It uses confidential computing for frontier Gemini models - hardware-enforced privacy, not just policy promises.
Not perfect, but radically better.
This should be table stakes!
OpenAI, Anthropic, every frontier lab: make this the new baseline.
This is like what Apple did with Private Cloud Compute, but with a frontier model.
Anthropic does use Confidential Compute for inference, but mostly to protect their weights from being stolen by the cloud host… for you as a user, they don't prove that they aren't logging via remote attestation.