Critical Assessor Notice: Boundaries of Cryptographic Evidence

SWT3 witness anchors prove that specific operational controls were active at a specific point in time. They do not replace the assessor's independent judgment, professional expertise, or regulatory authority. Assessors must verify that anchored evidence is sufficient, appropriate, and relevant to the specific assessment context. Each regulatory framework retains its own assessment authority, methodology, and determination standards.

1

Who This Guide Is For

This guide provides a structured onboarding path for any assessor encountering SWT3 cryptographic evidence for the first time. Two assessment tracks are supported:

Conformity Assessment Path

For Notified Bodies and conformity assessment bodies performing assessments under the EU AI Act (Article 43) or other regulatory frameworks requiring third-party evaluation.

Key guide: EU AI Act Conformity Assessment Evidence Guide

Compliance Assessment Path

For C3PAOs, auditors, ISSMs, and internal compliance teams performing assessments under NIST 800-53, CMMC, AI RMF, or other compliance frameworks.

Key guide: C3PAO Assessment Guide

2

What is SWT3 Evidence

SWT3 (Sovereign Witness Traceability) is a cryptographic witness protocol. It generates deterministic, independently verifiable attestation records -- called witness anchors -- that encode observed operational facts about AI systems.

What SWT3 IsWhat SWT3 Is Not
A deterministic cryptographic attestation computed from observed factsA log entry or assertion that could be edited after the fact
Independently verifiable using standard SHA-256 (no proprietary tools)A certificate issued by a trusted authority
Framework-neutral (maps to EU AI Act, NIST, CMMC, and 33 other frameworks)A compliance product tied to a single regulation
Evidence that controls were operational at runtimeEvidence that policies or documentation exist

The Three-Factor Matrix

Every witness anchor encodes exactly three numeric factors that capture the observed measurement:

Example: For AI-GRD.1 (Guardrail Enforcement), factor_a=2 means 2 guardrails required, factor_b=3 means 3 guardrails were active, factor_c=0. Verdict: PASS (3 >= 2).

Clearing Levels

Clearing levels control what metadata accompanies the anchor. Higher levels remove more metadata before transmission:

LevelNameWhat SurvivesAssessor Implication
0AnalyticsAll metadata (model ID, hashes, tokens, latency, context)Full forensic capability
1StandardHashes + model ID + context (raw evidence purged)Recommended default. Sufficient for most assessments.
2SensitiveHashes + model ID onlyReduced context. Check if traceability requirements are met.
3ClassifiedFactors only. Model ID hashed.Proves a fact occurred without revealing model identity or operational details.
CJT fields survive all clearing levels. Jurisdiction, legal basis, purpose class, agent ID, and cycle ID are never cleared, regardless of level. These fields ensure regulatory traceability across jurisdictions.
3

Assessor Competence Requirements

Evaluating SWT3 evidence requires the following technical knowledge. No proprietary tooling, platform access, or vendor-specific training is needed.

CompetenceLevel RequiredWhy
SHA-256 hash computationBasicFingerprint verification requires computing a SHA-256 hash and comparing the first 12 hex characters
JSON document structureBasicCompliance passports and witness payloads are JSON documents
Hexadecimal notationFamiliarityFingerprints, hashes, and signatures are hex-encoded
Anchor format decompositionFamiliarityReading the structured fields in an anchor token string
Verification uses standard cryptographic primitives. Any SHA-256 implementation (command-line sha256sum, Python hashlib, online calculator, or the public verifier at sovereign.tenova.io/verify) is sufficient. No proprietary SDK or API key is required.
4

How to Verify an Anchor

Obtaining Verification Inputs

An anchor token contains the fingerprint but not the inputs needed to verify it. To perform verification, request the witness payload JSON from the provider. This is available via the anchor chain export (JSON), the Compliance Passport, or the platform API. The payload contains the tenant ID, procedure ID, factors, and millisecond timestamp needed for recomputation.

Worked Example

Given this anchor token:

SWT3-E-VULTR-AI-AIINF1-PASS-1774800000-2e16e2fe92dd

And these input components (from the witness payload JSON):

ComponentValue
Tenant IDENCLAVE_PROD
Procedure IDAI-INF.1
factor_a1
factor_b1
factor_c0
Timestamp (ms)1774800000000

Step 1: Construct the fingerprint input string:

WITNESS:ENCLAVE_PROD:AI-INF.1:1:1:0:1774800000000

Step 2: Compute SHA-256 of that string (UTF-8 encoded):

$ echo -n "WITNESS:ENCLAVE_PROD:AI-INF.1:1:1:0:1774800000000" | sha256sum 2e16e2fe92dd... (full 64-char hex digest)

Step 3: Take the first 12 characters: 2e16e2fe92dd

Step 4: Compare to the anchor's claimed fingerprint (last segment): 2e16e2fe92dd

Result: VERIFIED. The fingerprints match. The anchor's factors, tenant, procedure, and timestamp are confirmed authentic.

