ISO/IEC 42001: Requirements, AI Controls, and Implementation Guide
- 13 minutes to read
Table of Contents
What is ISO/IEC 42001?
ISO/IEC 42001 is the first international standard for an Artificial Intelligence Management System (AIMS), published in December 2023 to help organizations responsibly balance AI innovation with governance and risk management. It provides a structured way for organizations to govern how they develop, provide, or use AI systems.
The standard covers areas such as AI risk management, accountability, transparency, data management, impact assessment, system monitoring, and continuous improvement. It follows a management-system approach similar to standards such as ISO/IEC 27001, allowing AI governance processes to be integrated with existing organizational controls.
ISO 42001 has the following clauses:
- Scope: Defines the purpose, applicability, and boundaries of the artificial intelligence management system.
- Normative references: Identifies ISO/IEC 22989:2022 as the normative reference for AI concepts and terminology.
- Terms and definitions: Establishes the management system and AI-specific terminology used throughout the standard.
- Context of the organization: Requires organizations to identify relevant internal and external issues, interested parties, AI roles, and AIMS scope.
- Leadership: Requires top management to establish an AI policy, assign responsibilities, provide support, and demonstrate commitment to the AIMS.
- Planning: Covers AI risks and opportunities, risk assessment and treatment, impact assessments, AI objectives, and planned changes.
- Support: Defines requirements for resources, competence, awareness, communication, and control of documented information.
- Operation: Requires organizations to implement planned controls, conduct risk and impact assessments, and manage AI-related operational activities.
- Performance evaluation: Covers monitoring, measurement, analysis, internal audits, and management reviews to evaluate AIMS effectiveness.
- Improvement: Requires continual improvement and processes for addressing nonconformities, root causes, and corrective actions.
ISO 42001 has the following annexes defining controls for AI risk:
- A.2 Policies related to AI: Establishes controls for defining, maintaining, reviewing, and communicating policies for responsible AI development and use.
- A.3 Internal organization: Defines responsibilities, accountability, and reporting mechanisms for AI governance and oversight.
- A.4 Resources for AI systems: Covers the data, software, infrastructure, personnel, tools, and other resources required for AI systems.
- A.5 Assessing impacts of AI systems: Provides controls for identifying and documenting potential impacts of AI systems on individuals, groups, and society.
- A.6 AI system life cycle: Covers controls for AI system design, development, testing, deployment, operation, maintenance, change, and retirement.
- A.7 Data for AI systems: Addresses data acquisition, preparation, quality, provenance, management, appropriate use, and documentation.
- A.8 Information for interested parties: Covers transparency about AI system purposes, limitations, appropriate use, and mechanisms for reporting concerns.
- A.9 Use of AI systems: Defines controls for responsible AI operation, monitoring, human oversight, user competence, and adherence to intended use.
- A.10 Third-party and customer relationships: Addresses AI risks involving suppliers, service providers, partners, and customers, including responsibilities and external dependencies.
What Are the Objectives of ISO 42001?
ISO/IEC 42001 is designed to help organizations manage AI systems responsibly through a structured AI management system. Its objectives focus on governance, risk management, transparency, accountability, and continual improvement across the AI lifecycle.
- Establish responsible AI governance: Define clear policies, roles, responsibilities, and decision-making processes for AI systems, aligned with business goals, legal requirements, and stakeholder expectations.
- Manage AI risks and opportunities: Identify, assess, and treat AI-related risks such as bias, inaccurate outputs, privacy issues, security weaknesses, and lack of oversight while pursuing AI benefits responsibly.
- Improve transparency and accountability: Maintain documentation and traceability for AI design, data use, controls, monitoring, and decision ownership to support audits, oversight, and stakeholder communication.
- Support continuous improvement of AI systems: Monitor performance, conduct audits, review controls, address weaknesses, and adapt the AI management system as technology, risks, and legal requirements change.
ISO 42001 Clauses and Requirements Explained
ISO/IEC 42001 is structured into ten clauses. Clauses 1 through 3 establish the standard’s scope, references, and terminology. Clauses 4 through 10 contain the core requirements organizations must follow to establish, operate, evaluate, and improve an artificial intelligence management system.
Clause 1: Scope
Clause 1 defines the purpose and applicability of ISO/IEC 42001. It states that the standard specifies requirements and provides guidance for establishing, implementing, maintaining, and continually improving an artificial intelligence management system.
