Written and maintained by CASRAI Editorial Board
Published · Last updated
On this page
Organizations building enterprise AI and cybersecurity governance frequently need to align three complementary international standards published by ISO/IEC JTC 1: ISO/IEC 42001:2023 (Artificial Intelligence Management System, AIMS), ISO/IEC 27001:2022 (Information Security Management System, ISMS), and ISO/IEC 23894:2023 (Artificial Intelligence — Guidance on Risk Management). While ISO/IEC 42001 and ISO/IEC 27001 are both certifiable Type A Management System Standards built on the identical ISO Harmonized Structure (Clauses 4–10), ISO/IEC 23894 is a non-certifiable Type B guidance standard that adapts ISO 31000:2018 risk methodology to execute the AI risk assessments mandated by ISO/IEC 42001 Clause 6.1.2.
Ask CASRAI · free to tryQuestion about this guide? Ask CASRAI answers with linked sources. 2 free questions a day, no account.
At-a-Glance Comparison: ISO/IEC 42001 vs. ISO/IEC 27001 vs. ISO/IEC 23894
| Evaluation Dimension | ISO/IEC 42001:2023 (AIMS) | ISO/IEC 27001:2022 (ISMS) | ISO/IEC 23894:2023 (AI Risk Guidance) |
|---|---|---|---|
| Standard Type & Certifiability | Type A Requirements standard (normative shall clauses); eligible for accredited third-party certification via ISO/IEC 42006 bodies. |
Type A Requirements standard (normative shall clauses); eligible for accredited third-party certification via ISO/IEC 27006-1 bodies. |
Type B Guidance standard (informative recommendations using should); not certifiable on its own. |
| ISO/IEC Subcommittee & Edition | ISO/IEC JTC 1/SC 42 (Artificial intelligence), 1st edition (December 2023). | ISO/IEC JTC 1/SC 27 (Information security, cybersecurity and privacy protection), 3rd edition (October 2022). | ISO/IEC JTC 1/SC 42 (Artificial intelligence), 1st edition (February 2023). |
| Primary Governance Scope | Responsible development, provision, and use of AI systems across the full AI lifecycle (fairness, transparency, human oversight, data quality, safety). | Preservation of the Confidentiality, Integrity, and Availability (CIA triad) of organizational information assets across all IT systems and media. | Practical methodology for identifying, analyzing, evaluating, and treating AI-specific risks within enterprise risk management (adapting ISO 31000:2018). |
| Clause Architecture | ISO Harmonized Structure (Annex SL): Clauses 4 through 10. | ISO Harmonized Structure (Annex SL): Clauses 4 through 10. | ISO 31000:2018 structure: Clauses 4 (Principles), 5 (Framework), and 6 (Process). |
| Risk vs. Impact Assessment Mandate | Dual mandate: (1) AI Risk Assessment & Treatment (Clauses 6.1.2–6.1.3) and (2) mandatory AI System Impact Assessment (Clause 6.1.4 & Annex A.5) evaluating foreseeable impacts on individuals, groups, and society. | Single focus on Information Security Risk Assessment & Treatment (Clauses 6.1.2–6.1.3) protecting organizational information assets; no societal/individual impact assessment requirement. | Provides the step-by-step risk identification, source analysis, and evaluation methodology cited by ISO/IEC 42001 Clause 6.1.2 Note. |
| Annex A Controls & Objectives | 38 controls across 9 control objectives (A.2 through A.10), with implementation guidance built into Annex B. |
93 controls consolidated into 4 themes (A.5 Organizational [37], A.6 People [8], A.7 Physical [14], A.8 Technological [34]), with guidance in ISO/IEC 27002:2022. |
No normative control catalog; Annex A provides a reference list of AI risk sources and trustworthiness objectives. |
| Statement of Applicability (SoA) | Required under Clause 6.1.3(d): compares selected AI controls against the 38 Annex A controls and justifies inclusions/exclusions by AI role (Provider, Developer, Producer, User). | Required under Clause 6.1.3(d): compares selected security controls against the 93 Annex A controls with implementation status and exclusion justifications. | Not applicable. |
What Is the Difference Between ISO/IEC 42001 (AIMS) and ISO/IEC 27001 (ISMS)?
