AI Governance Operating Model - ISO 42001 Alignment

This document defines how a centralised AI governance function operates in alignment with ISO 42001:2023 (AI Management System) while leveraging the control framework (guardrails, Judge, HITL) and enterprise AI platforms.

Executive Summary

ISO 42001 establishes requirements for an AI Management System (AIMS)-a structured approach to governing AI across the organisation. This document maps:

1. ISO 42001 clauses → Framework controls
2. Centralised governance function → Roles, responsibilities, processes
3. Platform capabilities → Technical implementation (Bedrock, Databricks, Foundry)
4. Operating rhythm → Governance cadence and decision rights

Core principle: Centralised policy, federated execution, unified assurance.

ISO 42001 Structure Overview

ISO 42001 follows the Annex SL high-level structure common to all ISO management systems:

Clause	Title	Focus
4	Context of the organisation	Understand internal/external factors affecting AI
5	Leadership	Management commitment and policy
6	Planning	Risk assessment and objectives
7	Support	Resources, competence, awareness, communication
8	Operation	AI system lifecycle controls
9	Performance evaluation	Monitoring, measurement, audit, review
10	Improvement	Nonconformity, corrective action, continual improvement

Annex A provides AI-specific controls (similar to ISO 27001 Annex A for information security).

Governance Operating Model

Three Lines Model for AI

Line	Function	AI Governance Role
1st Line	Business / Engineering	Build and operate AI systems; execute controls
2nd Line	AI Governance / Risk	Set policy; provide oversight; operate assurance
3rd Line	Internal Audit	Independent assurance; validate effectiveness

Centralised vs Federated

Aspect	Centralised (AI Governance Function)	Federated (Business Units)
Policy	✅ Sets AI policy and standards	Implements policy
Risk framework	✅ Defines risk tiers and methodology	Performs risk assessments
Control standards	✅ Defines control requirements	Implements controls
Tooling/platforms	✅ Selects and governs platforms	Uses platforms
Judge configuration	✅ Defines evaluation criteria	May customise for use case
HITL processes	✅ Defines requirements	Staffs and operates
Monitoring	✅ Aggregates and reports	Monitors own systems
Approval (CRITICAL)	✅ Approves CRITICAL deployments	Requests approval
Approval (HIGH)	Reviews	✅ Approves with risk sign-off
Approval (MEDIUM/LOW)	Spot checks	✅ Self-service with standards

ISO 42001 Clause Mapping

Clause 4: Context of the Organisation

4.1 Understanding the organisation and its context

Requirement	Framework Implementation
Identify external factors	Regulatory landscape (EU AI Act, sector regulations), competitive environment, technology trends
Identify internal factors	AI maturity, risk appetite, existing capabilities, culture

4.2 Understanding the needs and expectations of interested parties

Interested Party	Needs/Expectations	How Addressed
Regulators	Compliance, transparency, human oversight	Risk classification, HITL, audit trail
Customers	Fair treatment, privacy, accuracy	Guardrails, Judge quality monitoring, bias testing
Employees	Clear guidance, training, tools	Policy, standards, platform governance
Board/Executives	Risk management, value delivery	Governance reporting, metrics
Data subjects	Rights respected, explanations	HITL for decisions, documentation

4.3 Determining the scope of the AIMS

Scope statement should define: - Which AI systems are in scope (all? production only? above certain risk tier?) - Which lifecycle stages (development, deployment, operation, retirement) - Organisational boundaries - Exclusions and justifications

4.4 AI Management System

The AIMS comprises: - This governance operating model - The control framework (AI.1–AI.12) - Platform implementations (guardrails, Judge, HITL) - Supporting processes and documentation

Clause 5: Leadership

5.1 Leadership and commitment

Requirement	Implementation
Top management demonstrates commitment	Board/ExCo AI governance charter
Policy aligned with strategic direction	AI strategy linked to business strategy
Integration with business processes	AI controls embedded in SDLC, procurement, vendor management
Resources available	Funded AI governance function, platform investment
Achieving intended outcomes	Metrics and reporting to leadership

5.2 AI Policy

The AI Policy should address:

AI Policy - Key Elements

1. Purpose and scope
2. Alignment with organisational values
3. Commitment to responsible AI principles
4. Risk-based approach to AI governance
5. Roles and responsibilities
6. Compliance with applicable laws and regulations
7. Commitment to human oversight
8. Continuous improvement commitment
9. Review and update cycle

5.3 Organisational roles, responsibilities, and authorities

Role	Responsibilities	Authority
Board / ExCo	Approve AI strategy and policy; oversee AI risk	Final accountability
AI Governance Committee	Approve CRITICAL deployments; review AI risk posture; escalation body	Approval for CRITICAL tier
Chief AI Officer / Head of AI	Lead AI strategy and governance function	Policy decisions
AI Risk Lead	Manage AI risk framework; oversee assessments	Risk methodology
AI Ethics Lead	Advise on ethical considerations; review sensitive use cases	Ethics review
Business Unit AI Leads	Implement AI governance in business units; own AI systems	LOCAL execution
AI Platform Team	Operate AI platforms; implement technical controls	Platform standards
HITL Reviewers	Review Judge findings; make decisions on flagged items	Review decisions

Clause 6: Planning

6.1 Actions to address risks and opportunities

Framework Component	ISO 42001 Alignment
AI.2.1 Risk Classification	Systematic identification of AI risks
AI.2.2 Risk Assessment	Evaluation of likelihood and impact
Risk tier controls	Risk treatment through proportionate controls
Judge monitoring	Ongoing risk identification

Risk Assessment Integration:

AI Risk Assessment Process

1. Identify

• Use case analysis
• Data sensitivity
• Decision impact
• Regulatory scope

2. Classify

• Apply risk tier methodology
• Document classification rationale
• Obtain appropriate approval

3. Treat

• Select controls per tier
• Implement guardrails, Judge, HITL
• Document residual risk

4. Monitor

• Judge ongoing evaluation
• HITL review findings
• Periodic reassessment

6.2 AI objectives and planning to achieve them

Objective Category	Example Objectives	Metrics
Safety	Zero critical safety incidents	Incident count
Fairness	No significant bias in CRITICAL systems	Bias metrics
Quality	>95% Judge quality scores	Quality scores
Compliance	100% regulatory compliance	Audit findings
Oversight	100% HITL coverage for CRITICAL	HITL coverage
Transparency	Documentation complete for all HIGH+	Documentation audit

Clause 7: Support

7.1 Resources

Resource Category	Requirements
AI Governance Function	Staffed with AI risk, ethics, technical expertise
Platform investment	Guardrails, Judge, HITL tooling
HITL capacity	Reviewers scaled to volume and SLAs
Training	AI governance training for all AI practitioners
External expertise	Access to legal, regulatory, ethics advisors

Governance Function Sizing:

Organisation Size	AI Portfolio	Recommended FTE
Small (<5 AI systems)	Low complexity	1-2 FTE (may be part-time)
Medium (5-20 AI systems)	Mixed tiers	3-5 FTE
Large (20-100 AI systems)	Multiple CRITICAL	8-15 FTE
Enterprise (100+ AI systems)	Complex, regulated	15-30+ FTE

7.2 Competence

Role	Required Competencies
AI Governance Lead	AI/ML fundamentals, risk management, regulatory knowledge
AI Risk Analyst	Risk assessment, AI systems, data analysis
AI Ethics Specialist	Ethics frameworks, bias assessment, stakeholder engagement
Platform Engineer	AI platforms, guardrails implementation, MLOps
HITL Reviewer	Domain expertise, judgement, documentation

7.3 Awareness

All personnel involved with AI should be aware of: - The AI policy - Their contribution to AIMS effectiveness - Implications of not conforming - How to report concerns or incidents

7.4 Communication

What	Who	When	How
AI policy	All staff	Onboarding, annually	Intranet, training
Governance requirements	AI practitioners	Project initiation	Standards documents
Risk decisions	Stakeholders	As needed	Governance forum
Performance metrics	Leadership	Quarterly	Dashboard, report
Incidents	Affected parties	As required	Incident process

7.5 Documented information

Document Type	Retention	Owner
AI Policy	Current + 3 years	AI Governance
Risk assessments	Life of system + 7 years	Business owner
Control documentation	Life of system + 7 years	Platform team
Judge evaluations	Per tier (90 days – 7 years)	Platform team
HITL decisions	Per tier (1 – 7 years)	HITL team
Audit reports	7 years	Internal Audit

Clause 8: Operation

8.1 Operational planning and control

This is where the framework controls (AI.1–AI.12) are implemented.