The clause makes clear that the standard applies to any organization, regardless of size, type, or nature, that provides or uses products or services that utilize AI systems. It is intended to help organizations develop, provide, or use AI systems responsibly.
It also establishes that the management system operates within the context of the organization. This means the controls and processes used should reflect the organization’s size, objectives, AI use cases, and level of risk.
Clause 2: Normative References
Clause 2 identifies the documents that are indispensable for applying ISO/IEC 42001. It contains a single normative reference: ISO/IEC 22989:2022, which covers artificial intelligence concepts and terminology.
This reference supplies many of the AI-specific terms the standard relies on. Organizations implementing ISO/IEC 42001 may need to consult it when interpreting requirements or defining terms in their own AI policies and procedures.
This clause is short, but it is important because it ties ISO/IEC 42001 to the broader ISO/IEC vocabulary for AI and keeps terminology consistent across related standards.
Clause 3: Terms and Definitions
Clause 3 defines the terminology used throughout ISO/IEC 42001. It combines terms drawn from ISO/IEC 22989 with standard management system terms and AI-specific concepts such as AI system impact assessment and data quality.
Consistent terminology is particularly important for AI governance because different teams may use the same terms in different ways. Technical teams, legal teams, risk managers, auditors, and executives need a shared vocabulary when developing policies and controls.
The definitions in this clause also support more consistent implementation. They reduce the chance that requirements are interpreted differently across departments or locations.
Clause 4: Context of the Organization
Clause 4 requires organizations to understand the internal and external factors that can affect their artificial intelligence management system. Organizations must also determine their role in relation to AI systems, such as AI provider, producer, customer, or partner, and the intended purpose of the AI systems they develop or use.
Organizations must identify interested parties and understand their relevant needs and expectations. These parties may include customers, employees, regulators, suppliers, business partners, and people affected by AI-supported decisions.
The clause also requires the organization to define the scope of the management system and then establish, implement, maintain, and continually improve it. The scope determines which AI systems, processes, departments, locations, and activities are covered.
Clause 5: Leadership
Clause 5 focuses on the role of top management in AI governance. Senior leaders must demonstrate commitment to the artificial intelligence management system and ensure that it supports the organization’s strategic direction.
Management must establish an AI policy, assign responsibilities, and make sure relevant roles have sufficient authority. This helps avoid situations where AI risks are recognized but no one is clearly responsible for addressing them.
Leadership must also ensure that AI governance is integrated into existing business processes. This includes providing resources, supporting compliance activities, and promoting awareness of responsible AI practices throughout the organization.
Clause 6: Planning
Clause 6 requires organizations to identify risks and opportunities that could affect the artificial intelligence management system. It also requires them to define an AI risk assessment process, with criteria for identifying, analyzing, and evaluating AI risks.
Organizations must define an AI risk treatment process, select controls, and compare them against Annex A to confirm that no necessary controls have been left out. The result is documented in a Statement of Applicability. The clause also requires a process for AI system impact assessments.
Planning also includes setting AI objectives, measurable where practicable, and defining how they will be achieved. When changes to the management system are needed, they must be carried out in a planned way.
Clause 7: Support
Clause 7 covers the resources and organizational support needed to operate the artificial intelligence management system. This includes personnel, technical resources, competence, awareness, communication, and documented information.
Organizations must ensure that people performing AI-related work have the necessary skills and knowledge. This may involve training on AI risks, governance requirements, data handling, monitoring procedures, or specific technical systems.
The clause also requires organizations to control documentation. Policies, procedures, records, assessments, and other relevant information should be kept current, protected from unauthorized changes, and available to the people who need them.
Clause 8: Operation
Clause 8 focuses on putting into practice the processes and controls planned under Clause 6. Organizations must manage AI-related activities in a controlled way, including externally provided processes, products, and services such as third-party AI systems.
Organizations must perform AI risk assessments at planned intervals and when significant changes are proposed or occur. They must then implement the AI risk treatment plan and retain documented information on the results.
AI system impact assessments follow the same schedule. Organizations must consider how an AI system could affect individuals, groups, and society, and keep records of the results. Planned changes must be controlled, and the consequences of unintended changes reviewed.
Clause 9: Performance Evaluation
Clause 9 requires organizations to determine whether the artificial intelligence management system is operating effectively. This involves monitoring, measurement, analysis, and evaluation of relevant processes and controls.