Although both standards govern digital technology risk and share the same management system skeleton, they solve fundamentally different engineering and governance problems:
- Organizational Asset Protection vs. Stakeholder & Societal Impact: In ISO/IEC 27001:2022, risk is evaluated primarily from the perspective of harm to the organization’s information assets (unauthorized disclosure, tampering, or downtime). In ISO/IEC 42001:2023, Clause 4.1 (Context of the Organization) and Clause 6.1.4 (AI System Impact Assessment) require the organization to evaluate how an AI model’s outputs, autonomy, or failure modes affect external individuals, patient cohorts, protected demographic groups, and society—covering algorithmic bias, safety hazards, human rights, and economic or clinical harm.
- Deterministic IT Security vs. Probabilistic AI Lifecycle Governance: ISO/IEC 27001 Annex A controls (such as access control, cryptography, network segregation, and vulnerability management) secure the infrastructure hosting an application. However, an AI system can be hosted inside a fully ISO/IEC 27001-compliant cloud environment and still produce hallucinated clinical advice, exhibit demographic bias due to unrepresentative training data, or suffer post-deployment covariate shift. ISO/IEC 42001 addresses these AI-intrinsic risks through dedicated Annex A domains:
A.5Assessing impacts of AI systems: Documenting intended consequences, foreseeable misuse, and demographic/societal impact assessments (further detailed in ISO/IEC 42005:2025).A.6AI system life cycle: Defining responsible design objectives, verification and validation metrics, continuous post-deployment monitoring, and event logging across model training and inference.A.7Data for AI systems: Governing training, validation, and test data acquisition, provenance, bias screening, preparation, and data quality lineage.A.8Information for interested parties: Providing external users, regulators, and affected individuals with model transparency documentation, system capabilities, limitations, and incident reporting channels.A.9Use of AI systems: Enforcing human oversight, human-in-the-loop override authority, and strict adherence to intended use boundaries.
- Supply-Chain Role Scoping (Provider, Developer, Producer, User): Unlike ISO/IEC 27001, where most controls apply uniformly to the scoped information boundary, ISO/IEC 42001 Clause 4.1 requires the organization to explicitly determine its role along the AI value chain (AI Provider, AI Developer, AI Producer, or AI User/Deployer), which directly shapes which of the 38 Annex A controls are applicable in its Statement of Applicability (SoA).
How Do ISO/IEC 42001 and ISO/IEC 27001 Share the Harmonized Structure (Annex SL) for Integrated Audits?
Because both standards conform to the ISO/IEC Directives, Part 1, Annex SL (Harmonized Structure for Management System Standards), organizations that already hold an ISO/IEC 27001:2022 certificate can avoid building a separate AI governance silo. Instead, they can implement an Integrated Management System (IMS) and pursue a combined ISO/IEC 42001 certification audit:
| Harmonized Clause | Shared ISMS / AIMS Governance Backbone (Reusable ~80%+) | AI-Specific Delta Required for ISO/IEC 42001 |
|---|---|---|
| Clause 4: Context of the Organization | External/internal issue tracking (4.1), interested-party register (4.2), and management system scope definition (4.3). | Explicit determination of organizational AI roles (Provider, Producer, Developer, User) and regulatory mapping (e.g., EU AI Act, FDA SaMD). |
| Clause 5: Leadership | Executive commitment (5.1), policy approval lifecycle (5.2), and governance committee charters (5.3). | Adoption of an explicit AI Policy (`A.2.2`) and assignment of AI oversight and human-in-the-loop accountability roles (`A.3.2`). |
| Clause 6: Planning | Risk assessment/treatment mechanics, risk register format, Statement of Applicability (SoA) workflow, and measurable objectives (6.2). | Addition of Clause 6.1.4 AI System Impact Assessment, AI-specific risk sources (`ISO/IEC 23894`), and the 38-control AIMS Statement of Applicability. |
| Clause 7: Support | Resource allocation (7.1), training/competence tracking (7.2), awareness (7.3), communication (7.4), and Documented Information control (7.5). | Role-specific AI ethics, ML engineering, and human-oversight competence verification, plus documentation of data/compute/tooling resources (`A.4`). |
| Clause 8: Operation | Operational planning, change management, and risk treatment execution (8.1–8.3). | Execution of Clause 8.4 AI System Impact Assessment at defined lifecycle stages and operationalization of Annex A.6–A.10 controls. |
| Clauses 9 & 10: Evaluation & Improvement | KPI measurement (9.1), Internal Audit program (9.2), Management Review agenda (9.3), and Nonconformity/Corrective Action tracker (10.1–10.2). | Expanding internal auditor competence to cover ISO/IEC 42001 clauses and adding model drift/fairness metrics to Management Review inputs. |
Furthermore, ISO/IEC 42001 Annex D explicitly maps cross-domain integration with cybersecurity (ISO/IEC 27001) and privacy information management (ISO/IEC 27701), allowing security controls such as A.8.25 (Secure development life cycle) and A.5.19 (Information security in supplier relationships) in ISO/IEC 27001 to link directly to AI lifecycle (`A.6`) and third-party AI supplier (`A.10`) controls in ISO/IEC 42001.
