security
Intended Documentation
Security and Compliance
How Intended enforces authority — the per-tenant SHA-256 audit hash chain, HMAC-signed evidence export, fail-closed defaults, and the runtime mechanisms a security reviewer should evaluate.
Security and Compliance#
Intended is an authority runtime: it decides whether an AI action may proceed, and it records that decision so the answer to "why was this permitted?" is provable after the fact. This section documents the mechanisms the platform actually implements — a per-tenant cryptographic audit chain, HMAC-signed evidence export, fail-closed defaults, and single-use authority tokens — so a security reviewer can evaluate them directly against the code.
What this section does and does not claim
Intended provides tamper-evidence mechanisms (an immutable hash chain, signed evidence bundles, tamper detection). It does not ship certified compliance-framework control mappings. The runtime encodes no SOC 2, ISO 27001, or EU AI Act control IDs in the audit chain, and Intended is not itself a certification authority. Where a page references a compliance program, it describes evidence and workflow tooling, not an attestation Intended can issue on its own behalf.
What a reviewer should evaluate#
| Mechanism | What it guarantees | Where it is documented |
|---|---|---|
Per-tenant SHA-256 hash chain (AuditEntry) | Tamper-evident, append-only decision history; any edit breaks the chain | Enforcement Lineage |
HMAC-signed evidence export (generateEvidenceBundle) | A self-contained bundle whose integrity is verifiable by a holder of the tenant secret | Enforcement Lineage |
| Default-DENY runtime | No action proceeds without an explicit allow; failures deny | Fail-Closed Controls |
| Single-use Ed25519 authority tokens | Replay-resistant, short-lived (300 s) execution authorization | Fail-Closed Controls |
| Per-tenant key isolation (Ed25519, AES-256-GCM at rest) | Cryptographic tenant separation and key rotation | Tenant Trust Boundary |
Pages in this section#
- Enforcement Lineage — the audit hash chain, what it links, and how to verify it
- Fail-Closed Controls — default-deny behavior and the exact failure modes
- Operational Readiness — verifying health and integrity before going live
- Third-Party Runtime Boundaries — shared responsibility for governed external runtimes
- OpenShell Integration Threat Model — threats and controls for OpenShell / NemoClaw governance