VirtEngine

Learn · Security architecture

Confidential computing on VirtEngine

Confidential computing — running workloads inside hardware-isolated enclaves the host cannot inspect — solves a technical problem. VirtEngine solves the adjacent marketplace problem: proving to a paying counterparty that confidentiality actually holds, before secrets are delivered.

Security architecture

The trust gap in rented compute

When a workload runs on someone else's hardware, the operator can normally read its memory. Trusted execution environments (TEEs) close that hole in silicon, producing attestations: hardware-signed evidence of exactly what code, in exactly what configuration, is running inside the enclave.

But an attestation is only useful if the counterparty can verify it and act on it. That is the part VirtEngine puts on-chain.

Security architecture

Attestation as chain state

The enclave module (x/enclave) records and verifies TEE attestations against the state machine's expectations. A provider's confidential-compute capability becomes a verifiable on-chain claim — and a filterable attribute. Tenants can constrain orders to attested enclave execution, so unverified capacity never even matches.

Security architecture

Secrets released only after proof

The encryption module (x/encryption) implements envelope encryption to specific recipients. In confidential workflows, workload secrets — keys, model weights, sensitive configuration — are sealed so they can only be delivered into an enclave whose attestation has verified. The sequence is proof first, secrets second, and it is enforced by protocol machinery rather than provider goodwill.

Security architecture

The rest of the assurance stack

Confidential workloads inherit the marketplace's full trust apparatus: VEID-verified counterparties, chain-anchored mTLS (x/cert) on every off-chain connection, auditor-signed provider attributes (x/audit), and fraud enforcement with dispute intake if conduct fails. For enterprises, this is the difference between a confidentiality feature and a confidentiality argument that survives procurement review.

Security architecture

Where to apply it

The pattern fits wherever data or models must not be exposed to the infrastructure operator: regulated datasets, proprietary model weights during training or inference, key-handling services, and multi-party computations where participants trust the enclave but not each other.

Related

Continue from here