Complementary management, risk, security and ethics frameworks come together in AI governance.

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 standardWhat they requireTypical evidence
4 and 5: context and leadershipScope of the system, AI policy and assigned rolesScope document, approved policy, named owners
6: planningObjectives, risks and impactsRisk register, impact assessment, treatment plan
7 and 8: support and operationResources, competence, documentation and life cycle controlProcedures, training, controls over development and use
9 and 10: evaluation and improvementMonitoring, audit, review and correctionMetrics, 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:

StandardWhat it’s forPublished
ISO/IEC 23894:2023Guidance on managing AI riskFebruary 2023
ISO/IEC 42005:2025Assessing the impact of AI systems on people and societyMay 2025
ISO/IEC 38507:2022Governance implications of AI, aimed at top managementApril 2022
ISO/IEC 5338:2023Life cycle processes: design, testing, deployment and retirementDecember 2023
ISO/IEC 5259 seriesData quality for analytics and machine learningSince 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:

CriterionISO/IEC 42001NIST AI RMFCSA AICMOWASP
Organizational governanceVery highVery highMediumLow
Risk managementVery highVery highHighTechnical
LLM and agent securityGeneralGeneralHighVery high
Impact assessmentHighHighMediumPartial
Specific technical controlsMediumMediumVery highVery high
Organizational certificationYesNoNo equivalentNo

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 useMain riskKey control
Assistants for customers or employeesFabricated answers or wrong guidanceValidated knowledge base, evaluation and a correction channel
Decisions about people (hiring, credit, performance reviews)Bias, unfairness, lack of explanationHuman oversight and a right to review
Search over internal documents (RAG)Exposure of information without authorizationPer-document access control and leakage testing
Agents that use toolsUnauthorized actions and prompt injectionLeast privilege, human approval, logs and rollback
AI-generated codeVulnerabilities and dependency licensesHuman review and automated security scanning

Where to start: a six-month roadmap

You don’t need certification to govern AI well. An indicative roadmap:

  1. Month 1: inventory of AI systems and vendors, initial risk classification and scope.
  2. Month 2: AI policy, committee and owners, and criteria for approving new uses.
  3. Months 3 and 4: impact assessments, vendor controls, rules for developing with AI, and logging.
  4. Month 5: metrics, adversarial testing, internal audit and corrective actions.
  5. 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:

MetricWhat it measuresReference
Inventory coverageAI systems registered versus those in use across the companyNIST GOVERN 1.6
Risk assessmentsIn-scope systems with a current risk assessmentISO/IEC 42001, clause 8.2
Oversight of critical casesHigh-impact cases with documented human approval and oversightNIST GOVERN 3.2
AI securityCritical findings detected and resolved in each cycleNIST MEASURE 2.7
AI incidentsFrequency, severity, time to containment and time to closureNIST MANAGE 4.3
VendorsPercentage of critical vendors with a documented assessmentNIST GOVERN 6 and MANAGE 3.1
Quality and effectivenessResults of tests for accuracy, bias, information leakage and robustnessNIST 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

Similar Posts