Trusted processing: the lock and the key

A trusted processing unit is a vault built out of the computer's own hardware. The technical names are trusted execution environments and confidential virtual machines — AMD SEV-SNP, Intel SGX and AWS Nitro are the three the system supports. Inside one, the processor itself encrypts the running memory, so even the machine's owner, looking at their own hardware, sees only ciphertext.

Ordinary servers have doors: an administrator account, a debug console, a memory dump. A trusted processing unit is designed with no doors at all — no operator login, no console, no way to read memory from outside. That is not a policy someone promises to follow; it is how the silicon works.

Before encrypted evidence is released to a vault, the vault must prove what it is. It produces an attestation — a signed statement from the hardware manufacturer's own certificate chain that says, in effect: "I am genuine sealed hardware, and I am running exactly this code, with this measurement."

The VirtEngine chain verifies that attestation on-chain, in its x/enclave module, checking the signature and comparing the code's measurement — a cryptographic fingerprint of every byte the enclave will run — against the approved record. If a single byte of code were different, the measurement would not match and no data would be released to it. Because the check is on-chain, it is public: anyone can confirm which enclave measurements the network accepts.

The key that decrypts a scope never exists outside the enclave. It is derived inside the hardware, from the hardware's own secrets, and sealed so it can only ever be used by that exact enclave running that exact code — hardware-based key derivation with vTPM sealing, in the technical terms.

This is why "no one can open it" is a design fact rather than a promise: there is no key to hand over. Not to an operator with root access, not to the provider hosting the machine, not to the foundation that stewards the protocol, not in response to a demand. A key that never leaves the silicon cannot be copied, stolen, or surrendered.

Attested enclave capability is recorded in x/enclave, and orders can require it as a placement constraint — so unverified capacity never even matches. Workload secrets are sealed with x/encryption envelopes and released only after attestation verifies. The sequence is proof first, secrets second, and it is enforced by protocol machinery rather than provider goodwill.