Mapping CB audit requirements to SWT3 witness procedures so your evidence portfolio is ready before the auditor arrives
Who this is for: Notified Bodies, C3PAOs, accredited certification bodies, ISO 42001 auditors, and organizations preparing for AIMS (AI Management System) certification. This guide assumes familiarity with ISO/IEC 17021-1 and the ISO 42001 control set.
ISO/IEC 42006:2025 was published on July 7, 2025 by ISO/IEC JTC 1/SC 42 (Artificial Intelligence). It specifies requirements for certification bodies that audit and certify AI management systems under ISO/IEC 42001:2023.
This is not a standard for organizations building AI. It is a standard for the bodies that certify those organizations. Think of it as the rulebook for auditors, not the rulebook for builders.
ISO 42006 builds on ISO/IEC 17021-1 (general requirements for bodies providing audit and certification of management systems). Where 17021-1 provides the generic framework for any management system audit, 42006 adds AI-specific requirements covering:
In November 2025, the European co-operation for Accreditation (EA) adopted a resolution making ISO 42006 mandatory for all accredited certification bodies offering AIMS certification. This means any CB operating in the European accreditation framework must comply with 42006 before issuing ISO 42001 certificates.
National standards bodies are adopting 42006 as identical national standards. Slovenia (SIST) published its adoption in 2025. The Standards Council of Canada issued an implementation bulletin. ANAB in the United States has been running CB training courses throughout 2026. As adoption spreads, the bar for what constitutes a credible ISO 42001 audit rises globally.
If your organization is pursuing ISO 42001 certification, the auditor arriving at your door is now held to a higher, AI-specific standard. They are required to bring collective competence in AI. They must spend adequate time examining your AI governance evidence. They must document their findings against AI-specific criteria.
This raises the bar for your evidence portfolio. Generic policy documents that might satisfy a general management system audit will not satisfy a 42006-compliant audit team that understands AI risk, data governance, and model lifecycle management.
ISO 42006 establishes a three-layer chain of accountability from national accreditation bodies down to the certified organization.
The CB reviews your AIMS documentation to confirm the management system is designed and documented. At this stage, the CB examines policies, risk assessments, and governance structures. SWT3 anchors for AI-GOV.1 (governance framework), AI-DATA.1 (data provenance policy), and AI-RISK.1 (risk assessment) provide cryptographic proof that these documents were reviewed and approved on specific dates.
The CB verifies your AIMS is implemented and effective. This is where operational evidence matters. The audit team examines whether your organization actually performs the activities documented in Stage 1. SWT3 anchors create a continuous, tamper-evident record of each governance activity, allowing the auditor to verify implementation frequency, consistency, and outcomes.
After certification, the CB conducts periodic surveillance audits (typically annually). The SWT3 ledger provides a complete timeline between audits, demonstrating continuous governance rather than point-in-time compliance. The auditor can examine the full chain of anchors since the last visit.
The following table maps ISO 42001 clauses to SWT3 procedures and the evidence types a 42006-compliant certification body will examine during an AIMS audit.
| ISO 42001 Clause | Requirement Area | SWT3 Procedure | Evidence Artifact |
|---|---|---|---|
| 5.1, 5.2 | Leadership, AI Policy | AI-GOV.1 | Governance framework approval records, policy review timestamps |
| 7.2, 7.3 | Competence, Awareness | AI-GOV.2 | AI literacy program completion, workforce training records |
| 6.1 | Risk Assessment | AI-RISK.1 | Risk assessment execution records, risk register updates |
| A.6.2.4 | AI System Inventory | AI-MDL.1 | Model card registration, version lineage, architecture documentation |
| A.6.2.6 | Data Quality | AI-DATA.1 | Data provenance chain, quality assessment records |
| A.6.2.5 | Bias and Fairness | AI-FAIR.1 | Bias measurement results, threshold compliance records |
| 9.1 | Monitoring, Measurement | AI-DRIFT.1 | Output drift monitoring records, threshold breach alerts |
| A.8.3 | Human Oversight | AI-HITL.1 | Human review execution records, escalation logs |
| A.8.4 | Transparency | AI-EXPL.1 | Explainability method application records, interpretability scores |
| A.8.5 | Transparency (External) | AI-TRANS.1 | Stakeholder communication records, disclosure documentation |
| 9.2 | Internal Audit | AI-AUDIT.1 | Audit trail integrity records, log completeness verification |
| A.6.2.8 | Safety | AI-SAFE.1 | Safety evaluation records, test result documentation |
| 10.2 | Nonconformity, Corrective Action | AI-INCIDENT.1 | Incident response records, corrective action tracking |
Each card below describes what a 42006-compliant CB audit team examines, the SWT3 procedure that provides evidence, and what the CB auditor should look for in the evidence chain.
