Chio/Docs

PlatformCompliance Mappings

ISO/IEC 42001

A technical mapping from Chio controls to selected ISO/IEC 42001:2023 clauses for an operator's Statement of Applicability and self-assessment.

Technical mapping for self-assessment

This page maps Chio's tool-call controls and signed evidence to selected ISO 42001 clauses. An operator uses the mapping as input to its own Statement of Applicability, policies, management review, and assessment.

Where Chio Contributes

Chio contributes to operational clauses and reference controls for logging and third-party relationships.

Clause FamilyChio Role
Clause 7.5 Documented informationSigned receipts, checkpoints, compliance certificates, and evidence bundles. Each signed record uses canonical JSON with a timestamp and key reference.
Clause 8.2 and 8.3 Operational control and life cycleGuard pipeline, capability lifecycle (issue, delegate, revoke, expire), tool-server lifecycle evidence via signed manifests.
Clause 9.1 Monitoring and measurementReceipt query API, checkpoint monitor, receipt analytics, SIEM export.
Annex A.7 AI system loggingSigned receipt per invocation, Merkle-committed batches, DPoP attribution.
Annex A.5 Data for AI systemsData-layer guards: SQL and warehouse access guards, vector store constraints, column-level restrictions, response sanitization.

Clauses 4 (context), 5 (leadership), most of 6 (planning), 7.1 through 7.4 (resources, competence, awareness, communication), and 10 (improvement) are largely organizational and are marked org-owned below.


Coverage Legend

LevelMeaning
supportsChio's shipped controls directly implement the clause or control at the tool-governance layer.
evidence-forChio supplies part of the control; the organization adds procedures, documents, or review cadence.
org-ownedClause is organizational, procedural, or policy-level; Chio does not implement it.
out-of-scopeAddresses concerns outside Chio's governance boundary (e.g., training-data curation, workforce).

Main Body Clause Mapping

Clauses 4 and 5: Context and Leadership

ClauseChio MappingCoverage
4.1 / 4.2 Context and interested partiesNot applicable at the protocol layer.org-owned
4.3 Scope of the AIMSCapability issuance demarcates which tool servers are inside the governed scope; deployment-level scope is documented per site.evidence-for
4.4 AI management systemOperational layer (guards, receipts, revocation) runs continuously and produces signed artifacts.evidence-for
5.1 Leadership and commitmentNot applicable at the protocol layer.org-owned
5.2 AI policyPolicy-as-code artifacts compiled from chio-policy; policy hash embedded in every receipt.evidence-for
5.3 Roles, responsibilities, authoritiesIssuer and delegation chain on every capability attribute technical authority; human roles external.evidence-for

Clause 6: Planning

ClauseChio MappingCoverage
6.1 Actions to address risks and opportunitiesGuard pipeline deny paths, capability scoping, budget caps.supports
6.1.2 AI risk assessmentUnderwriting tier classification; receipt analytics surface aggregates.evidence-for
6.1.3 AI risk treatmentRevocation runtime, scope tightening, approval tokens for step-up review.supports
6.1.4 AI system impact assessmentUnderwriting risk reasons plus receipt impact data; artifact format is operator-owned.evidence-for
6.2 AI objectivesNot applicable at the protocol layer.org-owned
6.3 Planning of changesPolicy hash in every receipt ties each call to a policy version; request-matching ties the call to a request-schema version.supports

The underwriting inputs in the rows above come from chio-underwriting, which classifies an agent into one of four risk tiers: Baseline, Guarded, Elevated, and Critical, from evidence references (receipt history, reputation, certification, runtime assurance, settlement exposure, and metered billing). Its evaluator is fail-closed and emits typed reason codes, such as low_reputation, missing_certification, and pending_settlement_exposure. The tier and its reasons populate the risk column of a 6.1.2 risk assessment and an 8.2 impact assessment; the assessment artifact itself stays operator-owned.

Clause 7: Support

ClauseChio MappingCoverage
7.1 / 7.2 / 7.3 Resources, competence, awarenessNot applicable at the protocol layer.org-owned
7.4 CommunicationSIEM export, trust-control dashboard, compliance certificates.evidence-for
7.5 Documented informationSigned artifacts: receipts, checkpoints, compliance certificates, evidence bundles.supports
7.5.2 Creating and updatingCanonical JSON (RFC 8785) for every signed artifact; timestamp and key reference recorded.supports
7.5.3 Control of documented informationArchival via evidence retention; inclusion proofs guarantee integrity of archived entries.supports

Clause 8: Operation

ClauseChio MappingCoverage
8.1 Operational planning and controlGuard pipeline, grant-scoped operational controls, fail-closed default.supports
8.2 AI system impact assessmentUnderwriting tiers, capability-tier assignment on every grant.evidence-for
8.3 AI system life cycleCapability lifecycle: issue, delegate, revoke, expire. Tool-server lifecycle evidence via signed manifests. Design and training phases are external.evidence-for
8.4 Third-party relationshipsCapability scoping for tool servers; DPoP attribution for outbound calls; manifest verification for counterparties.supports

Clauses 9 and 10: Performance Evaluation and Improvement

ClauseChio MappingCoverage
9.1 Monitoring, measurement, analysis, evaluationReceipt query API, checkpoint monitor, receipt analytics.supports
9.2 Internal auditEvidence export bundles, compliance certificates, Merkle inclusion proofs.supports
9.3 Management reviewCompliance certificates summarize operational posture.evidence-for
10.1 Continual improvementReceipts feed improvement workflows; automated policy evolution is operator-owned.evidence-for
10.2 Nonconformity and corrective actionDeny receipts, revocation, scope reduction, approval escalation.evidence-for

Receipts are the CAPA anchor

Corrective action workflows benefit from the fact that every deny is signed. When an investigator asks "which policy blocked this action?" or "was a mitigation applied?", the signed receipt chain is the answer, not a support ticket.

