ISO 42001
Definition
ISO 42001
ISO 42001 is the first international standard for an AI management system (AIMS). Published by ISO/IEC in December 2023, it gives organizations a certifiable framework to build AI risk, transparency, and accountability controls into every stage of the AI lifecycle.
ISO 42001 borrows the Annex SL structure of ISO 27001, so teams running an existing management system can bolt AI governance onto familiar rails. It covers any organization that develops, provides, or uses AI — vendors, integrators, and internal buyers alike.
Certification is issued by accredited third parties after a two-stage audit: documentation review, then an on-site evidence check. It signals to boards and buyers that AI risk isn’t left to project owners; it lives inside a documented, auditable system.
Adoption is accelerating. In the year after publication, big cloud providers, frontier model labs, and Big Four audit firms built ISO 42001 programs, complementing frameworks like the NIST AI Risk Management Framework.
Key takeaways
- ISO 42001 is the first certifiable AI management system standard, published by ISO/IEC in December 2023.
- It uses the same Annex SL structure as ISO 27001 and ISO 9001, so it maps cleanly onto existing governance programs.
- Certification is voluntary today, but the EU AI Act, buyer procurement clauses, and board-level risk management pressure are pulling it toward market default.
- It covers the full AI lifecycle: design, data, deployment, monitoring, and retirement — not just model output.
- Any organization that develops, provides, or uses AI can be certified, regardless of sector.
How it works
ISO 42001 works like ISO 27001 for AI. Organizations set scope, assess AI-specific risks, implement controls from Annex A, then run the Plan-Do-Check-Act cycle. An accredited body audits the system every three years, with annual surveillance visits.
| Clause / Annex | What it covers | Typical evidence |
|---|---|---|
| Context & leadership | Purpose, stakeholders, top-management commitment | Board sign-off, AI policy |
| AI risk & impact assessment | Threats, harms, affected groups | Risk register, impact records |
| Annex A controls | 38 controls across data, lifecycle, third parties | Model cards, dataset lineage |
| Performance & improvement | KPIs, incidents, corrective action | Audit logs, retraining tickets |
Annex A is the working part. Its 38 controls cover data science practices, model lifecycle management, third-party AI, and human oversight. Auditors sample evidence against each in-scope control — a claim without a paper trail fails.
Certification is not a paperwork stamp. The auditor spends time with actual engineering teams, checks the risk register lives, and confirms incidents flow into corrective action, not archived slide decks nobody opens post-launch.
Examples
Early ISO 42001 certifications went to firms with something to prove: cloud providers, model builders, and enterprise AI vendors. Certification tells regulated buyers the AI they procure sits inside an audited governance system, not a founder’s laptop.
Amazon Web Services (AWS) achieved ISO 42001 certification in November 2024, covering the Amazon Bedrock generative AI platform and Amazon Q assistants. AWS published the scope in its compliance portal for buyers running procurement checks against the EU AI Act.
Anthropic announced its ISO 42001 certification in December 2024, becoming one of the first frontier model labs to publish a certified AIMS. The certificate covers the Claude family of models and the company’s underlying safety and evaluation processes.
Microsoft reported ISO 42001 certification across Azure AI services in early 2025, tying it explicitly to EU AI Act readiness for customers building high-risk systems on Azure OpenAI and Azure Machine Learning.
KPMG and PwC both stood up ISO 42001 audit practices during 2024, offering pre-audit gap assessments to clients scrambling to align with the EU AI Act’s August 2026 general-purpose model deadlines.
IBM and Fujitsu both reported ISO 42001 alignment in 2025 as part of enterprise AI service positioning aimed at buyers in Japan, the EU, and financial services.
The certification pattern reveals what buyers are pricing. Cloud infrastructure providers certify the platform layer; model labs certify training and safety; audit firms sell the audit itself. Together they compress a year of compliance work into a purchase decision.
Related terms
ISO 42001 sits inside a wider stack of AI governance instruments. Some are voluntary standards, others are hard law like the EU AI Act, and a few are enabling technical practices. Together they define what a defensible AI program looks like in 2026.
- Artificial Intelligence: the umbrella capability the standard governs across every model, tool, and downstream product.
- Machine Learning: the statistical model family that generates most of the risk iso 42001 targets in the Annex A controls.
- Risk Management: the discipline iso 42001 imports from ISO 31000 and adapts for AI-specific harms.
- Compliance: the corporate function that typically owns iso 42001 certification evidence and audit response.
- Generative AI: the model class driving the urgent 2024–25 uptake of the standard among cloud and SaaS providers.
- Data Science: the team whose everyday model-building work Annex A controls document and constrain.
FAQ
Is ISO 42001 mandatory?
No. It’s a voluntary certification, but regulated buyers (finance, healthcare, procurement) increasingly ask for it in vendor questionnaires. Some EU AI Act obligations are met faster by an ISO 42001 certificate than starting compliance from scratch.
How long does ISO 42001 certification take?
Most organizations take six to twelve months from scoping to certificate. The path is: gap assessment, control implementation, internal audit, then a two-stage external audit by an accredited body. Existing ISO 27001 holders tend to move faster.
How does ISO 42001 relate to the EU AI Act?
The EU AI Act is law; ISO 42001 is a voluntary standard. Certification doesn’t automatically satisfy the Act, but many controls overlap — an ISO 42001 program covers most of the governance evidence the Act requires from providers of high-risk systems.
What does Annex A cover?
Annex A has 38 controls in nine themes: policies, organization, resources, impact, lifecycle, data, stakeholder info, use, and third-party relationships. Auditors sample evidence, so undocumented practices fail even when they work.
Does ISO 42001 apply to internal AI use?
Yes. The standard covers any organization that develops, provides, or uses AI, including internal chatbots, HR resume screeners, and off-the-shelf copilots. Scope is defined by the certifying organization, but “we only use vendor AI” is not an exemption.
Who publishes ISO 42001?
ISO and IEC jointly published the standard as ISO/IEC 42001:2023 in December 2023.
Find outsourcing partners who can help stand up an ISO 42001-aligned AI governance program at Outsource Accelerator.
Independent