How Does ISO/IEC 23894 Complement Certifiable ISO/IEC 42001 Requirements?
A common misconception in AI compliance is treating ISO/IEC 23894 vs. ISO/IEC 42001 as competing standards. In reality, they occupy the exact same relationship in AI governance that ISO/IEC 27005 occupies relative to ISO/IEC 27001 in cybersecurity, or that the NIST AI Risk Management Framework (AI RMF) occupies alongside AIMS certification:
- Normative Requirement vs. Methodological Guidance: ISO/IEC 42001 Clause 6.1.2 requires the organization to define and apply an AI risk assessment process, and the official Note to Clause 6.1.2 states: “The organization can use ISO/IEC 23894 as guidance for performing risk management related to AI.”
- AI-Specific Risk Sources in ISO/IEC 23894 Annex A: Whereas traditional IT risk frameworks focus on confidentiality/integrity/availability threats, ISO/IEC 23894:2023 details how to identify and analyze risk sources unique to machine learning and autonomous systems, including: (1) complexity of the environment and degree of autonomy, (2) lack of transparency and explainability (black-box inference), (3) statistical and societal bias in training data or objective functions, (4) continuous learning and concept/covariate drift in production, (5) hardware/compute dependencies, and (6) adversarial ML attacks (data poisoning, model inversion, and prompt injection).
Frequently Asked Questions
What is the difference between ISO/IEC 42001 (AI Management System) and ISO/IEC 27001 (Information Security Management System)?
While both are certifiable Management System Standards built on the ISO Harmonized Structure (Clauses 4–10), ISO/IEC 27001:2022 protects organizational information assets across Confidentiality, Integrity, and Availability using 93 cybersecurity, physical, and organizational controls. ISO/IEC 42001:2023 governs the responsible lifecycle development, provision, and use of AI systems using 38 AI-specific Annex A controls—addressing training data provenance (A.7), model verification and drift monitoring (A.6), human oversight (A.9), stakeholder transparency (A.8), and mandatory AI System Impact Assessments on individuals and society (Clause 6.1.4 and A.5).
How do ISO/IEC 42001 and ISO/IEC 27001 share the Harmonized Structure (Annex SL) for integrated audits?
Because both standards share identical Clause 4–10 numbering and management system mechanics under the ISO Harmonized Structure (formerly Annex SL), organizations with an existing ISO/IEC 27001 ISMS can operate an Integrated Management System (IMS). Document control, competence tracking, internal audit schedules, management review meetings, and corrective action workflows (Clauses 4, 5, 7, 9, and 10) can be unified and audited concurrently during combined Stage 1, Stage 2, and surveillance audits.
How does ISO/IEC 23894 (AI risk management guidance) complement certifiable ISO/IEC 42001 requirements?
ISO/IEC 42001 Clause 6.1.2 mandates an AI risk assessment process and explicitly references ISO/IEC 23894 as guidance. ISO/IEC 23894:2023 is a non-certifiable guidance standard that adapts ISO 31000:2018 risk management principles to AI-specific risk sources such as algorithmic bias, autonomy, lack of explainability, model drift, and adversarial attacks—explaining how to execute the AI risk assessments that ISO/IEC 42001 requires for certification.