What the CB examines: Evidence that top management has established, maintained, and reviewed an AI policy. The policy must include commitments to responsible AI, compliance with applicable requirements, and continual improvement. The CB verifies the policy is communicated, understood, and applied.
How SWT3 provides evidence: Each governance framework review or approval event generates an AI-GOV.1 anchor. The anchor records the review date, the reviewing authority, and the outcome. A continuous chain of AI-GOV.1 anchors demonstrates periodic policy review, not just a one-time document creation.
Verify the AI-GOV.1 anchor chain covers the full certification period. Confirm review frequency aligns with the organization's stated policy review cycle. The cryptographic fingerprint ensures the record has not been retroactively created or modified. Cross-reference with meeting minutes or board resolutions for substance verification.
What the CB examines: Evidence that persons performing work under the organization's control have the necessary AI-related competence. ISO 42006 specifically requires the CB to verify that the organization's workforce understands AI risks relevant to their roles.
How SWT3 provides evidence: AI-GOV.2 anchors record training program completions, competence assessments, and literacy milestones. Each anchor proves a specific workforce development activity occurred on a specific date with a documented outcome.
Examine AI-GOV.2 anchors for coverage across roles. The audit team should confirm that training is role-appropriate (developers receive different training than executives). ISO 42006 requires the CB audit team itself to have collective AI competence, so auditors should be prepared to evaluate training content quality, not just completion records.
What the CB examines: Evidence that the organization monitors AI system performance over time and detects degradation. The CB verifies that monitoring is systematic, that thresholds are defined, and that threshold breaches trigger corrective action.
How SWT3 provides evidence: AI-DRIFT.1 anchors record each monitoring interval. PASS indicates performance within defined thresholds. FAIL indicates a threshold breach was detected. The anchor chain provides a complete drift timeline the CB can examine for monitoring consistency and incident response patterns.
Look for gaps in the AI-DRIFT.1 anchor timeline, which may indicate monitoring interruptions. After any FAIL anchor, verify a corresponding AI-INCIDENT.1 anchor exists, confirming the organization's corrective action process was triggered. The monitoring frequency visible in the anchor chain should match the organization's documented monitoring policy.
What the CB examines: Evidence that human oversight mechanisms are implemented, that humans can intervene in AI system decisions, and that oversight is proportionate to the risk level of the AI application.
How SWT3 provides evidence: AI-HITL.1 anchors record each human review event, capturing the review outcome and the decision made. The anchor proves a human was in the loop at a specific point in the AI system's operation, with a tamper-evident record of the review.
Verify AI-HITL.1 anchor frequency is proportionate to the AI system's risk classification. High-risk applications should show more frequent human review events. Cross-reference with the organization's AI system inventory (AI-MDL.1) to confirm that all high-risk systems have corresponding HITL evidence.
What the CB examines: Evidence that the organization maintains comprehensive audit trails for AI system activities and that those trails have integrity (they have not been altered or selectively deleted).
How SWT3 provides evidence: AI-AUDIT.1 anchors verify audit log completeness and integrity at each check interval. The SHA-256 fingerprint in each anchor creates a cryptographic binding between the audit trail state at a given moment and the witness record. Any subsequent modification to audit logs would break the fingerprint chain.
