Skip to content

Latest commit

 

History

History
337 lines (231 loc) · 23.6 KB

File metadata and controls

337 lines (231 loc) · 23.6 KB

ISO/IEC 42001 Certification Engagement Package

Audience: Compliance lead + engineering exec preparing the AIMS (AI Management System) for third-party certification. Scope: Selecting an accredited certification body, navigating Stage 1 (documentation review) and Stage 2 (implementation audit), and maintaining the certificate across the three-year cycle. Prerequisite: Internal readiness pack at packages/safety/compliance/iso-42001-readiness/ is drafted and minutes from Management Review #1 exist. If that pack is not yet in place, complete Section 2 (Pre-engagement readiness) before requesting cert-body quotes — quotes anchor to scope, and scope without a Statement of Applicability is a guess. Status of figures: All pricing in this document is estimated from public references, vendor sales conversations, and 2025-era engagement post-mortems. Quotes must be obtained directly; treat the ranges as planning anchors, not commitments.


1. Certification body comparison

ISO/IEC 42001 was published 2023-12. Accreditation for cert bodies under UKAS (UK) and ANAB (US, via ANSI National Accreditation Board) ramped through 2024-2025. Most major IMS (Integrated Management System) bodies offer 42001 today, but AI-specific track record is the differentiator — a body that has issued only 5-10 42001 certs will reuse 27001 auditors with a thin AI briefing, which surfaces as weaker Annex A coverage during Stage 2.

1.1 Established IMS bodies

Body 42001 accreditation AI-specific track record Stage 1+2 (Year 1, est.) Surveillance Y2 / Y3 (est.) Geo coverage Lead time
BSI Group (https://www.bsigroup.com/en-GB/products-and-services/standards/iso-iec-42001-ai-management-system/) UKAS-accredited 42001 (first body globally, Feb 2024) Strongest — co-authored the standard via committee; 100+ 42001 certs issued globally by Q1 2026 $35-55k $12-20k / $12-20k Global (60+ countries) 14-20 weeks from kick-off
DNV (https://www.dnv.com/services/iso-iec-42001-artificial-intelligence-management-system-certification-243947) UKAS + ANAB Strong — heavy in EU regulated industries (maritime, energy, healthcare AI) $30-50k $10-18k / $10-18k Global, EU-strongest 12-18 weeks
TÜV SÜD (https://www.tuvsud.com/en/services/auditing-and-system-certification/iso-iec-42001) DAkkS (Germany) + UKAS Strong — known for industrial AI, automotive, medical devices; tight on Annex A.6 (impact assessment) $35-60k $14-22k / $14-22k Global, EU/DACH-strongest 14-22 weeks
TÜV Rheinland (https://www.tuv.com/landingpage/en/iso-iec-42001/) DAkkS + ANAB Moderate-strong — overlaps TÜV SÜD turf; differentiator is faster turnaround in US $30-50k $11-18k / $11-18k Global, EU/US balanced 12-16 weeks
SGS (https://www.sgs.com/en/services/iso-iec-42001-aims-certification) UKAS + ANAB Moderate — broad IMS portfolio, AI specialism is newer (2024-onward); pricing competitive $25-45k $9-16k / $9-16k Global, broadest geographic footprint of any body listed 10-16 weeks
Bureau Veritas (https://www.bureauveritas.com/services-plus-solutions/iso-42001-artificial-intelligence-management-system-certification) UKAS + ANAB Moderate — strong in finance/insurance AI; weaker public track record on pure-play AI SaaS $28-48k $10-17k / $10-17k Global 12-18 weeks

1.2 AI-focused / newer entrants

Body 42001 accreditation AI-specific track record Stage 1+2 (Year 1, est.) Surveillance Y2 / Y3 (est.) Geo coverage Lead time
A-LIGN (https://www.a-lign.com/services/iso-42001) ANAB Moderate — SOC 2 / ISO 27001 specialist pivoting into 42001; strong for US tech orgs already in their ecosystem $25-40k $8-14k / $8-14k US-primary, EU via partner 8-14 weeks
Schellman (https://www.schellman.com/iso-42001) ANAB Moderate-strong — early ANAB-accredited 42001 body, US tech-first; bundles well with SOC 2 + 27001 $28-45k $9-15k / $9-15k US-primary 8-12 weeks
Mastermind / Coalfire / Insight Assurance ANAB (varies) Light-moderate — emerging; verify accreditation status at quote time $20-35k $7-12k / $7-12k US-primary 8-12 weeks
HKQAA, JQA, JIPDEC (regional) Regional accreditation Light on AI; strong domestic recognition Varies, often lower Varies APAC Varies

