For the complete documentation index, see llms.txt. This page is also available as Markdown.

Trust Model

Verification is only as strong as what the SDK trusts. The chain is:

  • Intel SGX Root CA β€” pinned in code; collateral is fetched from PCCS over HTTPS but the chain is anchored locally.

  • AMD ARK β€” pinned per product (Milan, Genoa, Turin) by SHA-256 of its SubjectPublicKeyInfo; HTTPS to kdsintf.amd.com is not load-bearing for trust.

  • NVIDIA NRAS JWKS β€” JWT signatures are verified against NVIDIA's published JWKS. The NRAS root CA is not currently pinned (NVIDIA does not publish it); the JWKS is trusted via HTTPS.

  • SecretVM registry β€” bundled with each release and refreshed from github.com/scrtlabs/secretvm-verify on a miss. Refresh content is trusted via HTTPS to GitHub.

The SDK does not trust the VM itself, the operator, or the network path beyond what these roots permit.

Was this helpful?