AI-AUDIT.1 is particularly important for 42006-compliant audits because it addresses the integrity of all other evidence. Verify continuous coverage. The CB can independently recompute the SHA-256 fingerprint to confirm the anchor has not been tampered with, using the public verification formula with no dependency on TeNova infrastructure.
What the CB examines: Evidence that the organization assesses AI systems for bias and discrimination, that measurement methods are appropriate for the application context, and that results are within acceptable thresholds.
How SWT3 provides evidence: AI-FAIR.1 anchors record each bias evaluation event with the measurement method, the result, and whether the result fell within the defined threshold. PASS indicates acceptable bias levels. FAIL indicates a threshold breach requiring remediation.
Examine the measurement methodology recorded in the anchor factors. ISO 42006 requires the CB audit team to have collective competence in AI, including understanding of fairness metrics. Verify that the chosen metrics are appropriate for the application domain (e.g., demographic parity for hiring, equalized odds for lending). Confirm remediation actions following any FAIL verdict.
What the CB examines: Evidence that the organization has an incident management process for AI system failures, that incidents are documented, root causes analyzed, and corrective actions implemented and verified.
How SWT3 provides evidence: AI-INCIDENT.1 anchors record incident detection, response initiation, and resolution. The anchor chain for a single incident provides a tamper-evident timeline from detection through corrective action, demonstrating that the organization's nonconformity process is active and effective.
Trace AI-INCIDENT.1 anchor chains end-to-end. Verify that incidents detected via AI-DRIFT.1 or AI-SAFE.1 have corresponding incident anchors. The time between detection and response anchors reveals the organization's actual incident response capability. Compare against the organization's documented response time commitments.
What the CB examines: Evidence that the organization manages AI training and operational data with documented provenance, quality controls, and retention policies.
How SWT3 provides evidence: AI-DATA.1 anchors record data governance activities including provenance verification, quality assessment completion, and data lineage documentation updates. Each anchor proves that data governance is an active, ongoing process rather than a one-time documentation exercise.
Data governance is a frequent area of nonconformity in AIMS audits. Verify AI-DATA.1 anchor coverage across all AI systems in the organization's inventory. Look for regular cadence indicating systematic data quality reviews, not just initial dataset documentation. Cross-reference with AI-MDL.1 to confirm every registered model has associated data governance evidence.
Organizations preparing for an AIMS certification audit under a 42006-compliant CB should build their evidence portfolio well before the Stage 1 audit. The following steps ensure your evidence is ready for examination.
Begin by ensuring every AI system is registered with an AI-MDL.1 anchor. The CB will request a complete inventory. Each model card should document the system's purpose, architecture, intended use, and risk classification. The anchor proves when each system was registered and when its documentation was last updated.
Configure regular governance activities that generate AI-GOV.1 anchors. The CB will look for evidence of periodic policy review, not just initial policy creation. Quarterly reviews with documented outcomes create a strong evidence chain.
Enable AI-DRIFT.1 monitoring for all production AI systems. The CB will examine monitoring frequency and consistency. Gaps in the monitoring timeline raise questions about the organization's commitment to ongoing measurement. Automated monitoring with SWT3 anchoring eliminates manual evidence collection gaps.
For high-risk AI applications, ensure AI-HITL.1 anchors are generated at each human review point. The CB must verify that human oversight is proportionate to risk. Having anchored evidence of each review event demonstrates a functioning oversight mechanism.
Before the Stage 1 audit, assemble your evidence portfolio by procedure. The SWT3 ledger provides a single, filterable view of all governance activities. Export the ledger by procedure and date range to produce audit-ready evidence packages. Each anchor is independently verifiable by the CB without requiring access to your internal systems.
Review your anchor chain for completeness. Verify that every ISO 42001 clause in the mapping table above has corresponding SWT3 evidence. Identify gaps early. A missing procedure or an inconsistent monitoring cadence is better discovered internally than during the Stage 2 audit.
