ISO/IEC 42001 and other AI governance standards: what each one offers and how to combine them
Artificial intelligence is already part of customer service, analytics, software development and decision-making at many companies. The question is no longer just what AI can do, but who is accountable for it, how its risks are controlled and how that is demonstrated. That is what governance standards and frameworks are for. The best known is ISO/IEC 42001, but it isn’t the only one, and each addresses a different need.
ISO/IEC 42001: the AI management system
ISO/IEC 42001:2023, published on December 18, 2023, by the joint ISO and IEC committee on artificial intelligence (JTC 1/SC 42), sets out the requirements for establishing, implementing, maintaining and continually improving an artificial intelligence management system (IEC, 2023). It applies to organizations that develop, provide or use AI, whether that’s a predictive model, a generative assistant or an agent.
It follows the same logic as other management system standards, such as ISO 9001 or ISO/IEC 27001: it doesn’t tell you which technology to use, but how to organize yourself.
| Clauses of the standard | What they require | Typical evidence |
|---|---|---|
| 4 and 5: context and leadership | Scope of the system, AI policy and assigned roles | Scope document, approved policy, named owners |
| 6: planning | Objectives, risks and impacts | Risk register, impact assessment, treatment plan |
| 7 and 8: support and operation | Resources, competence, documentation and life cycle control | Procedures, training, controls over development and use |
| 9 and 10: evaluation and improvement | Monitoring, audit, review and correction | Metrics, internal audits, corrective actions |
The standard also includes an annex of reference controls. Each organization selects and justifies the ones it applies based on its scope and risks; the point is not to apply all of them indiscriminately.
It is the only one in this group that allows organizational certification by an independent body. Since July 2025, ISO/IEC 42006 has set the requirements those certification bodies must meet (IEC, 2025). Major AI providers have already certified several of their services: AWS in November 2024 (AWS, 2024), Anthropic in January 2025 (Anthropic, 2025) and Microsoft in July 2025 (Microsoft, 2025).
The ISO/IEC family of AI standards
Around 42001 there are standards that help with specific parts of the work. These are the most useful for a company:
| Standard | What it’s for | Published |
|---|---|---|
| ISO/IEC 23894:2023 | Guidance on managing AI risk | February 2023 |
| ISO/IEC 42005:2025 | Assessing the impact of AI systems on people and society | May 2025 |
| ISO/IEC 38507:2022 | Governance implications of AI, aimed at top management | April 2022 |
| ISO/IEC 5338:2023 | Life cycle processes: design, testing, deployment and retirement | December 2023 |
| ISO/IEC 5259 series | Data quality for analytics and machine learning | Since July 2024 |
Dates are taken from the official IEC listings, linked in the references.
The security and privacy standards many companies already use, such as ISO/IEC 27001 and ISO/IEC 27701, are still necessary, but they don’t replace AI-specific governance.
Complementary frameworks outside ISO
- NIST AI RMF 1.0 (United States, January 2023): organizes AI risk management into four functions (govern, map, measure and manage) and offers a free practical playbook. It is voluntary and under revision, so it may change soon. Its generative AI profile, NIST AI 600-1 (July 2024), details the risks specific to this technology (NIST).
- CSA AI Controls Matrix 1.1 (Cloud Security Alliance, June 2026): 247 control objectives across 18 domains, mapped to ISO/IEC 42001, ISO/IEC 27001, the NIST AI RMF and the EU AI Act. It is very useful for evaluating platforms and vendors (CSA, 2026).
- OWASP GenAI Security Project: in September 2026 it published its 2026 Top 10 risks for large language model applications. It also maintains a separate list for agentic applications. It is the go-to reference for the technical security of assistants, RAG systems and agents (OWASP, 2026).
- MITRE ATLAS: a knowledge base of the tactics and techniques used to attack AI systems. It helps design tests and defenses (MITRE).
- International principles: the UNESCO Recommendation on the Ethics of AI (November 2021), the first global standard on the subject, applicable to its 194 member states (UNESCO), and the OECD AI Principles (May 2019, updated in May 2024) (OECD). They are not control checklists, but they guide decisions that affect people.
What about the EU AI Act? It is mandatory regulation for anyone operating in the European Union, not a voluntary framework. Its first harmonized standard, EN 18286:2026, on the quality management system the Act requires of high-risk AI providers (Article 17), was cited in the Official Journal of the EU in October 2026; applying it gives a presumption of conformity with those requirements (CEN-CENELEC, 2026). ISO/IEC 42001 is not a harmonized standard: it is a solid foundation, but on its own it does not amount to compliance with the Act (European Commission).
How they complement each other
They don’t compete: each covers a different front. This is our indicative assessment of where each one stands out:
| Criterion | ISO/IEC 42001 | NIST AI RMF | CSA AICM | OWASP |
|---|---|---|---|---|
| Organizational governance | Very high | Very high | Medium | Low |
| Risk management | Very high | Very high | High | Technical |
| LLM and agent security | General | General | High | Very high |
| Impact assessment | High | High | Medium | Partial |
| Specific technical controls | Medium | Medium | Very high | Very high |
| Organizational certification | Yes | No | No equivalent | No |
To decide: ISO/IEC 42001 if you want certification; NIST AI RMF to start flexibly; CSA AICM, OWASP and MITRE ATLAS for the technical side.
