In short: ISO/IEC 42001:2023 is the first international standard for an AI management system: a set of rules, roles and processes that lets an organisation develop, buy and operate AI responsibly. It is certifiable and follows the same structure as ISO/IEC 27001. For the EU AI Act (Regulation (EU) 2024/1689) it provides a solid framework of inventory, risk assessment, roles and documentation. A certificate does not, however, replace the legal conformity of an individual AI system.

What an AI management system under ISO 42001 is

A management system is not a tool but a framework of policies, responsibilities, objectives, processes and their regular review. ISO/IEC 42001 applies this to AI. The standard calls it an AIMS (Artificial Intelligence Management System) and addresses every organisation that provides, develops or uses AI systems, regardless of size or sector.

The core consists of clauses 4 to 10. They follow the common structure of all ISO management system standards, the Harmonized Structure:

  • Context (clause 4). The organisation's role in relation to AI, its interested parties and the scope of the system.
  • Leadership (clause 5). Top management adopts an AI policy and assigns responsibilities.
  • Planning (clause 6). AI risk assessment, risk treatment and an impact assessment covering effects on individuals, groups and society.
  • Support and operation (clauses 7 and 8). Resources, competence, documentation and carrying out the planned measures.
  • Evaluation and improvement (clauses 9 and 10). Internal audits, management review, corrective action.

Annex A adds AI-specific controls. They range from the AI policy, resources and impact assessment to the system life cycle, data quality, information for affected parties, responsible use, and suppliers and customers. As with ISO 27001, a Statement of Applicability records which controls apply and why others are excluded. Annex B gives implementation guidance, Annex C lists typical objectives and risk sources.

What the EU AI Act requires

The EU AI Act classifies AI systems by risk. The risk class determines the obligations:

  • Prohibited practices (Article 5). Such as social scoring or manipulative AI.
  • High-risk AI (Article 6, Annexes I and III). For example AI in recruitment, credit scoring, education, law enforcement or critical infrastructure. These systems need risk management, data governance, technical documentation, logging, human oversight and cybersecurity (Articles 9 to 15).
  • Transparency obligations (Article 50). Such as making chatbots recognisable and labelling AI-generated content.
  • Minimal risk. No specific obligations.

Who carries which obligation depends on the role. Providers (Article 3(3)) develop an AI system or place it on the market under their own name. Deployers (Article 3(4)) use it under their own authority, which covers most companies and public bodies. Anyone who substantially modifies a high-risk system or offers it under their own name becomes a provider (Article 25).

Article 4 applies regardless of risk class: providers and deployers promote AI literacy among staff who work with AI systems. The duty has applied since 2 February 2025 and was reframed in July 2026 as a duty to promote: the law no longer requires a specific level of competence per person.

Deadlines after the Digital Omnibus

The AI Act applies in stages. Prohibited AI practices and the AI literacy duty have applied since 2 February 2025, obligations for providers of general-purpose AI models since 2 August 2025. Regulation (EU) 2026/1744 of 8 July 2026, the so-called Digital Omnibus on AI, amended the remaining deadlines. It entered into force on 27 July 2026.

  • 2 August 2026. Most transparency duties under Article 50 still apply from this date.
  • 2 December 2027. Obligations for high-risk systems under Annex III.
  • 2 August 2028. Obligations for high-risk systems embedded in products under Annex I.

In Germany, the KI-MIG market surveillance act has defined since 29 July 2026 which authority supervises compliance, as a rule the Federal Network Agency (Bundesnetzagentur). What this means for public bodies is covered in our article EU AI Act High-Risk Systems: Checklist for Public-Sector Authorities. The extra time is best used for the inventory, classification and building governance.

How ISO 42001 helps and where its limits are

Many AI Act requirements are organisational: an overview of all AI systems, risk management across the life cycle, clear responsibilities, documented decisions, trained staff. These are exactly the processes ISO 42001 establishes. Providers of high-risk systems also need a quality management system under Article 17. An existing AI management system supplies building blocks for it, such as document control, responsibilities, risk treatment and internal audits.

The limit is clear: an ISO 42001 certificate attests a working management system, not the legal conformity of an individual AI system. The standard is not a harmonised standard within the meaning of Article 40 of the AI Act and therefore does not create a presumption of conformity. Conformity assessment, the EU declaration of conformity and CE marking (Articles 43, 47 and 48) remain separate steps.

AI governance and ISO 27001 as an integrated management system

If you already run an information security management system (ISMS) under ISO/IEC 27001, you do not start from scratch with AI. Both standards share the Harmonized Structure, with the same clauses and terms. Policies, document control, internal audits and management review can be run jointly. Such an integrated management system avoids duplicate work and keeps evidence in one place.

In substance the two complement each other. ISO 27001 protects information in terms of confidentiality, integrity and availability. ISO 42001 additionally asks about fairness, explainability, data quality and effects on affected people. Attacks such as prompt injection, meaning manipulated inputs that make a language model take unintended actions, belong in both risk assessments. For the technical safeguards, the BSI IT-Grundschutz offers suitable modules, see BSI IT-Grundschutz for AI Models. We have been certified to ISO/IEC 27001:2022 ourselves since 26 May 2026 and know the audit process first-hand.

Six steps to get there

  1. AI inventory. Record all AI systems in use or planned, including AI features in purchased software and freely used chat services. For each system: purpose, data, provider, affected people, contribution to decisions.
  2. Role and risk class. For each system, record whether you are provider or deployer and which risk class applies. Obtain a legal assessment for borderline cases under Annex III.
  3. Scope and policy. Management defines which areas the AI management system covers, adopts the AI policy and appoints owners.
  4. AI risk assessment and impact assessment. Assess risks to the organisation and effects on affected people, select controls from Annex A, draw up the Statement of Applicability. Bodies governed by public law using high-risk systems also need the fundamental rights impact assessment under Article 27.
  5. AI literacy. Role-based training for management, business units, development and IT security. Document attendance as evidence for Article 4.
  6. Operation, audit, improvement. Bind suppliers by contract, monitor logs and incidents, plan internal audits. Certification by an accredited certification body can follow.

How GermanAI Defense helps

We build the AI inventory with you, classify roles and risk classes, set up the AI management system under ISO/IEC 42001 and integrate it with your ISMS. AI literacy training under Article 4 is available through the GAD Academy. Our AI Solutions support you in developing and operating your own AI applications.

→ More on our GRC and compliance consulting: GRC and compliance consulting

→ The information security foundation: ISO 27001 consulting

Sources

As of 28 October 2026. This article is a professional assessment and does not replace legal advice in individual cases.