1.3 Reading the table

  • Year-1 totals assume a single-site AIMS scope covering one to three AI products with ≤ 200 personnel in scope. Multi-site or > 500 personnel adds 25-50%.
  • Day rates drive the totals: BSI / TÜV SÜD anchor around $2,500-3,200/day; SGS / Schellman / A-LIGN around $1,800-2,500/day; emerging US-tech-focused bodies $1,500-2,200/day. Stage 1 is typically 2-4 audit-days; Stage 2 is 5-10.
  • Surveillance audits at year 2 and year 3 are roughly 40-50% of the Stage 2 effort. Recertification at year 3 is roughly 80% of the original Year-1 cost.
  • Lead time is the gap between signed engagement letter and Stage 1 onsite. UKAS-accredited bodies were back-logged through most of 2025; expect 12+ weeks for BSI/DNV/TÜV SÜD, 8-12 for Schellman/A-LIGN.

2. Pre-engagement readiness assessment

Run this checklist against the existing pack at packages/safety/compliance/iso-42001-readiness/ before requesting quotes. Every "no" is a Stage 1 or Stage 2 finding waiting to happen.

2.1 Clause 4 — Context of the organization

  • iso-42001-readiness/clause-4/context-of-organization.md documents internal/external issues, interested parties (regulators, customers, employees, society), and AIMS scope statement.
  • Scope statement names the AI systems in scope (model families, deployment surfaces) and exclusions with justification.
  • Interested-party register reviewed within last 12 months.

2.2 Clause 5 — Leadership

  • AIMS policy signed by top management within last 12 months (iso-42001-readiness/clause-5/aims-policy.md).
  • Roles, responsibilities, and authorities documented and assigned by name, not just by title (iso-42001-readiness/clause-5/raci.md).
  • Top management's accountability for AIMS effectiveness is demonstrable via Management Review minutes (see 2.6).

2.3 Clause 6 — Planning

  • AI risk register (iso-42001-readiness/clause-6/ai-risk-register.xlsx or .md) — current within 90 days, covering at minimum: bias/fairness, robustness, privacy, security, transparency, accountability, environmental impact.
  • AI impact assessments for each in-scope AI system (iso-42001-readiness/clause-6/aiia/) — per ISO/IEC 42005 guidance.
  • AIMS objectives are SMART, owner-assigned, with measurement cadence (iso-42001-readiness/clause-6/objectives.md).
  • Treatment plan exists for each unacceptable residual risk.

2.4 Clause 7 — Support

  • Resources, competence, awareness, communication, documented information clauses each have a one-pager (iso-42001-readiness/clause-7/).
  • Competence matrix evidences training of personnel on AI risks, AIMS responsibilities, and incident response.

2.5 Clause 8 — Operation

  • Operational planning and control: ML lifecycle, data management, model deployment runbooks linked (iso-42001-readiness/clause-8/operational-controls.md).
  • AI system impact assessments operationalized — each new model release triggers the assessment workflow.
  • Data quality and provenance controls (iso-42001-readiness/clause-8/data-management.md).
  • Third-party / supplier AI controls (iso-42001-readiness/clause-8/supplier-controls.md).

2.6 Clause 9 — Performance evaluation

  • Monitoring/measurement plan with KPIs for each AIMS objective (iso-42001-readiness/clause-9/monitoring.md).
  • Internal audit #1 completed with audit report, nonconformity log, and closure evidence (iso-42001-readiness/clause-9/internal-audit-1/). This is a Stage 2 gating item.
  • Management Review #1 minuted covering all required inputs (audit results, KPI trends, risk changes, opportunities for improvement) and all required outputs (decisions, actions, resources). (iso-42001-readiness/clause-9/management-review-1.md). Also a Stage 2 gating item.

2.7 Clause 10 — Improvement

  • Nonconformity and corrective action procedure with worked examples from the last cycle (iso-42001-readiness/clause-10/ncar-procedure.md).
  • Continual improvement evidence — at least 3 closed improvements traceable to MR #1 or internal audit findings.

2.8 Annex A — Statement of Applicability (SoA)

  • SoA (iso-42001-readiness/soa.xlsx or .md) covers all Annex A controls (A.2 through A.10) with: applicable Y/N, justification, implementation reference, evidence pointer.
  • No Annex A control marked "N/A" without written justification.
  • Each "applicable" control has at least one piece of implementation evidence (procedure, log, configuration, training record).

