ISO 42001 is the first international standard for AI management systems, published in December 2023. If your organization develops, deploys, or relies on AI systems, this standard defines how to govern them responsibly, with a certifiable framework that regulators and customers are beginning to expect.
This article breaks down what ISO 42001 requires, how certification works, how it relates to ISO 27001 and the EU AI Act, and what implementation actually looks like for IT and compliance teams managing AI-powered tools across a distributed device fleet.
ISO/IEC 42001:2023 establishes requirements for an Artificial Intelligence Management System (AIMS). An AIMS is a structured set of policies, processes, and controls that an organization uses to develop, provide, or use AI responsibly across its full lifecycle.
The standard exists because existing frameworks were not designed with AI-specific risks in mind. ISO 27001 governs information security. ISO 9001 governs quality. Neither addresses algorithmic bias, model drift, AI transparency obligations, or the risk of deploying a system that makes consequential decisions without adequate human oversight.
ISO 42001 fills that gap. It applies to any organization regardless of size, sector, or whether it builds AI internally or procures it from vendors. If you use a third-party AI tool that touches sensitive data or drives business decisions, ISO 42001 is relevant to you.
ISO 42001 follows the Annex SL high-level structure used by ISO 27001, ISO 9001, and other management system standards. This is deliberate: organizations already certified in other ISO frameworks can integrate AIMS requirements into their existing management system architecture without building everything from scratch.
Here is what each clause covers:
The Plan-Do-Check-Act cycle runs through all of this. Planning (Clauses 4-6) feeds into operation (Clause 8), which feeds into evaluation (Clause 9), which drives improvement (Clause 10).
Annex A contains 38 controls organized into 9 control categories. Like ISO 27001's Annex A, these controls are not all mandatory: you select controls based on your risk assessment and document your rationale for any exclusions in a Statement of Applicability.
The 9 control categories are:
1. Policies related to AI (governance frameworks and policy documentation)
2. Internal organization (roles, responsibilities, and escalation paths for AI risk)
3. Resources for AI systems (compute, data, human expertise)
4. Assessing impacts of AI systems (structured impact assessments before deployment)
5. AI system lifecycle (design, development, testing, deployment, and decommissioning controls)
6. Data for AI systems (data provenance, quality, labeling, and bias controls)
7. Information for interested parties about AI systems (transparency and disclosure obligations)
8. Use of AI systems (controls for organizations that use rather than build AI)
9. Third-party and customer relationships (supply chain AI risk management)
For IT teams specifically, several controls have direct operational implications. Control category 5 (AI system lifecycle) requires documented processes for deploying and retiring AI systems, which maps closely to device lifecycle management workflows if AI capabilities are embedded in endpoint tooling. Control category 8 is particularly relevant for organizations procuring AI-powered security or management platforms: you are responsible for governing how those tools are used, not just what they do.
One of the most operationally significant requirements in ISO 42001 is the AI system impact assessment, introduced in Clause 6.1.4 and elaborated in Annex C guidance.
This is distinct from a standard information security risk assessment. An AI system impact assessment asks:
For an IT team managing a fleet of Apple devices, this has practical consequences. If your MDM platform uses AI to flag anomalous device behavior, classify risk scores, or automate remediation actions, that system needs an impact assessment. The same applies to AI-powered endpoint detection tools, automated patch prioritization engines, and any AI assistant your employees access on managed devices.
Building an accurate inventory of AI systems in use is the prerequisite. This is where shadow AI becomes a real problem: employees installing AI tools on managed devices without going through a procurement or risk review process. Your AIMS cannot govern what it cannot see.
The structural similarity between ISO 42001 and ISO 27001 makes integration straightforward, but the two standards address different risk domains.
| Dimension | ISO 27001 | ISO 42001 |
|---|---|---|
| Focus | Confidentiality, integrity, availability of information | Responsible development and use of AI systems |
| Risk type | Security threats to data and systems | Algorithmic bias, opacity, misuse, unintended harm |
| Controls | 93 controls across 4 themes (Annex A) | 38 controls across 9 categories (Annex A) |
| Certification body | Accredited ISO certification bodies | Accredited ISO certification bodies |
| Existing integrations | Widely recognized, mature ecosystem | Newer standard, growing auditor ecosystem |
If you are already ISO 27001 certified, your ISMS documentation, internal audit process, and management review cadence all carry over. You will need to add AI-specific scope documentation, conduct impact assessments for AI systems, and map new controls from ISO 42001 Annex A. Many organizations pursuing ISO 42001 treat it as an extension of their existing ISMS rather than a separate program.
One important clarification: ISO 27001 controls that govern data handling (encryption, access control, logging) remain relevant for AI systems that process personal or sensitive data. You do not replace your ISMS with an AIMS. You extend it.
The EU AI Act, which entered into force in August 2024, takes a risk-tiered regulatory approach to AI. Systems classified as high-risk (medical devices, employment screening, critical infrastructure) face mandatory conformity assessments, technical documentation requirements, and human oversight obligations.
ISO 42001 is not a direct path to EU AI Act compliance, but there is significant overlap. The standard's requirements for AI risk assessment, transparency, data governance, and human oversight align with what the EU AI Act demands of high-risk AI providers and deployers.
