AI Management System Standards (ISO/IEC 42001 and 23894)

1. Why This Topic Matters

Organizations are deploying artificial intelligence faster than their governance structures can keep up. Regulators (such as the EU AI Act), customers, auditors and boards are now asking a simple question: Can you prove that your AI is managed responsibly? Ad hoc ethics statements no longer satisfy that question. What satisfies it is a structured, auditable management system with documented policies, risk assessments, controls and evidence of continual improvement.

ISO/IEC 42001 and ISO/IEC 23894 are the two international standards that provide this structure. For the ISACA Advanced in AI Risk (AAIR) candidate, they sit at the heart of the AI Risk Governance and Framework Integration domain because they answer the questions an AI risk professional is paid to answer: How do we govern AI across the enterprise? How do we assess and treat AI risk in a repeatable way? How do we integrate AI governance with the quality, security and privacy management systems the organization already runs? Expect multiple exam questions that test whether you understand the purpose, structure and relationship of these two standards rather than simply their names.

2. What ISO/IEC 42001 Is

ISO/IEC 42001:2023 (published December 2023) is the world's first certifiable AI Management System (AIMS) standard. It specifies requirements for establishing, implementing, maintaining and continually improving a management system for organizations that develop, provide or use AI systems. Key characteristics:

• It is a requirements standard: it uses the word shall, and conformance can be independently audited and certified (certification body requirements are defined in ISO/IEC 42006).
• It follows the ISO Harmonized Structure (formerly Annex SL), the same high-level structure used by ISO/IEC 27001 (information security), ISO 9001 (quality) and ISO/IEC 27701 (privacy). This is what makes integration with existing management systems practical.
• It is technology-neutral and sector-neutral: it governs the organization's processes around AI, not the technical design of any specific model.
• It is built on the Plan-Do-Check-Act (PDCA) cycle, emphasizing continual improvement rather than a one-time compliance exercise.

The main clauses (4 to 10)

• Clause 4 Context of the organization: determine internal and external issues, interested parties and their requirements, the organization's role (developer, provider, user), and the scope of the AIMS.
• Clause 5 Leadership: top management must demonstrate commitment, establish an AI policy, and assign roles, responsibilities and authorities. Accountability sits at the top.
• Clause 6 Planning: this is the risk heart of the standard. It requires an AI risk assessment process (6.1.2), an AI risk treatment process (6.1.3) producing a Statement of Applicability (SoA) that justifies inclusion or exclusion of Annex A controls, and an AI system impact assessment (6.1.4) that evaluates consequences for individuals, groups and society. It also requires measurable AI objectives and planning of changes.
• Clause 7 Support: resources, competence, awareness, communication and documented information.
• Clause 8 Operation: operational planning and control, and actually carrying out the risk assessment, risk treatment and impact assessment at planned intervals or when significant changes occur.
• Clause 9 Performance evaluation: monitoring and measurement, internal audit and management review.
• Clause 10 Improvement: nonconformity, corrective action and continual improvement.

The annexes

• Annex A (normative): a reference set of 38 controls grouped under 9 control objectives: policies related to AI; internal organization; resources for AI systems (data, tooling, computing, human); assessing impacts of AI systems; AI system life cycle; data for AI systems; information for interested parties; use of AI systems; and third-party and customer relationships. Like ISO 27001's Annex A, these are selected and justified via the Statement of Applicability.
• Annex B (informative): implementation guidance for each Annex A control.
• Annex C (informative): potential AI-related organizational objectives (such as fairness, safety, security, transparency, explainability, accountability, privacy, robustness, environmental impact) and risk sources that feed the risk assessment.
• Annex D (informative): guidance on using the AIMS across sectors and integrating it with other management system standards.

3. What ISO/IEC 23894 Is

ISO/IEC 23894:2023 (published February 2023) is Guidance on AI risk management. Its defining features:

• It is a guidance standard: it uses should, and it is not certifiable. Nobody earns an ISO 23894 certificate.
• It is explicitly built on ISO 31000:2018, the generic enterprise risk management standard. It takes the ISO 31000 principles, framework and process and adds AI-specific considerations at each step.
• It is intended for any organization that develops, produces, deploys or uses AI products, systems and services, and it helps them integrate AI risk management into existing enterprise risk management (ERM).