One of the strongest advantages of SWT3 witness anchors is that the same evidence artifacts satisfy multiple regulatory and certification frameworks simultaneously. Organizations pursuing ISO 42001 certification often face parallel compliance obligations.
| SWT3 Procedure | ISO 42001 | EU AI Act | NIST AI RMF |
|---|---|---|---|
| AI-GOV.1 | Clause 5.1, 5.2 | Art. 9 (Risk Management) | GOVERN 1.1 |
| AI-GOV.2 | Clause 7.2, 7.3 | Art. 4 (AI Literacy) | GOVERN 2.2 |
| AI-MDL.1 | Annex A.6.2.4 | Art. 11 (Technical Documentation) | MAP 1.1 |
| AI-FAIR.1 | Annex A.6.2.5 | Art. 10 (Data Governance) | MEASURE 2.6 |
| AI-DRIFT.1 | Clause 9.1 | Art. 9(2) (Ongoing Monitoring) | MEASURE 3.2 |
| AI-HITL.1 | Annex A.8.3 | Art. 14 (Human Oversight) | GOVERN 1.4 |
| AI-EXPL.1 | Annex A.8.4 | Art. 13 (Transparency) | MEASURE 2.8 |
| AI-DATA.1 | Annex A.6.2.6 | Art. 10 (Data Quality) | MAP 2.3 |
| AI-SAFE.1 | Annex A.6.2.8 | Art. 9(7) (Testing) | MEASURE 3.3 |
| AI-INCIDENT.1 | Clause 10.2 | Art. 62 (Incident Reporting) | MANAGE 4.1 |
| AI-TRANS.1 | Annex A.8.5 | Art. 13 (Transparency) | GOVERN 4.2 |
| AI-AUDIT.1 | Clause 9.2 | Art. 12 (Record-Keeping) | GOVERN 5.2 |
| AI-RISK.1 | Clause 6.1 | Art. 9 (Risk Management) | MAP 5.1 |
ISO 42001 defines what an AI management system must include. ISO 42006 defines what a certification body must do when auditing that system. 42001 is for organizations. 42006 is for auditors.
Not directly. Your evidence requirements come from ISO 42001. However, 42006 raises the competence bar for CB audit teams, meaning auditors will examine your evidence more rigorously and with deeper AI domain knowledge. Generic, superficial evidence that might pass a general management system audit will face closer scrutiny.
For certification bodies operating under European accreditation (EA members), yes, per the November 2025 EA resolution. For organizations seeking certification, it is not directly applicable, but it determines the quality and rigor of the audit you will undergo.
Yes. Each anchor contains a SHA-256 fingerprint that can be independently recomputed using a publicly documented formula. The CB does not need access to TeNova infrastructure, accounts, or APIs. The verification is purely mathematical.
ISO 42006 addresses access to sensitive AI documentation. The SWT3 Clearing Protocol provides configurable clearing levels that control what metadata is retained alongside the cryptographic anchor. At higher clearing levels, sensitive model details remain with the organization while the cryptographic proof of governance activity is independently verifiable.
CBs outside the EA framework may not yet be required to comply with 42006. However, the standard represents the emerging global benchmark for AI certification audit quality. Organizations should prefer CBs that voluntarily adopt 42006 practices. Your SWT3 evidence portfolio works regardless of the CB's 42006 status, but a 42006-compliant CB will evaluate it more thoroughly.
The EU AI Act requires conformity assessments for high-risk AI systems. Notified Bodies performing those assessments may draw on ISO 42001 certification as supporting evidence. ISO 42006 ensures the CBs conducting those certifications meet AI-specific competence requirements. The SWT3 evidence portfolio serves both pathways.
ISO 42006 includes guidance on audit time calculation that accounts for AI system complexity, the number of AI systems in scope, and the organization's risk profile. Expect longer audits than a typical ISO 27001 or ISO 9001 certification. Having a structured, cryptographically anchored evidence portfolio reduces the time the CB spends locating and validating evidence.