An AI Builder’s Request through the Attested Sealed Compute Lane

The starting point

Illustrative flow: an AI builder has a workload that requires a defined confidentiality boundary. Routing it to an ordinary consumer GPU would not make the prompts private from the operator. The sealed-lane design instead requires eligible confidential GPUs and a TDX environment with current attestation.

The intended flow

The builder will select the sealed lane (in testing) in the API request. The OHTTP relay passes the encrypted body to the gateway and the eligible node decrypts it within the confidential environment. The completed job returns a response and a signed receipt linked to environment, model and GPU evidence. The design leaves the host operator with ciphertext.

Design outcome

The intended outcome is private execution inside the defined sealed-lane boundary, with attested work records. This is an illustrative architecture flow rather than a customer deployment or a security audit. The open lane is not confidential, and the disclosed TEE limitations still apply.