AI System Lifecycle Governance:

Stage	Governance Activities	Controls
Ideation	Use case review, initial risk screening	AI.2.1
Design	Risk assessment, control selection	AI.2.2, AI.3
Development	Secure development, testing	AI.4, AI.5, AI.6
Pre-deployment	Security review, approval	AI.4.3
Deployment	Guardrails, logging enabled	AI.7, AI.11
Operation	Judge monitoring, HITL review	AI.8, AI.9
Change	Change assessment, re-approval if needed	AI.2, AI.4
Retirement	Decommissioning, data retention	AI.3

8.2 AI risk assessment

See AI.2 Risk Management controls.

8.3 AI risk treatment

Risk Tier	Treatment Approach
CRITICAL	Full controls, 100% Judge, 100% HITL decisions
HIGH	Full controls, 20-50% Judge sampling, HITL escalation
MEDIUM	Standard controls, 5-10% Judge sampling, periodic HITL
LOW	Basic controls, optional Judge, spot checks

8.4 AI system impact assessment

For HIGH and CRITICAL systems, conduct: - Fundamental Rights Impact Assessment (FRIA) - required for EU AI Act high-risk - Data Protection Impact Assessment (DPIA) - if personal data involved - Bias and Fairness Assessment - especially for consequential decisions

Clause 9: Performance Evaluation

9.1 Monitoring, measurement, analysis, and evaluation

Governance Metrics Framework:

Category	Metric	Target	Source
Coverage	% AI systems in inventory	100%	Inventory
Coverage	% HIGH+ with guardrails	100%	Platform
Coverage	% CRITICAL with 100% Judge	100%	Platform
Quality	Judge quality score (avg)	>90%	Judge logs
Quality	HITL false positive rate	<20%	HITL records
Timeliness	HITL SLA compliance	>95%	HITL records
Incidents	AI safety incidents	0 critical	Incident log
Compliance	Regulatory findings	0 major	Audit reports
Maturity	Maturity assessment score	Per target	Assessment

Governance Dashboard:

9.2 Internal audit

Audit Type	Frequency	Scope	Auditor
AIMS audit	Annual	Full management system	Internal Audit
Control effectiveness	Annual	Technical controls sample	Internal Audit / 2nd Line
CRITICAL system review	Annual per system	Deep dive on CRITICAL systems	Internal Audit
HITL quality	Quarterly	Sample of HITL decisions	AI Governance
Judge calibration	Quarterly	Judge accuracy vs human	AI Governance

9.3 Management review

AI Governance Committee - Quarterly Review Agenda:

1. Performance metrics review
2. Incident summary
3. Risk posture update
4. Regulatory developments
5. CRITICAL system status
6. Audit findings and actions
7. Resource and capability needs
8. Emerging risks and opportunities
9. Improvement actions

Clause 10: Improvement

10.1 Continual improvement

Improvement sources: - Judge findings → guardrail updates - HITL feedback → Judge calibration - Incidents → process improvements - Audits → control enhancements - Regulatory changes → policy updates - Technology advances → capability improvements

10.2 Nonconformity and corrective action

Nonconformity Type	Example	Response
Policy violation	AI deployed without approval	Stop, assess, remediate, prevent recurrence
Control failure	Guardrails bypassed	Investigate, fix, strengthen controls
HITL failure	SLAs consistently missed	Root cause, capacity adjustment
Incident	Customer harm from AI output	Incident response, customer remediation, process fix

Platform Integration

How Platforms Support ISO 42001

ISO 42001 Requirement	Bedrock	Databricks	Foundry
Risk treatment (8.3)	Guardrails	AI Gateway + Judge	AIP governance
Monitoring (9.1)	CloudWatch + custom	Unity Catalog + Lakehouse Monitoring	Inference history
Audit trail (7.5)	Inference logging	Inference Tables	Audit logs
Human oversight (8.4)	Build yourself	Review App + Deployment Jobs	Ontology workflows
Documentation (7.5)	Build yourself	MLflow tracking	Notepad + Model docs
Change control (8.1)	Build yourself	Model Registry + Deployment Jobs	Versioning + releases