2.9 Readiness gate

If any of the bolded items above are "no" — internal audit #1, management review #1, SoA — do not engage a cert body yet. Stage 1 will fail and you'll burn the engagement fee on a dry run.


3. Stage 1 preparation timeline — 8 weeks

Stage 1 is a documentation and readiness review. The cert body confirms the AIMS exists on paper and you're ready for the implementation audit. Findings are usually "areas of concern" rather than nonconformities — but unresolved areas of concern become Stage 2 NCs.

Weeks 1-2: Internal audit dry-run

  • Re-run the internal audit (clause 9.2) against the full clause-4-to-10 + Annex A checklist as if you were an external auditor.
  • Audit team must be independent of the area audited (i.e., the ML team doesn't audit ML processes).
  • Output: refreshed nonconformity log with severities (minor / major / observation).

Weeks 3-4: Close minor nonconformities

  • Close every minor NC from the dry run. Document root cause, correction, corrective action.
  • Defer major NCs only if they truly cannot be closed in 8 weeks — but flag them in your cert-body application so the auditor isn't surprised.
  • Refresh Management Review minutes if any closure changes the risk profile.

Week 5: Cert body application + contract

  • Submit application to selected cert body (see Section 6 for decision matrix). Application requires: AIMS scope statement, number of personnel, sites, AI systems in scope, prior certifications (ISO 27001, SOC 2).
  • Cert body returns a proposal with day count and price. Sign within the week; lead time runs from signature.

Weeks 6-7: Stage 1 documentation package

  • Compile the Stage 1 evidence package — auditors request this 2-3 weeks pre-onsite:
    • AIMS policy, scope, objectives
    • SoA + Annex A evidence
    • Internal audit #1 report + NC closure evidence
    • Management Review #1 minutes
    • Risk register + treatment plan
    • AI impact assessments for in-scope systems
    • Org chart + RACI
  • Upload to the cert body's audit portal (most use proprietary portals; some accept Drive/SharePoint shares).

Week 8: Stage 1 onsite (or remote)

  • 2-4 audit-days. Auditor reviews the documentation package, interviews top management + AIMS owner, walks through 2-3 sample controls.
  • Output: Stage 1 report listing readiness for Stage 2 + any areas of concern.
  • Decision point: Stage 2 can proceed only after Stage 1 closure. If major findings surface, expect a 4-8 week gap before Stage 2.

4. Stage 2 preparation timeline — 8-12 weeks post-Stage 1

Stage 2 is the implementation audit. The auditor traces controls from policy through implementation to evidence, interviews personnel at all levels, and tests sampled processes end-to-end.

Weeks 1-3 post-Stage 1: Gap closure

  • Address every Stage 1 area of concern with documented action.
  • For each finding, produce: root cause, correction, corrective action, evidence of effectiveness, closure date.
  • If a finding touches a recurring process, run the process at least once post-fix to demonstrate the corrective action works.

Weeks 4-6: Evidence dossier compilation

  • Build a per-control evidence dossier indexed against the SoA. Each Annex A control should have ≥ 2 pieces of evidence from the last 90 days (logs, screenshots, signed documents, training records).
  • Specifically over-prepare on the controls in Section 5 (common findings) — those are the auditor's hot list.
  • Evidence freshness matters: a 2024 training record won't carry a 2026 audit.

Weeks 7-9: Personnel interview prep

  • Auditors sample 5-15 personnel across roles: top management, AIMS owner, ML engineers, data engineers, security, legal, customer-facing.
  • Each interviewee should be able to:
    • State their role in the AIMS in one sentence
    • Name the AIMS policy and where to find it
    • Describe one AI risk relevant to their work and how it's controlled
    • Describe the incident response path for an AI failure
  • Run mock interviews in week 7-8. The most common Stage 2 finding from personnel interviews is inconsistency between policy and practice — a documented control that nobody on the team can describe.

Weeks 10-11: Walkthrough rehearsal

  • Pick 3-5 sample AI systems and rehearse end-to-end walkthroughs: design → risk assessment → data sourcing → training → evaluation → deployment → monitoring → incident handling.
  • Time-box each walkthrough to 60-90 minutes. Stage 2 walkthroughs run that long; rehearse to that bound.

Week 12: Stage 2 audit window

  • 5-10 audit-days depending on scope. Onsite or remote (most bodies accept remote post-2024).
  • Closing meeting on the final day: auditor presents draft findings.
  • 2-4 weeks post-audit: final report + cert decision. Major NCs require 90-day closure plan; minor NCs require 90-day evidence of corrective action.

