Framework overlap

Does Azure Security Benchmark cover ISO/IEC 23894:2023?

You hold Azure Security Benchmark and have been told to do ISO/IEC 23894:2023. Here is how much overlaps, control by control.

24% of ISO/IEC 23894:2023 you already have

Azure Security Benchmark already covers about 24% of ISO/IEC 23894:2023, leaving 61 of 80 controls as genuinely new work.

Already covered 7 Likely covered 12 New work 61

What is genuinely new work

Nothing in Azure Security Benchmark reaches these. This is the list to scope.

23894-4.1
AI Risk Management Principles
23894-5.2
Leadership and Commitment
23894-5.3
Integration into Organizational Processes
23894-5.4.2
AI Risk Management Policy
23894-5.4.3
Roles, Authorities, Responsibilities
23894-5.4.4
Allocation of Resources
23894-5.5
Communication and Consultation
23894-6.3
AI Risk Assessment Scope and Criteria
23894-6.4.3
AI Risk Analysis
23894-6.4.4
AI Risk Evaluation
23894-6.5
AI Risk Treatment
23894-6.6
Monitoring and Review
23894-6.7
Recording and Reporting
23894-A.2
AI Objectives and Risk Sources
23894-A.3
Lifecycle Risk Considerations
23894-A.4
AI System Impact Assessment
23894-A.5
Human Oversight Controls
23894-A.6
Transparency and Explainability
23894-A.7
Data Quality and Provenance
23894-A.8
Robustness and Resilience Testing
23894-A.9
Third-Party AI Components
ISO23894-4.1
Integrated AI Risk Management
ISO23894-4.2
Structured and Comprehensive Approach
ISO23894-4.3
Customized to AI Context
ISO23894-4.4
Inclusive Stakeholder Engagement
ISO23894-4.5
Dynamic and Responsive
ISO23894-4.6
Best Available Information
ISO23894-4.7
Human and Cultural Factors
ISO23894-4.8
Continual Improvement
ISO23894-5.3
AI Risk Management Design
ISO23894-5.4
AI Risk Management Implementation
ISO23894-5.6
Framework Improvement
ISO23894-6.1
Communication and Consultation
ISO23894-6.3.2
AI Risk Analysis
ISO23894-6.4
AI Risk Treatment
ISO23894-6.5
Monitoring and Review
ISO23894-6.6
Recording and Reporting
ISO23894-A.2
Model Transparency and Explainability
ISO23894-A.3
Algorithmic Bias and Fairness
ISO23894-A.4
AI System Robustness
ISO23894-A.7
Human Oversight of AI
ISO23894-A.8
AI Accountability and Governance
5.2
Leadership and commitment
5.3
Integration
5.4
Design
5.4.2
Articulating risk management commitment
5.4.3
Assigning organizational roles, authorities, responsibilities and
5.4.4
Allocating resources
5.4.5
Establishing communication and consultation
5.7.1
Adapting
5.7.2
Continually improving
6.2
Communication and consultation
6.3.2
Defining the scope
6.3.3
External and internal context
6.3.4
Defining risk criteria
6.4.4
Risk evaluation
6.5
Risk treatment
6.5.2
Selection of risk treatment options
6.5.3
Preparing and implementing risk treatment plans
6.6
Monitoring and review
6.7
Recording and reporting
Show the 19 you already have
ISO23894-1
Scope of AI Risk Management
ISO23894-3
AI-Specific Terminology
ISO23894-6.2
Scope, Context and Criteria
ISO23894-6.3
AI Risk Assessment
ISO23894-6.3.1
AI Risk Identification
ISO23894-6.3.3
AI Risk Evaluation
ISO23894-A.5
Privacy and Data Protection in AI
23894-5.4.1
Understanding Organization and Context
23894-6.4.2
AI Risk Identification
ISO23894-5.1
Leadership and Commitment
ISO23894-5.2
AI Risk Management Integration
ISO23894-5.5
Framework Evaluation
ISO23894-A.1
Data Quality and Representativeness
ISO23894-A.6
AI System Security
5.4.1
Understanding the organization and its context
5.5
Implementation
6.4
Risk assessment
6.4.2
Risk identification
6.4.3
Risk analysis

How this is calculated

Already covered means a mapping runs from a control in Azure Security Benchmark to that control. Likely covered means no direct mapping exists but both frameworks map to the same control in a third standard. New work means neither. We keep those separate rather than adding them into one friendlier number, because blending them would present a two-hop inference as a verified fact.

Coverage is not symmetric. Run it the other way and you will get a different number; both are correct.

From 332,959 cross-framework control mappings across 723 frameworks, 531 of them verified against their source documents. It does not tell you that you are compliant: a mapped control means the two standards ask for the same thing, not that you have done it.

Try another pair ยท Today's edition