threat model demands verifiable privacy over raw capability, that is a reasonable trade. For everyone else, the existing anonymous mode probably still makes more sense.