Organizations should define what will be measured, how measurements will be performed, and how results will be reviewed. Relevant metrics may include incidents, control effectiveness, system performance, audit findings, or progress toward AI objectives.
The clause also requires internal audits and management reviews. These activities provide structured ways to identify weaknesses, verify whether requirements are being followed, and determine whether the management system remains suitable and effective.
Clause 10: Improvement
Clause 10 first requires organizations to continually improve the suitability, adequacy, and effectiveness of their artificial intelligence management system. Audit results, monitoring data, incidents, feedback, and management reviews should all feed into this.
The clause then addresses nonconformities and corrective action. When a problem is identified, the organization must react to it, correct it, and deal with its consequences.
Corrective action should address the root cause, not only the immediate issue. Organizations must also check whether similar nonconformities exist or could occur in other AI systems, processes, or business units, and confirm that the actions they took were effective.
ISO 42001 Annex A: Controls for Managing AI Risk
Annex A of ISO/IEC 42001 provides a reference set of controls that organizations can use to manage AI-related risks and support the requirements of the artificial intelligence management system. Organizations select applicable controls based on their risks, objectives, regulatory obligations, and use of AI.
The controls are organized into several areas covering governance, resources, impact assessment, system development, data, transparency, AI use, and external relationships.
A.2: Policies Related to AI
This section focuses on establishing and maintaining policies for the responsible development and use of AI. Organizations should define principles, responsibilities, and rules that guide AI-related activities.
Policies should be reviewed regularly and updated when technologies, risks, business objectives, or regulatory requirements change. They should also be communicated to relevant employees and other stakeholders.
A.3: Internal Organization
These controls address the internal roles and responsibilities required to govern AI effectively. Organizations should clearly assign responsibility for AI systems, risk management, oversight, and decision-making.
The controls also support reporting mechanisms for concerns related to AI systems. Clear accountability helps ensure that issues are identified, escalated, and addressed by the appropriate people.
A.4: Resources for AI Systems
This section covers the resources needed to develop, operate, and manage AI systems. Resources may include data, software, computing infrastructure, personnel, tools, and technical documentation.
Organizations should understand how these resources affect the performance and risk profile of an AI system. They should also ensure that resources remain suitable throughout the system lifecycle.
A.5: Assessing Impacts of AI Systems
These controls require organizations to assess the potential impacts of AI systems on individuals, groups, and society. Assessments should consider both intended and unintended consequences.
Impact assessments can help identify issues such as unfair outcomes, privacy risks, safety concerns, or inappropriate use. Organizations should document the results and use them when determining appropriate controls and risk treatments.
A.6: AI System Life Cycle
This section covers controls for managing AI systems throughout their lifecycle, including design, development, testing, deployment, operation, maintenance, and retirement.
Organizations should define requirements for AI systems, verify that those requirements are met, and document key development activities. Controls should also address changes to systems and ensure that risks are reassessed when significant modifications occur.
A.7: Data for AI Systems
These controls focus on the data used to develop and operate AI systems. Organizations should establish processes for acquiring, preparing, managing, and maintaining data.
Data quality, relevance, origin, and appropriate use should be considered because poor or unsuitable data can directly affect AI system behavior. Organizations should also document important characteristics and limitations of datasets where appropriate.
A.8: Information for Interested Parties
This section addresses the information organizations provide to people and organizations affected by AI systems. Relevant information may include the purpose of the system, its limitations, how it should be used, and how users can report problems.
The controls support transparency by helping interested parties understand important aspects of an AI system. The level and type of information provided should reflect the system’s context, risks, and intended audience.
A.9: Use of AI Systems
These controls apply to organizations that use AI systems, including systems obtained from external providers. Organizations should define processes for responsible use and ensure that AI systems are operated according to their intended purpose.
This includes monitoring system operation, following usage instructions, applying appropriate human oversight, and addressing risks that arise during use. Organizations should also consider whether users have the competence needed to operate AI systems appropriately.
A.10: Third-Party and Customer Relationships
This section covers risks related to suppliers, service providers, partners, and customers involved with AI systems. Organizations should define responsibilities and requirements when AI-related activities depend on external parties.
Controls may address procurement, contractual requirements, information sharing, monitoring, and changes in third-party services. These measures help ensure that external dependencies do not create unmanaged risks within the organization’s AI management system.
