A year ago, most AI governance questions in vendor questionnaires could be answered with a policy document and a few paragraphs about responsible AI principles. That is no longer enough. Procurement teams, regulators and boards increasingly want evidence that AI risk is managed systematically, and the instrument they reach for is ISO/IEC 42001, the international standard for an AI management system (AIMS).
Adoption is accelerating. BSI announced in November 2026 that it had issued its 50th ISO/IEC 42001 certificate, less than two years after its first, with certified organizations in more than 16 countries including Vietnam, Japan, India, the UK, the Netherlands and the United States. Its 2026 certificate count was more than triple the 11 it issued in 2025. This article is a practical playbook for organizations considering certification: what the standard requires, how to scope it, how to build the evidence, and how to avoid the mistakes that stretch a nine-month project into eighteen.
What ISO/IEC 42001 Is — and What It Is Not
ISO/IEC 42001:2023 is a management system standard. Like ISO/IEC 27001 for information security or ISO 9001 for quality, it does not prescribe a particular technology or model architecture. It requires an organization to establish, implement, maintain and continually improve a system of policies, roles, processes and controls for the responsible development, provision or use of AI.
It applies to organizations of any size that develop AI, provide AI-based products, or simply use AI in their operations. That breadth is important: a company that only consumes third-party AI services can still certify, with a scope focused on procurement, use and oversight.
What it is not: a product certification, a guarantee that a model is fair or accurate, or a legal safe harbor. Certification under ISO/IEC 42001 does not, on its own, satisfy the EU AI Act. But the risk assessments, impact assessments and lifecycle records it requires overlap heavily with what the Act expects, which is why many organizations treat it as the operating backbone for regulatory compliance.
The Structure: Clauses 4–10 and Annex A
ISO/IEC 42001 uses the harmonized structure shared by ISO management system standards. If you already run ISO/IEC 27001, the skeleton will feel familiar.
- Clause 4 – Context: internal and external issues, interested parties, your role in the AI value chain (developer, provider, user), and the scope of the AIMS.
- Clause 5 – Leadership: top-management commitment, an AI policy, and assigned roles and authorities.
- Clause 6 – Planning: AI risk assessment and treatment, AI system impact assessment, and measurable AI objectives.
- Clause 7 – Support: resources, competence, awareness, communication and documented information.
- Clause 8 – Operation: executing the planned risk treatments and impact assessments across the AI lifecycle.
- Clause 9 – Performance evaluation: monitoring and measurement, internal audit and management review.
- Clause 10 – Improvement: nonconformity, corrective action and continual improvement.
Annex A lists reference controls — commonly counted as 38 — grouped into areas such as AI policies, internal organization, resources for AI systems, assessing impacts of AI systems, the AI system life cycle, data for AI systems, information for interested parties, use of AI systems, and third-party and customer relationships. Annex B gives implementation guidance for each control. As with ISO/IEC 27001, you select applicable controls through a Statement of Applicability (SoA) that records which controls you implement, which you exclude, and why.
Companion standards worth knowing
- ISO/IEC 23894 gives guidance on AI risk management and pairs naturally with Clause 6.
- ISO/IEC 42005:2025, published in May 2025, provides guidance on AI system impact assessments: when to perform them, at which lifecycle stages, and how to document them.
- ISO/IEC 42006 sets requirements for the bodies that audit and certify AI management systems, which matters when you choose a certification body.
Scoping: The Decision That Determines Everything
The single biggest driver of cost and timeline is scope. A scope that covers “all AI across the group” sounds ambitious and usually stalls. A scope that is too narrow may not satisfy the customers who asked for certification in the first place.
Good scoping answers four questions:
- Which role are you certifying? Developer of AI systems, provider of AI products, user of AI, or a combination. Your role drives which Annex A controls apply.
- Which systems and services? Name the products or internal systems. Customers buying a specific AI product will check that it is inside the boundary.
- Which locations and teams? Include the engineering, data, product, legal and support teams that touch in-scope systems, including offshore delivery centers.
- Which interfaces are outside? Third-party foundation models, cloud platforms and data suppliers are usually outside the boundary but must be controlled through supplier management.
A useful rule of thumb: certify the AI that customers and regulators care about first, design the AIMS so it can extend to additional products, and expand scope at a surveillance or recertification audit.
The Core Artifacts Auditors Will Ask For
Auditors do not certify intentions; they test evidence. Plan to produce and maintain the following, each with a clear owner:
- AI policy approved by top management, with commitments that are specific enough to be audited.
- AI system inventory listing each in-scope system, its purpose, owner, data sources, models used, users, and risk rating.
- AI risk methodology and risk register covering technical risks (performance degradation, robustness, security), ethical and societal risks (bias, transparency, misuse), and legal risks.
- AI system impact assessments that consider effects on individuals, groups and society, performed at defined lifecycle points and refreshed on material change.
- Statement of Applicability with justified inclusions and exclusions.
- Lifecycle procedures for requirements, data acquisition and quality, model development, verification and validation, deployment, monitoring and retirement.
- Supplier and third-party controls, including due diligence on foundation-model providers and contract clauses on documentation, incident notification and data use.
- Competence and awareness records showing that people in key roles are trained.
- Monitoring evidence: dashboards, evaluation reports, incident logs, and corrective actions.
- Internal audit report and management review minutes from at least one full cycle before the certification audit.
ISO/IEC 42001 is won or lost on evidence, not documents. Every control in your Statement of Applicability should point to a living record — a log, a ticket, a test report — that an auditor can sample.