The NIST AI Risk Management Framework (AI RMF) provides a complementary voluntary framework organized around four functions: Govern, Map, Measure, and Manage. ISO 42001's clause structure maps reasonably well to NIST AI RMF functions, and organizations pursuing both will find that evidence collected for one often satisfies requirements of the other.
For organizations subject to GDPR compliance requirements, ISO 42001's data controls for AI systems (Annex A category 6) directly address how training data and inference data must be handled, which intersects with GDPR's data minimization and purpose limitation principles.
ISO 42001 certification follows the same two-stage audit process used for ISO 27001:
Stage 1 (Documentation Review): The certification body reviews your AIMS documentation to confirm you have addressed all clause requirements. This includes your AI policy, scope statement, Statement of Applicability, risk assessment methodology, AI system impact assessments, and internal audit records. Stage 1 typically surfaces gaps that need to be addressed before the main audit.
Stage 2 (Conformity Assessment): Auditors verify that your AIMS is implemented and operating as documented. They will interview personnel, test that controls are functioning, and examine evidence of management review and continual improvement. For AI-specific controls, expect auditors to ask for your AI system inventory, completed impact assessments, and records showing human oversight mechanisms are active.
Surveillance Audits: After initial certification, annual surveillance audits confirm the AIMS remains effective. Full recertification occurs every three years.
Practical timeline for a mid-market organization starting from scratch: 9 to 18 months from kickoff to initial certification, depending on the complexity of your AI system landscape and whether you are integrating with an existing ISO 27001 program.
Choosing a certification body matters. Look for accreditation from a national accreditation body (UKAS in the UK, ANAB in the US) and ask auditors specifically about their AI management system audit experience, since ISO 42001 is still a relatively new standard.
A functional AIMS requires a complete inventory of AI systems in scope. For organizations managing distributed device fleets, this is harder than it sounds.
Employees routinely install AI tools without IT visibility: browser extensions that process company data, AI writing assistants connected to cloud services, locally-run models on developer machines. On unmanaged or loosely managed devices, this inventory gap is significant.
ISO 42001 Clause 8.4 requires organizations to control the use of AI systems, which means having a process to discover, evaluate, and either approve or block AI tools that fall within scope. This is an IT operations problem before it is a compliance problem. You need device-level visibility to build a defensible AI system inventory.
Application management and software inventory capabilities in a mature MDM deployment are the starting point. If you can see what applications are installed across your fleet and what processes are running, you can identify AI tools that need to go through a formal impact assessment before they are approved for use.
ISO 42001 compliance involves two things that have to work together: the policies and evidence that satisfy auditors, and the technical controls that enforce your AI governance commitments at the device level. Without enforcement, your documentation is aspirational. Without documentation, your enforcement is invisible to auditors.
Iru Compliance maps ISO 42001 controls to your environment and automates the collection of audit evidence across your AIMS. Instead of manually assembling screenshots and configuration exports before an audit, Iru Compliance continuously monitors your control posture and surfaces gaps in real time. For organizations already using Iru for ISO 27001, adding ISO 42001 as an additional framework means your existing control evidence base carries over wherever controls overlap, and new AI-specific controls are added incrementally rather than rebuilt from the ground up.
Iru Endpoint handles the device-level enforcement that ISO 42001's operational controls depend on. For Apple-first enterprise environments, Iru Endpoint gives IT teams the application visibility and policy enforcement needed to maintain an accurate AI system inventory on managed devices. When your AIMS requires that only approved AI tools are used on managed endpoints, Iru Endpoint is where that restriction is configured and monitored. The same device telemetry that feeds your MDM workflows feeds your compliance evidence, so there is no manual reconciliation between what IT configured and what compliance reported.
For organizations building a shadow AI discovery process as part of their AIMS implementation, Iru Endpoint's application inventory and hardware inventory management capabilities provide the fleet-wide visibility required to identify AI tools in use before they go through a formal impact assessment.
ISO 42001 certification is achievable for mid-market organizations that approach it methodically. Here is a practical sequence:
1. Define scope. Identify which AI systems fall within your AIMS boundary. Start narrow: the AI systems that pose the most significant risk to your organization or that are most likely to face regulatory scrutiny.
2. Build your AI system inventory. Use your MDM and software inventory tooling to discover AI applications across managed devices. Document each system's purpose, data inputs, decision outputs, and affected parties.
3. Conduct impact assessments. For each in-scope AI system, work through the assessment questions in Annex C. Document your findings. This is what auditors will examine first.
4. Map controls to your Statement of Applicability. Select applicable controls from Annex A, document your rationale for any exclusions, and assign ownership.
5. Implement and enforce. Configure technical controls at the device and application level. For endpoint-related controls, this means your MDM policies need to reflect your AIMS commitments.
6. Run internal audits. Before Stage 1, conduct at least one full internal audit cycle to identify gaps and demonstrate that your management review process is functioning.
7. Select a certification body and schedule Stage 1. Allow 6 to 8 weeks between your internal audit and Stage 1 to address identified gaps.
Organizations that treat compliance automation as an ongoing operational discipline rather than an audit-season event will find the certification process significantly less disruptive.
AI governance gets easier when enforcement and evidence live in the same place. Iru offers device management and compliance automation as separate products or together, depending on where you are today. If you are building toward ISO 42001 certification and want device-level enforcement and audit evidence collection working in parallel, Book a demo.