Common risks and their key control
| AI use | Main risk | Key control |
|---|---|---|
| Assistants for customers or employees | Fabricated answers or wrong guidance | Validated knowledge base, evaluation and a correction channel |
| Decisions about people (hiring, credit, performance reviews) | Bias, unfairness, lack of explanation | Human oversight and a right to review |
| Search over internal documents (RAG) | Exposure of information without authorization | Per-document access control and leakage testing |
| Agents that use tools | Unauthorized actions and prompt injection | Least privilege, human approval, logs and rollback |
| AI-generated code | Vulnerabilities and dependency licenses | Human review and automated security scanning |
Where to start: a six-month roadmap
You don’t need certification to govern AI well. An indicative roadmap:
- Month 1: inventory of AI systems and vendors, initial risk classification and scope.
- Month 2: AI policy, committee and owners, and criteria for approving new uses.
- Months 3 and 4: impact assessments, vendor controls, rules for developing with AI, and logging.
- Month 5: metrics, adversarial testing, internal audit and corrective actions.
- Month 6: management review, closing gaps and a decision on ISO/IEC 42001 certification.
Every system should have a business owner and a technical owner. Timelines depend on the company’s size and maturity; they are not requirements of the standard. A practical way to get started is to adopt the NIST AI RMF as a working guide and keep traceability to ISO/IEC 42001 in case you decide to certify later.
What to measure: metrics to track
Governing AI doesn’t end with a policy: you have to show that it works. ISO/IEC 42001 devotes clause 9 to performance evaluation (monitoring and measurement, internal audit and management review) (ISO/IEC 42001, table of contents), and the NIST AI RMF details what is worth reviewing (NIST AIRC). These indicative metrics are a good place to start:
| Metric | What it measures | Reference |
|---|---|---|
| Inventory coverage | AI systems registered versus those in use across the company | NIST GOVERN 1.6 |
| Risk assessments | In-scope systems with a current risk assessment | ISO/IEC 42001, clause 8.2 |
| Oversight of critical cases | High-impact cases with documented human approval and oversight | NIST GOVERN 3.2 |
| AI security | Critical findings detected and resolved in each cycle | NIST MEASURE 2.7 |
| AI incidents | Frequency, severity, time to containment and time to closure | NIST MANAGE 4.3 |
| Vendors | Percentage of critical vendors with a documented assessment | NIST GOVERN 6 and MANAGE 3.1 |
| Quality and effectiveness | Results of tests for accuracy, bias, information leakage and robustness | NIST MEASURE 2.5, 2.10 and 2.11 |
The standard also requires defining roles, responsibilities and authorities (clause 5.3). A common split:
- Leadership or AI committee: approves the policy, the acceptable level of risk and high-impact decisions.
- Process owners: validate the purpose, quality and consequences of each use.
- Technology: designs the development and operational controls.
- Security, privacy and legal: assess access, data, contracts and regulations.
- Internal audit: verifies the evidence and the effectiveness of the controls.
A note for Ecuador: the standard is available as NTE INEN-ISO/IEC 42001, Ecuador’s national adoption, identical to the international standard and voluntary. It was made official in the Registro Oficial (Ecuador’s official gazette) on July 20, 2026 (Registro Oficial, 2026, in Spanish).
No standard on its own guarantees that an AI system will make the right decisions, nor does it replace legal obligations. What they do provide is order: knowing which AI is in use, who is accountable and how it is controlled. At Wasivia we help you assess your maturity and design AI governance tailored to your company. Learn about our AI consulting and strategy service or book a 30-minute assessment.
References
- IEC Webstore — ISO/IEC 42001:2023, AI management system (Dec 18, 2023)
- ISO/IEC 42001:2023 — table of contents (preview of the official document)
- IEC Webstore — ISO/IEC 42006:2025, requirements for certification bodies (Jul 7, 2025)
- IEC Webstore — ISO/IEC 42005:2025, impact assessment (May 28, 2025)
- IEC Webstore — ISO/IEC 23894:2023, AI risk management (Feb 6, 2023)
- IEC Webstore — ISO/IEC 38507:2022, governance (Apr 8, 2022)
- IEC Webstore — ISO/IEC 5338:2023, life cycle (Dec 20, 2023)
- IEC Webstore — ISO/IEC 5259-1:2024, data quality (Jul 2, 2024)
- AWS — Accredited ISO/IEC 42001 certification (Nov 25, 2024)
- Anthropic — ISO/IEC 42001 certification (Jan 13, 2025)
- Microsoft — Azure AI Foundry and Security Copilot achieve ISO/IEC 42001 (Jul 17, 2025)
- NIST — AI Risk Management Framework
- NIST AIRC — AI RMF 1.0, core: the GOVERN, MAP, MEASURE and MANAGE functions
- Cloud Security Alliance — AI Controls Matrix v1.1 (Jun 22, 2026)
- OWASP GenAI Security Project — 2026 Top 10 for LLM Applications (Sep 1, 2026)
- MITRE — ATLAS
- UNESCO — Recommendation on the Ethics of Artificial Intelligence
- OECD — AI Principles
- CEN-CENELEC — EN 18286:2026 published in the Official Journal of the EU (Oct 15, 2026)
- European Commission — Regulatory framework for AI
- Registro Oficial del Ecuador — Supplement No. 329 (Jul 20, 2026) (in Spanish)