How to Implement ISO 42001 in Your Organization
Implementing ISO/IEC 42001 involves building a formal artificial intelligence management system around the organization’s AI activities. The process should connect governance, risk management, operational controls, monitoring, and continual improvement rather than treating them as separate tasks.
Define the Scope of the AI Management System
Start by defining which business units, locations, processes, products, and AI systems fall within the AIMS. The scope should reflect how the organization develops, provides, procures, or uses AI.
Document important boundaries and dependencies, including shared infrastructure, external providers, and business processes that rely on AI. The scope should also consider regulatory obligations, customer requirements, organizational objectives, and the needs of relevant interested parties.
A clearly defined scope helps prevent gaps later in implementation. It also makes it easier to determine which systems need risk assessments, controls, monitoring, and audit coverage.
Inventory AI Systems and Use Cases
Create and maintain an inventory of AI systems and significant AI use cases across the organization. This should include internally developed systems, third-party tools, embedded AI features, production deployments, and relevant pilots or prototypes.
For each system, record details such as its purpose, business owner, technical owner, provider, users, data sources, model type, deployment environment, and intended outcomes. It is also useful to capture whether the system supports or makes decisions that affect people.
The inventory should be reviewed regularly because AI use can change quickly. New systems, features, and integrations should be added through a defined intake or approval process.
Related content: Read our article about shadow AI risks and how to prevent them
Identify Stakeholders and Governance Responsibilities
Identify the internal and external parties that are involved in or affected by AI systems. These may include developers, users, customers, suppliers, regulators, legal teams, security teams, risk managers, and people affected by AI-assisted decisions.
Define clear responsibilities for approving AI systems, managing risks, conducting assessments, monitoring performance, handling incidents, and maintaining documentation. Each important activity should have an accountable role rather than relying on informal ownership.
Governance responsibilities should also define escalation paths. Teams need to know who can approve higher-risk use cases, suspend a system, accept residual risk, or require corrective action.
Establish an AI Governance Policy
Develop an AI governance policy that sets the organization’s expectations for responsible AI development and use. The policy should cover topics such as accountability, transparency, risk management, human oversight, data quality, security, privacy, and regulatory compliance.
The policy should explain which AI activities require review or approval and who is responsible for those decisions. Supporting standards and procedures can provide more detailed requirements for specific teams or system types.
Leadership should approve the policy and ensure that it is communicated to relevant personnel. The policy should also be reviewed periodically so that it remains aligned with changes in technology, regulation, and organizational priorities.
Related content: Read our overview of AI regulations and LLM regulations
Perform AI Risk Assessments
Establish a documented method for identifying, analyzing, evaluating, and treating AI-related risks. The method should define risk criteria so that different systems can be assessed consistently.
Assess risks such as inaccurate outputs, bias, privacy violations, security weaknesses, poor data quality, lack of explainability, model drift, misuse, and insufficient human oversight. The assessment should also consider the severity and likelihood of potential impacts.
Risk assessments should not be static. Repeat them when there are significant changes to models, datasets, intended uses, system integrations, operating environments, or regulatory requirements.
Conduct AI System Impact Assessments
Perform impact assessments to understand how an AI system may affect individuals, groups, organizations, or society. These assessments should consider both expected benefits and possible negative consequences.
Document the system’s intended use, affected stakeholders, decision context, potential harms, existing safeguards, and residual impacts. Higher-impact systems may require deeper analysis and stronger review before deployment.
Impact assessments should be connected to the broader risk management process. Their findings should influence control selection, approval decisions, monitoring requirements, and whether additional human oversight is necessary.
Select and Implement Appropriate Controls
Use the results of risk and impact assessments to determine which controls are necessary. Annex A of ISO/IEC 42001 provides a reference set of controls covering governance, data, lifecycle management, transparency, AI use, and third-party relationships.
Document which controls are selected, why they are applicable, and how they are implemented. Where controls are not selected, the organization should be able to explain the reasoning based on its risk treatment approach.
Controls should be practical and proportionate to risk. A low-impact internal automation tool may require fewer safeguards than an AI system used for employment, credit, safety, or other high-impact decisions.
Establish AI Monitoring and Documentation Processes
Define how AI systems will be monitored after deployment. Monitoring may cover system performance, output quality, incidents, user complaints, model drift, changes in input data, and deviations from intended use.
Set thresholds or criteria for when issues require investigation or escalation. Monitoring should also define who reviews results and what actions can be taken if performance or risk levels become unacceptable.