Recommended Architecture

Governance Operating Rhythm

Daily

Activity	Owner	Inputs
HITL queue processing	HITL Reviewers	Judge findings
Immediate escalation handling	On-call	Critical findings
Monitoring review	Platform Team	Dashboards

Weekly

Activity	Owner	Inputs
HITL metrics review	HITL Lead	Queue metrics
Guardrail tuning	Platform Team	False positive data
Incident triage	AI Risk Lead	Open incidents

Monthly

Activity	Owner	Inputs
Portfolio review	AI Governance Lead	Inventory changes
Metrics reporting	AI Governance Lead	All metrics
Judge calibration review	Platform Team	Calibration data

Quarterly

Activity	Owner	Inputs
AI Governance Committee	Committee Chair	Performance report
Maturity assessment	AI Governance Lead	Assessment criteria
Policy review	AI Governance Lead	Regulatory changes
HITL quality audit	AI Governance	Sample reviews

Annually

Activity	Owner	Inputs
AIMS internal audit	Internal Audit	Full scope
Management review	Senior Leadership	Annual report
Policy update	AI Governance Lead	All inputs
External audit (if certified)	External Auditor	Full scope
Training refresh	AI Governance	All AI practitioners

Certification Considerations

Preparing for ISO 42001 Certification

Phase	Activities	Duration
Gap assessment	Compare current state to ISO 42001	4-6 weeks
Remediation	Close gaps, implement missing controls	3-6 months
Documentation	Complete AIMS documentation	2-3 months
Internal audit	Verify readiness	2-4 weeks
Management review	Confirm readiness	1-2 weeks
Stage 1 audit	Documentation review	1-2 days
Stage 2 audit	Implementation audit	3-5 days
Certification	Certificate issued	-

Common Certification Gaps

Gap	Remediation
Incomplete AI inventory	Systematic discovery and registration
Missing risk assessments	Retrospective assessments for existing systems
No HITL for CRITICAL	Implement HITL processes
Inadequate documentation	Document existing practices
No Judge/quality assurance	Implement async evaluation
Governance not formalised	Establish committee, define roles

Annex A Control Mapping

ISO 42001 Annex A provides AI-specific controls organised into nine sections (A.2 through A.10). The mapping below shows how the framework's core control families (AI.1 through AI.16) align with Annex A requirements. For the complementary infrastructure-level mapping (IAM, LOG, NET, DAT, etc.), see the ISO 42001 Annex A Infrastructure Mapping.

A.2 Policies Related to AI

ISO 42001 Control	Framework Mapping	Notes
A.2.2 AI policy	AI.1.1 AI Policy Framework	✅ Covered
A.2.3 Roles and responsibilities	AI.1.2 Governance Structure, AI.1.3 Accountability	✅ Covered
A.2.4 Resources	AI.14 Security Awareness (training)	✅ Covered

A.3 Internal Organisation

ISO 42001 Control	Framework Mapping	Notes
A.3.2 AI system inventory	AI.3.1 AI System Inventory	✅ Covered
A.3.3 AI system requirements	AI.4 Development Security	✅ Covered
A.3.4 Third-party components	AI.13 Supplier Management, AI.6 Model Security	✅ Covered

A.4 Resources for AI Systems

ISO 42001 Control	Framework Mapping	Notes
A.4.2 AI risk assessment	AI.2.2 Risk Assessment	✅ Covered
A.4.3 AI risk treatment	AI.2.3 + Risk tier controls	✅ Covered
A.4.4 Residual risk acceptance	Governance approval workflows	✅ Covered

A.5 Assessing Impacts of AI Systems

ISO 42001 Control	Framework Mapping	Notes
A.5.2 Data acquisition	AI.5 Data Governance	✅ Covered
A.5.3 Data quality	AI.5.2 Data Quality	✅ Covered
A.5.4 Data provenance	AI.13.3 Model Provenance (includes data)	✅ Covered
A.5.5 Data preparation	AI.5 Data Governance	✅ Covered