Structure of ISO/IEC 23894

• Principles (mirroring ISO 31000): risk management should be integrated, structured and comprehensive, customized, inclusive, dynamic, based on the best available information, consider human and cultural factors, and drive continual improvement, all with AI-specific commentary.
• Framework: leadership and commitment, integration, design, implementation, evaluation and improvement of the risk management framework within the organization.
• Process: communication and consultation; establishing scope, context and criteria; risk assessment (risk identification, risk analysis, risk evaluation); risk treatment; monitoring and review; and recording and reporting.
• Annex A: common AI-related objectives such as accountability, AI expertise, availability and quality of training and test data, environmental impact, fairness, maintainability, privacy, robustness, safety, security, and transparency and explainability.
• Annex B: AI-specific risk sources including complexity of the environment, lack of transparency and explainability, level of automation, machine-learning-related risks (data quality, drift, bias, adversarial inputs), system hardware issues, system life cycle issues, and technology readiness.
• Annex C: a mapping of the risk management process to the AI system life cycle (inception, design and development, verification and validation, deployment, operation and monitoring, re-evaluation, retirement), showing where each risk activity belongs.

4. How the Two Standards Work Together

The simplest way to remember the relationship: 42001 is the management system; 23894 is the risk management method inside it.

• ISO/IEC 42001 Clause 6.1.2 requires an AI risk assessment process and points to ISO/IEC 23894 for guidance on how to perform it.
• ISO/IEC 23894 tells you how to identify, analyze, evaluate and treat AI risk consistently with ISO 31000; ISO/IEC 42001 tells you what the organization must have in place (policy, roles, objectives, SoA, audits, management review) and makes it auditable.
• Both draw on the same terminology from ISO/IEC 22989 (AI concepts and terminology) and both reference the AI life cycle described in ISO/IEC 5338.
• Related standards you may see in distractors: ISO/IEC 38507 (governance implications of AI for governing bodies, aimed at boards), ISO/IEC TR 24028 (trustworthiness in AI), ISO/IEC TR 24027 (bias), ISO/IEC 42005 (AI system impact assessment guidance) and ISO/IEC 42006 (requirements for AIMS certification bodies).

Integration with other frameworks

• ISO/IEC 27001 and 27701: because of the shared Harmonized Structure, an organization can run a single integrated management system covering security, privacy and AI, sharing the context analysis, leadership commitments, document control, internal audit and management review. Annex D of 42001 encourages exactly this.
• NIST AI RMF: a voluntary US framework organized into Govern, Map, Measure and Manage functions. It is not certifiable and is often mapped to 42001 controls; NIST and ISO publish crosswalks. Exam questions may ask you to distinguish a voluntary framework (NIST AI RMF) from a certifiable standard (42001) from guidance (23894).
• EU AI Act: requires risk management systems and quality management systems for high-risk AI; ISO/IEC 42001 is widely used as a vehicle to demonstrate alignment, although certification to 42001 alone is not legal compliance.
• COBIT and ISACA guidance: 42001's leadership and performance evaluation clauses align with COBIT's governance (EDM) and management objectives, supporting board-level oversight.

5. How It Works in Practice: Implementation Flow

1. Define context and scope: inventory AI systems, determine the organization's role for each (developer, provider, user), identify stakeholders and legal obligations.
2. Secure leadership commitment: top management issues an AI policy and assigns accountability (for example an AI governance committee and system owners).
3. Establish risk criteria consistent with ERM and risk appetite (ISO 23894 context step).
4. Perform AI risk assessment using 23894 guidance and Annex B risk sources; assess likelihood and consequence against Annex A objectives (fairness, safety, robustness, etc.).
5. Perform AI system impact assessment for affected individuals, groups and society (42001 6.1.4, with 42005 as detailed guidance).
6. Select and justify controls from 42001 Annex A, producing the Statement of Applicability and a risk treatment plan with risk owners.
7. Operate controls across the AI life cycle: data governance, documentation, human oversight, testing, monitoring for drift, incident handling, third-party management.
8. Measure, audit and review: KPIs on AI objectives, internal audit, management review, corrective action and continual improvement.
9. Optionally seek certification by an accredited body against 42001.