Public Verifier

For convenience, the public verifier at sovereign.tenova.io/verify provides a web interface for single and bulk anchor verification. Enter an anchor token and the verifier recomputes the fingerprint. No account required.

Enclave Integrity Check

To verify an entire collection of anchors:

  1. Collect all anchor fingerprints
  2. Sort lexicographically (ascending)
  3. Join with colons: fp1:fp2:fp3:...
  4. Compute SHA-256 of the joined string (full 64-char hex)
  5. The result is the enclave integrity signature. Any modification to any anchor changes this value.
5

How to Interpret a Compliance Passport

A Compliance Passport is a signed summary of a tenant's compliance posture. It is available in three formats:

FormatTierVerification
HTML (printable report)Pro+Visual review
JSON (HMAC-signed)Pro+Server-side HMAC verification via POST /api/v1/passport/verify
W3C Verifiable CredentialEnclave+Ed25519 signature verification, DID resolution via did:web:sovereign.tenova.io

Key Fields

The W3C Verifiable Credential format is self-verifiable. The VC is signed with Ed25519 (eddsa-jcs-2022 proof suite). The public key is in the DID document at did:web:sovereign.tenova.io. Any W3C VC verifier can validate it without contacting the issuer.
6

How to Incorporate SWT3 Evidence into an Assessment

Notified Body / Conformity Assessment

SWT3 anchor chains provide the runtime evidence layer for conformity assessments under Article 43 of the EU AI Act.

  • Map witnessed procedures to Art. 9-15 and Art. 51-56 requirements
  • Use anchor fingerprint verification to confirm evidence integrity
  • Evaluate coverage breadth (which procedures) and depth (continuous vs sampled)
  • Identify supplementary evidence needed for PARTIAL coverage articles

Detailed guide: EU AI Act Conformity Assessment Evidence Guide

Integration with ISO/IEC 17065: SWT3 evidence can be referenced in the assessment body's quality management system as a standardized evidence source. The deterministic verification algorithm provides a repeatable, auditable evaluation procedure.

C3PAO / Auditor / Internal Assessment

SWT3 anchor chains map to NIST 800-53A assessment objectives and CMMC practice families through the UCT Registry crosswalk engine.

  • Each SWT3 procedure maps to one or more NIST 800-53 controls
  • AI-INF.1 maps to AU-2 (Event Logging), AI-GRD.1 maps to SI-10 (Information Input Validation)
  • Anchor chains provide continuous monitoring evidence for CA-7
  • Use the Assessor Evidence Matrix for cross-framework mapping

Detailed guide: C3PAO Assessment Guide

7

Assessment Report Template

Evidence Citation Format

When referencing SWT3 anchors in assessment findings, use the full anchor token string as the evidence citation:

[SWT3-E-VULTR-AI-AIINF1-PASS-1774800000-2e16e2fe92dd]

Include the following context with each citation:

Sample Findings

Finding: Conformity (Art. 12 -- Record-Keeping) The provider demonstrates continuous automatic recording of inference events via AI-INF.1 witness anchors. A 30-day sample was reviewed: 47,231 anchors with 100% fingerprint recomputation success rate. Clearing level 1 preserves model identity for traceability. Evidence: [SWT3-E-VULTR-AI-AIINF1-PASS-1774800000-2e16e2fe92dd] and 47,230 additional anchors. Conformity with Art. 12(1) is established.
Finding: Non-Conformity (SC-7 -- Boundary Protection) AI-GRD.1 anchors are absent for the customer-facing chatbot deployment. The provider's SSP documents guardrail enforcement as a compensating control for SC-7, but no witness anchors were found for the assessment period. Non-conformity. Corrective action: implement AI-GRD.1 witnessing for all production AI deployments referenced in compensating controls.
Observation (AI-DRIFT.1 -- Model Drift) AI-DRIFT.1 anchors show drift monitoring is active with a 30-day cadence. However, no AI-DRIFT.2 (consequence-mapped drift) lifecycle chains were observed. The provider may benefit from implementing automated drift response workflows to strengthen continuous monitoring evidence. Opportunity for improvement.
+

Working in the Audit Portal

When an organization shares an audit portal link with you, the portal provides interactive tools beyond what the SDK and CLI offer:

The portal also features a Guided Walkthrough for first-time users that covers the score overview, framework toggle, control matrix, lifecycle chains, and verification workflow.