A.6 AI System Life Cycle

ISO 42001 Control	Framework Mapping	Notes
A.6.2.2 Design	AI.4.1 Secure Development	✅ Covered
A.6.2.3 Data processing	AI.5 Data Governance	✅ Covered
A.6.2.4 Model building	AI.4 Development Security	✅ Covered
A.6.2.5 Validation	AI.6.2 Model Validation	✅ Covered
A.6.2.6 Verification	AI.4.2 Testing	✅ Covered
A.6.2.7 Deployment	AI.4.3 Deployment Security	✅ Covered

A.7 Data for AI Systems

ISO 42001 Control	Framework Mapping	Notes
A.7.2 Operation and monitoring	AI.8 Judge, AI.11 Logging & Monitoring	✅ Covered
A.7.3 Human oversight	AI.9 Human Oversight (HITL)	✅ Covered
A.7.4 Explainability	AI.3.2 Documentation, Judge reasoning capture	✅ Covered
A.7.5 Bias management	AI.8 Judge (bias evaluation), AI.9 HITL	✅ Covered

A.8 Information for Interested Parties

ISO 42001 Control	Framework Mapping	Notes
A.8.2 Performance monitoring	AI.8 Judge, AI.11 Monitoring	✅ Covered
A.8.3 Performance evaluation	Judge metrics, calibration	✅ Covered

A.9 Use of AI Systems

ISO 42001 Control	Framework Mapping	Notes
A.9.2 Change management	AI.4.4 Change Management (added)	✅ Covered
A.9.3 Incident management	AI.12 Incident Response	✅ Covered
A.9.4 Problem management	AI.12, Feedback loops	✅ Covered

A.10 Third-Party and Customer Relationships

ISO 42001 Control	Framework Mapping	Notes
A.10.2 Corrective action	Clause 10 processes, feedback loops	✅ Covered
A.10.3 Continual improvement	PDCA cycle, Judge→Guardrail feedback	✅ Covered

Agentic AI Mapping (Framework Extension)

ISO 42001 does not explicitly address agentic AI. The framework extends coverage:

Agentic Concern	ISO 42001 Nearest	Framework Control
Plan approval	A.7.3 Human oversight	AG.1.3 Plan Approval
Action constraints	A.4.3 Risk treatment	AG.2.1-2.4 Execution Controls
Circuit breakers	A.8.2 Monitoring	AG.2.2 Circuit Breakers
Trajectory evaluation	A.8.3 Performance	AG.3.2 Trajectory Evaluation
Multi-agent governance	A.3.2 Inventory	AG.4 Multi-Agent Controls

Cross-Reference: Framework to ISO 42001

Framework Control	ISO 42001 Clause	ISO 42001 Annex A
AI.1 Governance	5 Leadership	A.2
AI.2 Risk Management	6 Planning	A.4
AI.3 Inventory & Documentation	7.5 Documented info	A.3.2
AI.4 Development Security	8.1 Operational planning	A.6
AI.5 Data Governance	8.1	A.5
AI.6 Model Security	8.1	A.6.2
AI.7 Guardrails	8.1	A.4.3, A.7.2
AI.8 Judge	9.1 Monitoring	A.8
AI.9 Human Oversight	8.1	A.7.3
AI.10 Agentic Controls	- (extension)	-
AI.11 Logging & Monitoring	9.1	A.7.2, A.8.2
AI.12 Incident Response	10.2 Nonconformity	A.9.3
AI.13 Supplier Management	8.1 External provision	A.3.4
AI.14 Security Awareness	7.2, 7.3 Competence	A.2.4
AI.15 Business Continuity	8.1	-
AI.16 Intellectual Property	7.5	-

Summary

Key Success Factors

1. Executive sponsorship - AI governance needs visible leadership support
2. Clear accountability - Named owners for AI systems and governance
3. Proportionate controls - Match controls to risk tier
4. Platform enablement - Technical controls make governance practical
5. Feedback loops - Continuous improvement from Judge findings and HITL
6. Metrics-driven - Measure what matters, report to leadership

The Operating Model in One Page

References

• ISO/IEC 42001:2023 Artificial Intelligence Management System
• ISO 42001 Annex A Infrastructure Mapping
• Controls: Guardrails, Judge, and Human Oversight
• Risk Tiers and Control Selection