Maintain documented information such as policies, approvals, risk assessments, impact assessments, test results, monitoring records, incidents, and corrective actions. This documentation supports traceability and provides evidence during audits and management reviews.
Measure AIMS Performance
Define metrics that show whether the artificial intelligence management system is operating effectively. Measures may include the number of unresolved AI risks, audit findings, incident trends, control effectiveness, training completion, or progress against AI governance objectives.
Metrics should be useful for decision-making rather than collected only for reporting. Organizations should define how each measure is calculated, who owns it, how often it is reviewed, and what thresholds indicate a problem.
Performance results should be reported to appropriate governance bodies and management. Trends can help identify weak controls, repeated issues, emerging risks, or areas where additional resources are needed.
Conduct Internal Audits and Management Reviews
Perform internal audits at planned intervals to assess whether the AIMS conforms to ISO/IEC 42001 requirements and the organization’s own policies and procedures. Audits should examine both documentation and evidence that controls are functioning in practice.
Audit programs should consider the importance and risk level of different AI activities. Higher-risk systems or areas with previous findings may need more frequent or detailed review.
Top management should also conduct formal management reviews. These reviews should consider audit results, risk changes, incidents, stakeholder feedback, performance metrics, resource needs, and opportunities to improve the AIMS.
Address Nonconformities and Continuously Improve the AIMS
When a requirement is not met, document the nonconformity and take action to correct the immediate issue. The organization should then investigate the underlying cause rather than only treating the visible symptom.
Corrective actions should be tracked to completion and reviewed to confirm that they are effective. Organizations should also determine whether similar problems could exist in other AI systems, teams, or processes.
Continual improvement should use evidence from audits, monitoring, incidents, risk assessments, stakeholder feedback, and management reviews. This helps the AIMS remain effective as AI technologies, business activities, and regulatory expectations change.
Supporting AI Oversight and Monitoring with Exabeam Agent Behavior Analytics
Exabeam Agent Behavior Analytics (ABA) gives security teams visibility into AI agent activity across the digital workforce. It applies behavioral intelligence to establish baseline normal activity and surface subtle anomalies across AI agents and autonomous workflows, then detects misuse, drift, abnormal tool usage, risky access, and actions outside an agent’s intended role. To support compliance with ISO 42001, Exabeam contributes to open-source initiatives like Praxen, which provides pre-deployment Agent Behavior Verification (ABV), and Observra, which standardizes runtime telemetry to ensure AI systems operate within their intended boundaries. ABA covers major AI platforms as well as custom agents, and correlates agent activity with users, entities, applications, and security events, helping organizations monitor AI use and investigate emerging risk before it impacts operations.
Key capabilities of Exabeam Agent Behavior Analytics:
- Behavioral detections for AI agents: Baselines expected agent behavior and flags autonomous installation and execution, tool usage and MCP activity, and activity outside expected roles.
- Coverage for any agent: Analyzes activity from Claude, ChatGPT, Google Gemini, Microsoft Copilot, GitHub Copilot, and custom agents. Observra standardizes agent telemetry, while Exabeam Agent Sensor collects and normalizes activity across models, agents, and runtime environments.
- AI security posture measurement: Outcomes Navigator maps ABA coverage to the OWASP Agentic Top 10, identifies gaps in data and detections, and uses Exabeam Nova AI advisor capability to recommend data sources and detections that improve readiness.
- Automated investigation of agent activity: Machine-built timelines correlate agent actions with related user, entity, and application activity, while Nova AI summarizes threats, explains behavioral context, and recommends next steps with analysts remaining in control.
- Customizable detection rules: ABA detections are available in Threat Detection Management, where teams can review, clone, tune, and customize them. Exabeam Nova AI rule creator generates behavioral models and correlation rules from natural language.
- Flexible deployment: Add ABA to an existing SIEM with New-Scale Analytics, or deploy New-Scale Fusion for integrated detection, investigation, and response across human and non-human identities.
Learn More About Exabeam
Learn about the Exabeam platform and expand your knowledge of information security with our collection of white papers, podcasts, webinars, and more.
-
Webinar
The Insider Risk Program Just Got Bigger: How Digital Workers Are Changing Identity, Behavior, Ownership, and...
-
Blog
What Is the Difference Between Exabeam and CrowdStrike for SIEM and Security Operations?
- Show More