A Realistic Nine-Month Roadmap
Industry guidance typically puts first-time certification at six to twelve months, depending on maturity. The roadmap below assumes a mid-sized organization with some existing ISO/IEC 27001 or SOC 2 discipline.
Months 1–2: Gap assessment and design
Confirm scope, map current practices to clauses 4–10 and Annex A, and draft the SoA. Identify what can be reused from existing management systems: document control, internal audit, corrective action, supplier management, and incident response. Appoint an AIMS owner and a cross-functional steering group with engineering, data science, product, legal, security and HR represented.
Months 3–5: Build and embed
Write the AI policy, risk and impact methodologies and lifecycle procedures. Then — crucially — run them on real systems. Perform risk and impact assessments for every in-scope system, update the backlog with treatments, add evaluation and monitoring to release pipelines, and run supplier due diligence on model providers. Train staff in key roles.
Months 6–7: Operate and measure
Let the system run long enough to generate evidence: monitoring reports, at least one change that triggers a reassessment, incidents or near misses handled through the process, and objective tracking. Conduct a full internal audit by someone independent of the work being audited, followed by a management review.
Months 8–9: Certification audit
The external audit has two stages. Stage 1 reviews documentation and readiness, confirming the system design is complete. Stage 2 tests whether controls operate effectively, through interviews, sampling and walk-throughs. Nonconformities must be closed before the certificate is issued. After certification, expect annual surveillance audits and recertification every three years.
Integrating With ISO/IEC 27001, Privacy and the EU AI Act
Running AI governance as a separate island is the fastest way to create audit fatigue. Integration points that pay off quickly include:
- Shared management system core: one document control process, one internal audit program, one management review, one corrective-action register.
- Security controls: ISO/IEC 27001 controls for access, logging, supplier security and secure development cover much of the security dimension of AI risk; extend them with AI-specific threats such as prompt injection, data poisoning and model extraction.
- Privacy: data protection impact assessments and AI system impact assessments should cross-reference each other rather than duplicate analysis.
- EU AI Act: maintain a mapping from your controls to Act obligations — risk management, data governance, technical documentation, record-keeping, transparency, human oversight, accuracy and robustness — so the same evidence supports both certification and regulatory conformity work. With high-risk Annex III obligations now applying from December 2027, the AIMS is a natural vehicle for that preparation.
Worked Example: A SaaS Provider With Three AI Features
Imagine a document-management SaaS company with an AI search assistant built on a commercial foundation model, an in-house classification model that tags sensitive documents, and a summarization feature. Large enterprise customers in Europe and Asia-Pacific are asking for ISO/IEC 42001.
The company scopes the AIMS to its role as an AI developer and provider for those three features, including the product, data science and platform teams and its offshore engineering center, with the foundation-model vendor outside scope but under supplier control. Its gap assessment finds that ISO/IEC 27001 processes cover document control and supplier security, but there is no AI risk methodology, no impact assessment, and evaluation happens ad hoc before releases.
Over the next five months it adopts a risk methodology aligned with ISO/IEC 23894, runs impact assessments informed by ISO/IEC 42005 for each feature, and adds automated evaluation suites for retrieval quality, summarization faithfulness and classification precision to its CI pipeline. Model-provider changes now trigger a reassessment. When a vendor model update degrades summarization faithfulness in testing, the change is caught, logged as a nonconformity, and rolled back — exactly the kind of evidence a Stage 2 auditor wants to sample. The company passes certification with two minor nonconformities related to training records.
Pitfalls That Delay Certification
- Policy-only programs. Beautifully written procedures that nobody follows produce major nonconformities at Stage 2.
- No operating history. Booking the audit before the system has run a full internal audit and management review cycle.
- Ignoring third-party AI. Foundation models, embedded vendor AI and shadow AI tools used by staff are all in the risk picture, even when outside the certification boundary.
- Generic risk registers. Risks like “model may be biased” without a specific system, metric, owner or treatment do not survive sampling.
- Choosing an auditor on price alone. Check that the certification body is accredited for ISO/IEC 42001 by a recognized accreditation body and has auditors with genuine AI experience.
- Treating certification as the finish line. Surveillance audits will test whether the system keeps running as models, vendors and use cases change.
Is Certification Worth It?
For organizations that sell AI-enabled products to enterprises or regulated sectors, the commercial case is usually clear: certification shortens security and procurement reviews, differentiates in tenders, and gives the board a structured view of AI risk. For organizations that only use AI internally, an AIMS aligned with the standard but not externally certified can deliver most of the governance benefit at lower cost, with certification kept as an option for later.
Either way, the work is not wasted. The inventory, risk and impact assessments, lifecycle controls and monitoring that the standard requires are the same capabilities you need to ship AI responsibly and to meet the AI Act’s obligations as they phase in.
Conclusion
ISO/IEC 42001 has moved from a niche standard to an expected credential for serious AI providers in under three years. The organizations that certify smoothly share three habits: they scope deliberately, they build controls into the engineering lifecycle rather than around it, and they let the system operate long enough to generate real evidence before inviting an auditor in. Start with a gap assessment, reuse what your existing management systems already do well, and treat every control as something your teams will run every week — because they will.
How Hanoi Technology Can Help
Hanoi Technology helps organizations run ISO/IEC 42001 gap assessments and engineer the inventories, evaluation pipelines and monitoring evidence an AI management system audit requires.