5. Common Stage 1 + Stage 2 findings

Surveyed from publicly available audit summaries (BSI, DNV, Schellman case studies 2024-2025), ISO/IEC JTC 1/SC 42 working group post-mortems, and certified-org disclosures.

5.1 Top 10 findings + preemptive mitigations

# Finding Root cause Preemptive mitigation
1 SoA control marked applicable but no evidence Control adopted from template, not operationalized Audit each "applicable" row in iso-42001-readiness/soa.md for ≥ 2 evidence artifacts dated within 90 days
2 AI impact assessments stale or missing for recent model releases Process exists but doesn't fire on every release Wire the AIIA into the release checklist (iso-42001-readiness/clause-6/aiia/release-trigger.md); evidence the last 3 releases each have one
3 Risk register not reviewed within 90 days Quarterly cadence slipped Calendar invite + MR agenda item; refresh iso-42001-readiness/clause-6/ai-risk-register.md no less than 90 days pre-Stage 1
4 Top management cannot articulate AIMS objectives Policy signed but not internalized One-page exec brief; rehearse the interview in week 7-8 of Section 4
5 Personnel inconsistent on incident response path Procedure exists but training is patchy Tabletop exercise documented in iso-42001-readiness/clause-8/incident-response-tabletop.md; refresh competence matrix
6 No evidence of supplier AI controls Vendor onboarding doesn't include AI-specific clauses Updated DPA / supplier addendum at iso-42001-readiness/clause-8/supplier-controls.md; evidence: 2-3 executed addenda
7 Data provenance gaps on training datasets Older datasets ingested without lineage metadata Lineage register at iso-42001-readiness/clause-8/data-management.md; remediate top 5 datasets minimum
8 Internal audit didn't cover all clauses + Annex A Audit scope drift Audit plan template explicitly enumerates clauses 4-10 + every applicable Annex A control (iso-42001-readiness/clause-9/internal-audit-plan.md)
9 Management Review missed required inputs/outputs MR minutes informal Template at iso-42001-readiness/clause-9/management-review-template.md with the 9.3.2 input list and 9.3.3 output list as headers
10 Continual improvement not traceable to inputs Improvements happen but aren't logged against findings Improvement log links each closed item to its source (audit / MR / incident / complaint) at iso-42001-readiness/clause-10/improvement-log.md

5.2 AI-specific findings (rarer in 27001, common in 42001)

  • Bias / fairness controls described abstractly without metrics. Auditor wants to see the metric (demographic parity, equalized odds, etc.), the threshold, the monitoring cadence, the action on breach.
  • No transparency artifact for in-scope systems. Model cards, system cards, or equivalent — referenced in iso-42001-readiness/clause-8/transparency-artifacts/.
  • Human oversight described, not designed. Auditor wants to see the human-in-the-loop control point in the actual product flow, with logs of human decisions.
  • Environmental impact omitted. Annex A.6.2.2 is explicit; have a one-page energy / compute footprint estimate per in-scope system.

6. Decision matrix

Weighted criteria for selecting among the bodies in Section 1.

Criterion Weight Score 1-5 per body Weighted
AI-specific expertise (track record, AI-trained auditors, published guidance) 35% per body × 0.35
Pricing (Year-1 total + 3-year TCO) 25% per body × 0.25
Geographic + lead time fit (audit-team location, time-zone alignment, weeks-to-Stage-1) 20% per body × 0.20
Surveillance terms (Y2/Y3 pricing, scope flexibility, re-audit triggers) 10% per body × 0.10
Reputation / customer recognition (does the cert carry weight with our customers?) 10% per body × 0.10

6.1 Default ranking for a US-primary, EU-secondary AI SaaS

Based on the weights above and the public information in Section 1:

  1. BSI — top on AI expertise + reputation; mid on pricing; long lead time is the main drag.
  2. Schellman — top on US lead time + bundle with SOC 2 / 27001; mid on AI expertise.
  3. DNV — top in regulated-industry recognition; competitive pricing; strong EU coverage.
  4. TÜV SÜD — top for DACH / industrial customer base; otherwise comparable to DNV.
  5. A-LIGN — strong if already in their SOC 2 / 27001 program; lower reputation weight than BSI.
  6. Bureau Veritas / SGS / TÜV Rheinland — viable mid-tier choices; pick on geographic fit.

6.2 When to deviate

  • Customer mandate. If a flagship customer requires a specific body, that body wins regardless of matrix score.
  • Regulated sector. Healthcare → DNV/TÜV SÜD. Automotive → TÜV SÜD/TÜV Rheinland. Financial services → BSI/Bureau Veritas.
  • Bundle economics. If SOC 2 and 27001 are also in play, bundled-audit savings with Schellman / A-LIGN can exceed BSI's reputation premium.