6. Exam Tips: Answering Questions on AI Management System Standards (ISO/IEC 42001 and 23894)

Tip 1 - Lock in the core distinction. If a question asks which standard is certifiable, specifies requirements, or establishes a management system, the answer is ISO/IEC 42001. If it asks for guidance on AI risk management, something based on ISO 31000, or help integrating AI risk into ERM, the answer is ISO/IEC 23894. Watch for the words shall (42001) versus should (23894).

Tip 2 - Know the Harmonized Structure clause numbers. Questions often describe an activity and ask where it fits. Context equals Clause 4, Leadership equals Clause 5, Planning and risk assessment equal Clause 6, Support equals Clause 7, Operation equals Clause 8, Performance evaluation (internal audit, management review) equals Clause 9, Improvement equals Clause 10.

Tip 3 - Distinguish risk assessment from impact assessment. AI risk assessment (6.1.2) considers risk to the organization and its objectives; AI system impact assessment (6.1.4) considers consequences for individuals, groups and society. Both are required by 42001. An answer that collapses them into one is usually wrong.

Tip 4 - Remember the Statement of Applicability. Just as in ISO 27001, Annex A controls are a reference set; the organization must document which controls are included or excluded and why. If a question asks how an organization justifies omitting an Annex A control, the SoA is the answer.

Tip 5 - Think like top management. ISACA questions on management systems reward answers that emphasize leadership commitment, defined accountability, documented policy and continual improvement over purely technical fixes. When two answers both seem correct, the one that embeds the control in governance (policy, roles, review) is usually the best answer.

Tip 6 - Place standards correctly among their neighbors. Boards and governing bodies point to ISO/IEC 38507; terminology points to ISO/IEC 22989; life cycle points to ISO/IEC 5338; trustworthiness characteristics point to ISO/IEC TR 24028; certification bodies point to ISO/IEC 42006; detailed impact assessment points to ISO/IEC 42005. Distractors frequently swap these.

Tip 7 - Framework integration questions. If asked how to avoid duplication when an organization already holds ISO 27001, the answer involves leveraging the shared Harmonized Structure to build an integrated management system (shared context, policy framework, audit and management review), not running a separate parallel program. If asked to compare with NIST AI RMF, remember NIST is voluntary and non-certifiable with Govern, Map, Measure, Manage functions, and that crosswalks to 42001 exist.

Tip 8 - Life cycle thinking. ISO/IEC 23894 Annex C maps risk activities to AI life cycle stages. Questions about when to reassess risk expect answers such as at planned intervals, when significant changes occur, upon model retraining or redeployment, or when drift or incidents are detected, rather than only at project inception.

Tip 9 - Eliminate absolutes. Options stating that 42001 certification guarantees legal compliance, eliminates AI risk, or replaces the need for ERM are wrong. The standards support compliance and manage risk; they do not eliminate it.

Tip 10 - Read for role. 42001 applies to developers, providers and users of AI. A scenario about an organization that only procures AI from a vendor still falls within scope; relevant Annex A areas include third-party and customer relationships and use of AI systems. Do not assume the standard applies only to AI builders.

7. Quick Reference Summary

• ISO/IEC 42001:2023: AI Management System requirements; certifiable; Harmonized Structure clauses 4 to 10; PDCA; AI policy; AI risk assessment, risk treatment and Statement of Applicability; AI system impact assessment; Annex A with 38 controls in 9 objectives; Annexes B, C, D for guidance, objectives/risk sources and integration.
• ISO/IEC 23894:2023: Guidance on AI risk management; not certifiable; based on ISO 31000 principles, framework and process; Annex A AI objectives; Annex B AI risk sources; Annex C mapping to the AI life cycle.
• Relationship: 42001 requires a risk process and references 23894 as the method; together they let an organization govern AI in an auditable, risk-based, continually improving way that integrates with existing security, privacy and quality management systems.

Master these distinctions, be able to describe the flow from context to leadership to risk assessment to controls to audit, and you will handle any AAIR question on AI management system standards with confidence.