Artificial Intelligence Management Systems (ISO 42001)
ISO/IEC 42001 is the first certifiable international standard for managing artificial intelligence. It specifies the requirements for an AI Management System (AIMS): the policies, processes, and controls an organization uses to develop, provide, or use AI responsibly, and follows the same management system model as ISO 27001.
Who Needs It
ISO 42001 applies to any organization with AI in its products or operations, in any of three roles: developing AI models or systems, providing AI-based products and services, or using AI supplied by others. You do not need to be an AI company; a SaaS vendor embedding a third-party large language model (LLM) into a feature is in scope just as much as the model developer behind it.
The standard exists because AI systems behave differently from conventional software. They learn from data, adapt over time, and produce outputs that are probabilistic rather than deterministic, which introduces risks that traditional IT governance does not address: model drift, training data bias, lack of explainability, hallucinated outputs, adversarial manipulation, and harm to people affected by automated decisions. ISO 42001 gives organizations a management system for identifying and treating those risks, and gives customers and regulators a recognized way to verify that the work is real. With AI-specific regulation emerging in multiple jurisdictions, certification is increasingly showing up in enterprise procurement and vendor due diligence.
How the Standard Is Structured
Like other International Organization for Standardization (ISO) management system standards, ISO 42001 follows the harmonized structure: clauses 4 through 10 define the mandatory management system requirements, covering organizational context, leadership, planning, support, operation, performance evaluation, and improvement. Two AI-specific requirements sit at the center of the operation clause: an AI risk assessment and treatment process, and an AI system impact assessment that evaluates consequences for individuals, groups, and society as well as for the organization itself.
Annex A contains 38 reference controls organized into nine control objectives, spanning AI policy, internal organization and accountability, resources for AI systems, impact assessment, the AI system life cycle, data management, information for interested parties, responsible use of AI, and third-party and supplier relationships. As with ISO 27001, you select which Annex A controls apply based on your risk assessment and justify inclusions and exclusions in a Statement of Applicability.
The remaining annexes are guidance rather than requirements: Annex B provides implementation guidance for the Annex A controls, Annex C catalogs AI-related organizational objectives and risk sources to seed your risk assessment, and Annex D discusses applying the standard across domains and sectors.
Because the structure mirrors ISO 27001, organizations with an existing ISMS can extend it rather than building a parallel system. Much of the machinery (document control, internal audit, management review, corrective action) is shared; the AI-specific additions are the impact assessment, the AI life cycle controls, and the data and transparency requirements.
Certification Process
ISO 42001 certification follows the standard accredited audit model. A certification body first performs a Stage 1 audit reviewing your AIMS documentation and readiness, then a Stage 2 audit examining whether the management system and selected controls are implemented and operating. A successful audit yields a certificate valid for three years, maintained through annual surveillance audits and a full recertification audit at the end of the cycle.
Before engaging an auditor, you need the core AIMS artifacts in place and generating records: a defined scope, an AI policy, completed risk and impact assessments, a Statement of Applicability, measurable objectives, and evidence that internal audit and management review have run at least once.
For the artifacts that typically satisfy manually evidenced control areas, see Example Evidence.