Confidential inference and zero data retention protect different parts of an AI request

Compare evidence for data handling during computation with promises about storage after a request.

Follow a request through three places: the client that assembles it, the workload that processes it and any system that keeps a copy. Confidential computing can protect data in use inside a measured environment. A retention agreement governs what the covered party may keep.

Find the boundary of the evidence

Draw where encryption starts and terminates. Ask which workload produces the attestation and whether its measurement covers the inference process. Then check gateway logging and client observability separately. Even a verified workload leaves questions about a copy recorded before the request entered it.

Compare OpenRouter alternatives · What evidence should you keep when switching AI gateways?