Annex A Reference Controls

Annex A of ISO 42001 supplies a reference set of AI-specific controls. Control IDs follow the pattern A.x.y. Chio's strongest Annex A contribution is to A.6 (life cycle), A.7 (data), A.8 (information sharing), A.9 (use), and A.10 (third-party).

ControlDescriptionChio MappingCoverage
A.2Policies related to AI.Policy-as-code via chio-policy; policy hash in every receipt.evidence-for
A.3Internal organization.Issuer and delegation chain attribute authority at the protocol layer.evidence-for
A.4Resources for AI systems.Tool manifest catalogs tools; budget caps manage compute spend.evidence-for
A.5Assessing impacts of AI systems.Underwriting tier classification; receipt-based impact aggregation.evidence-for
A.6AI system life cycle.Capability lifecycle runtime; revocation; grant expiry.evidence-for
A.6.2Requirements and specification.Tool definition parameter schemas inside signed manifests.supports
A.6.3Design and development.Workspace-wide clippy -D warnings, denylist on unwrap_used, canonical JSON for every signed payload, Lean 4 proofs for protocol invariants.supports
A.6.4Verification and validation.Workspace-level cargo tests; guard integration tests; conformance harness for external implementations.evidence-for
A.6.5Deployment.Capability issuance, grant constraints, velocity guards.supports
A.6.6Operation and monitoring.Receipt store, dashboard, SIEM export, checkpoint monitor.supports
A.6.7Technical documentation.Signed tool manifests, compliance certificates.supports
A.6.8Recording of event logs.Signed receipt per invocation (allow or deny), Merkle checkpoints over batches.supports
A.7Data for AI systems.Data-layer guards: SQL, warehouse, vector store; response sanitization; column-level constraints.supports
A.7.2Data acquisition.Egress allowlist and internal-network guards limit the acquisition surface.evidence-for
A.7.4Data provenance.Workload identity, DPoP attribution, receipt capability chain.evidence-for
A.8Information for interested parties.Compliance certificates, evidence export bundles.evidence-for
A.9Use of AI systems.Capability scoping, delegation attenuation, approval tokens.supports
A.9.2Intended use.Tool manifest descriptions and GovernedAutonomyTier declarations.evidence-for
A.9.3Objectives for responsible use.Budget and velocity caps, guard suite.evidence-for
A.10Third-party and customer relationships.Capability-scoped access to third-party tools, manifest verification signatures.supports

Statement of Applicability Patterns

A typical ISO 42001 Statement of Applicability names each Annex A control, states whether the control is applicable, justifies the decision, and identifies implementation evidence. Chio makes the evidence column concrete.

soa-excerpt.yaml
A.6.8:
  title: Recording of event logs
  applicability: applicable
  implementation_evidence:
    - chio signed receipt per invocation (allow or deny)
    - merkle checkpoint batches signed by kernel keypair
    - chain integrity verified via chio receipt checkpoint verify
    - per-receipt inclusion proofs emitted to inclusion-proofs.ndjson on export

A.7:
  title: Data for AI systems
  applicability: applicable
  implementation_evidence:
    - SqlQueryGuard column constraints
    - vector store guard for retrieval governance
    - response_sanitization for output redaction
    - data_flow guard for cross-tool transfer limits

Audit Evidence Patterns

Internal audit under Clause 9.2 needs repeatable evidence assembly. Chio's evidence export plus the receipt query API are the mechanical answer.

bash
# Export the evidence package for the audit window.
# --since / --until are Unix seconds (Q1 2026 here).
$ chio evidence export \
    --policy-file ./policy.yaml \
    --receipt-db ./receipts.sqlite3 \
    --since 1767225600 \
    --until 1775001599 \
    --require-proofs \
    --output ./q1-iso-42001-evidence

# Sample the Annex A.6.8 event-log surface (100 allow + 100 deny).
# chio receipt list emits one JSON receipt per line (JSON Lines).
$ chio receipt list --receipt-db ./receipts.sqlite3 \
    --outcome allow --limit 100 > samples/allow.jsonl
$ chio receipt list --receipt-db ./receipts.sqlite3 \
    --outcome deny --limit 100 > samples/deny.jsonl

# Produce a compliance certificate for one governed session,
# an anchor for the management review.
$ chio cert generate \
    --session-id acp-session-01HQ0A... \
    --receipt-db ./receipts.sqlite3 \
    --output ./session-certificate.json

Training is not in scope

Clause 8.3 expects controlled treatment of the AI system life cycle. Chio implements the deployment and operational portions:ToolManifest at authoring, capability issuance at deployment, signed receipts during runtime, revocation plus evidence archival at retirement. The design, training, and evaluation phases of a model live outside chio's governance boundary and require training-pipeline tooling of their own.

Known Gaps

  • A.7.3 Data quality: out of scope at the tool-governance layer. Data-quality programs run in the data platform.
  • A.7.5 Data preparation: out of scope. Preparation pipelines are operator-owned.
  • A.6.3 SDLC evidence: Chio's own build pipeline enforces clippy, tests, and canonical JSON, but the customer's SDLC for their agent and its tool servers is operator-owned.
  • Management review outputs: Chio's compliance certificates summarize operational posture, but the written management-review minutes remain an organizational artifact.

For the regulation-driven mapping against Article 19, Article 14, and Annex IV, see EU AI Act. For SOC 2, HIPAA, and PCI DSS control mappings, see Compliance Frameworks.