7. Surveillance cycle

ISO 42001 certificates are valid 3 years, with annual surveillance audits.

7.1 Year 2 surveillance

  • Scope: ~40-50% of Stage 2 effort. 2-5 audit-days.
  • Auditor priorities: evidence that the AIMS is operating (not just documented), closure of any Year-1 NCs, Management Review and internal audit conducted since cert, risk register refreshed, sampled controls.
  • Common Year-2 trip wires: the MR / internal audit cadence slips because the urgency drained after cert. Calendar these immediately post-cert.

7.2 Year 3 surveillance + recertification

  • Surveillance: same as Year 2.
  • Recertification: runs alongside Year-3 surveillance or in the following months. ~80% of Year-1 cost. Full clause + Annex A coverage, but with the auditor's accumulated familiarity reducing onboarding overhead.
  • Output: new 3-year certificate.

7.3 Year-on-year cost envelope (estimated)

For a single-site, three-AI-product, ~200-person scope:

Year Activity Estimated cost
Y1 Stage 1 + Stage 2 $25-60k
Y2 Surveillance $8-22k
Y3 Surveillance + recert $25-50k
3-year TCO $58-132k

8. Certificate maintenance — what triggers a re-audit

Outside the scheduled cycle, the following changes require notifying the cert body and may trigger a special audit (typically 1-3 days, charged at day rate):

8.1 Mandatory notification

  • Major AIMS scope change — new AI system added to scope, new site, scope expansion to a new business unit, or scope reduction (the body confirms the reduction doesn't invalidate the cert).
  • Top management change — new CEO / AIMS sponsor; the body confirms continuity of leadership commitment.
  • Significant regulatory shift affecting in-scope AI — e.g., EU AI Act enforcement milestones, NIST AI RMF amendments, sector-specific AI rules that materially change controls.
  • Major nonconformity not closed within 90 days — escalates to suspension review.
  • Serious AI incident — material customer / regulator / public-facing incident; the body may run an unannounced or special surveillance audit per IAF MD 4 guidance.

8.2 Optional notification / recommended

  • Material residual-risk escalation — a risk previously rated acceptable now exceeds the tolerance threshold, with no immediate treatment plan.
  • Material technology shift — moving from in-house models to a third-party foundation model (or vice versa) — changes supplier-control surface materially.
  • Acquisition / divestiture affecting the AI portfolio.

8.3 Triggers for early recertification

Rare, but worth knowing:

  • Cumulative scope changes have made the original scope statement materially inaccurate.
  • The body's accreditation lapses or transfers (very rare; you may need to migrate the cert).
  • Standard revision (a hypothetical ISO 42001:2028) — transition window is typically 2-3 years.

8.4 Maintenance discipline

The cheapest path to a clean Year-2 and Year-3 audit is operating the AIMS as if the cert body could walk in any week. Concretely:

  • Quarterly risk register refresh (calendar invite, never skipped).
  • Quarterly Management Review (lighter cadence than the pre-cert one, but still minuted with the 9.3.2/9.3.3 structure).
  • Annual internal audit + annual full Management Review.
  • Continuous improvement log updated within 5 business days of any incident, finding, or customer complaint touching AI.

Drift here is the single biggest predictor of a hostile Year-2 surveillance.


9. Engagement checklist (one-page summary)

Pre-engagement:

  • Section 2 readiness checklist all green
  • Internal audit #1 closed
  • Management Review #1 minuted
  • SoA evidenced ≥ 2 artifacts per applicable control

Body selection:

  • Decision matrix (Section 6) scored across at least 3 bodies
  • 2-3 quotes obtained and compared on like-for-like scope
  • Customer/regulatory constraints applied per Section 6.2

Stage 1 (8 weeks):

  • Internal audit dry-run → NC closure → application → docs package → onsite

Stage 2 (8-12 weeks post-Stage 1):

  • Gap closure → dossier → interview prep → walkthrough rehearsal → audit window

Post-cert:

  • Y2 + Y3 surveillance calendared on day-1 post-cert
  • Change-notification triggers (Section 8) wired into change-management policy
  • Recertification budget allocated in Y3

Document version: v1 (2026-05-05) Next review: post-Stage 1 walkthrough, or 2026-Q4, whichever first. Owner: Compliance lead. References: packages/safety/compliance/iso-42001-readiness/ (full readiness pack), roadmap-2026/MACRO_ROADMAP.md (program-level context).