Security model

Trust boundaries should be inspectable.

This page states what Silentnox stores, what it protects against, and where the model ends. Treat it as an integration guide, not a substitute for your own threat analysis.

Cryptography

  • HKDF derives a per-request data-encryption key from shards K and S.
  • AES-256-GCM provides confidentiality and integrity.
  • Project keys are generated once and must remain server-side.

Silentnox retains

  • A protected platform shard
  • Project and record metadata
  • Operational and audit events

Silentnox does not retain

  • Your project key
  • Encrypted field blobs
  • Field plaintext after a request

Threats the split-custody model addresses

It reduces the impact of a database-only compromise, a Silentnox storage-only compromise, and accidental plaintext persistence in your application database. An attacker needs access to both custodians or to the live runtime path where both inputs meet.

Explicit limitations

  • A compromised application runtime may observe the project key, blob, and returned plaintext.
  • Incorrect logging in your application can leak request inputs or responses.
  • Client-side use exposes the project key and is unsupported.
  • Availability depends on access to both key shards and the Silentnox service.
  • Key rotation requires re-encrypting values that must use the new key.

Integration requirements

  • Keep the project key in an environment variable or managed secrets service.
  • Call encryption and decryption endpoints from trusted backend code only.
  • Never log plaintext, project keys, or complete request payloads.
  • Restrict outbound network access and rotate credentials after suspected compromise.
  • Review audit logs and disable a project key when activity is unexpected.
Implement the backend flow →