ISO/IEC 42001:2023 defines an AI Management System (AIMS) — the AI equivalent of what ISO 27001 is for information security. Certification means an accredited certification body has audited your AIMS and confirmed it meets the standard: you have an AI policy, defined roles, risk and impact assessments for your AI systems, lifecycle controls, and a working improvement loop.
Two clarifications buyers ask first:
Why it's surging in 2026: enterprise buyers have started asking for it in security reviews the way they ask for SOC 2, and it maps cleanly onto EU AI Act readiness. For what the standard contains clause-by-clause, see our ISO 42001 guide; this page is about getting certified.
Step 1 — Scope and gap analysis (2–4 weeks). Define which AI systems, teams, and sites the AIMS covers, then assess current state against the standard's clauses and Annex A. The output is your remediation backlog. Scoping decisions matter: too broad and the project balloons; too narrow and the certificate doesn't cover what customers ask about.
Step 2 — Build the AIMS (1–4 months). The core artifacts: - AI policy and objectives approved by leadership, with defined roles and accountability - AI risk assessment — the risks your AI systems create (see ISO 23894 for the companion risk-management methodology) - AI system impact assessments — 42001's signature requirement: documented analysis of how each in-scope AI system affects individuals and society - Annex A controls — the standard's 38 controls, covering the AI lifecycle, data quality and governance, transparency, human oversight, and third-party AI suppliers - Data protection for AI in practice — knowing what sensitive data your AI systems touch and controlling it; this is where a data security layer for AI does the enforcing
Step 3 — Operate, internally audit, and review (1–2 months). Auditors need to see the system running: records, metrics, at least one internal audit, and a management review with findings and actions. Certification bodies routinely reject systems that were written the month before the audit.
Step 4 — Certification audit. Stage 1 reviews documentation and readiness; Stage 2 tests operation — sampling records, interviewing owners, tracing controls to evidence. Nonconformities get corrective-action plans; major ones delay the certificate.
Step 5 — Keep it (3-year cycle). Annual surveillance audits, then full recertification at year three. Evidence collection never stops — which is exactly why manual screenshot-driven programs decay.
Two honest cost notes: quotes vary widely between certification bodies, so get several; and if you already hold ISO 27001, integrated audits (27001 + 42001 together) meaningfully reduce both cost and audit fatigue — see ISO 27001 AI compliance for how the two standards divide the work.
Most of the 3–9 months goes to two things: building the AI inventory and impact-assessment inputs, and accumulating operating evidence deep enough for Stage 2 sampling. Strac attacks both directly, because the platform that enforces your AI data controls is the same one filling the binder:
Most organizations take 3–9 months from gap analysis to certificate: a few weeks of scoping, one to four months building and operating the AIMS, then the two-stage certification audit. The variables that stretch timelines are AI system count, the depth of impact assessments, and how much evidence collection is manual. Teams that automate evidence typically land at the short end.
Certification-body fees typically run $15K–$50K+ for the initial two-stage audit, plus $5K–$15K per year for surveillance audits, scaling with company size and AI scope. Preparation — impact assessments, control implementation, evidence — is usually the larger real cost. Integrated audits with an existing ISO 27001 program and compliance automation both pull the total down meaningfully.
Any organization whose customers, regulators, or risk profile demand provable AI governance — AI product companies first, but increasingly any enterprise deploying AI on customer data. You don't have to build models: using AI tools and agents on regulated data is enough to be in scope, and enterprise security reviews have started asking for 42001 the way they ask for SOC 2.
No. Only an accredited certification body can issue an ISO 42001 certificate, through a Stage 1 (documentation) and Stage 2 (operation) audit. You can self-assess against the standard as preparation — and should — but "aligned with ISO 42001" is not "certified."
If AI is material to your product or operations, increasingly yes. SOC 2 and 27001 don't cover AI-specific obligations — impact assessments, AI lifecycle controls, transparency, human oversight — which is exactly what enterprise AI questionnaires now probe. The good news: your existing control set carries much of the weight, and the data-protection evidence you already produce can be mapped across all three frameworks at once.
The AI policy and risk/impact assessments, an AI system inventory (including third-party AI), records showing Annex A controls operating — data governance logs, access reviews, monitoring of AI data flows, incident records — plus internal audit and management review outputs. The recurring pain is proving data protection inside AI systems; controls that log their own enforcement, like Strac's redaction and policy events, turn that from a screenshot hunt into an export.
ISO 42001 certification is becoming the SOC 2 of AI: a buyer-driven, auditable proof that your AI is governed. The process is well-trodden — scope, build, operate, two-stage audit, surveillance — and the certificate goes to teams whose controls run continuously, not teams with the best-written policy binder. Automate the inventory, the data-protection enforcement, and the evidence, and certification becomes a project instead of a lifestyle.
Book a demo of Strac Comply to see ISO 42001 evidence collected continuously — with the AI data-security controls doing the proving.
Related: ISO 42001 explained · ISO 23894 · ISO 27001 AI compliance · AI compliance checklist · AI governance
.avif){kind=icon}
.gif)
.webp)
