Trust Centre

Trust Centre

What AWARE actually does, what it depends on, and where its limits are. Only claims with documentary evidence appear here.

Trust boundaries

Agent boundary

The agent is untrusted. It holds a credential and presents it. It is never trusted to authorise itself.

Control plane

Gateway, coordinator, identity registry and policy engine are the trust boundary’s inside. Decisions are made here, not by the agent.

Target boundary

The system being acted on. Today AWARE does not verify what happens inside this boundary — see execution assurance.

Cryptographic mechanisms

MechanismAlgorithmPurpose
Credential storagePBKDF2, per-credential salt, server-side pepperStored credentials are not recoverable
Credential comparisonConstant-timeNo timing oracle on verification
Credential generation32 bytes CSPRNGUnguessable provisioning secrets
Session tokenJWT, HS256Carries trust domain, blast radius, session id, execution context
Audit chainSHA-256 over canonical serialisation plus previous hashTamper-evident decision history
Key-bound identity—Not implemented. No X.509, Ed25519 or SPIFFE.

Compliance mappings

Seven frameworks are registered in the framework mapper, with the versions below.

FrameworkVersionCoverage
CSA AI Controls Matrixv1184 of 243 control IDs, across all 18 domains
NIST AI Risk Management Framework1.0Govern, Map, Measure, Manage
ISO/IEC 270012022Annex A control mapping
ISO/IEC 42001v1.0-2026-07-28AI management system controls
OWASP Agentic Skills Top 10v1.0-2026Full catalogue
OWASP Top 10 for MCP202510 of 10 classes
DORA2022Operational resilience mapping
MITRE ATLASv2026.06Technique mapping, 7 initial rule functions
A mapping is not a certification Registering a control against a framework supports preparation. It is not an audit opinion, not a certificate, and not a guarantee that a given requirement is met. The independent certification body decides. AWARE never guarantees an auditor’s decision.

Security testing methodology

  • Unit and integration suites run on the standard Node test runner, with separate standards suites per compliance framework.
  • Playwright covers browser-level behaviour of the operator UI.
  • Dependency and secret scanning run as part of the release process: gitleaks, npm audit and trivy reports are kept in the repository.
  • The npm publication boundary is enforced by a publish guard that packs from a clean detached worktree at the release tag and refuses to publish a tarball that differs from it.

Vulnerability disclosure

AWARE publishes a security policy and a SECURITY.md with a coordinated disclosure route. Report a vulnerability rather than disclosing it publicly. Contact: security@goodciso.org. A security.txt is served under /.well-known/.

Data handling

AWARE stores agent identity records, policy state and audit records. Credentials are stored as salted, peppered hashes and are never recoverable. Audit records can be exported and are intended to be retained by the operator; the retention period is the operator’s choice and is not imposed by the product.

Claims we do not make AWARE is not SOC 2 certified. It is not ISO certified. It has not been penetration tested by an independent firm. We do not claim enterprise customer deployments we cannot evidence. The product’s maturity is described by the capability register, and that register is public.