Appendix C

Standards crosswalk for compliance teams

The table maps the control plane’s main controls to three frameworks that compliance and audit teams report against:

ISO/IEC 42001 (its Annex A control objectives), with ISO/IEC 42005 for impact assessments.⁷²
The NIST AI Risk Management Framework (AI RMF 1.0 subcategories).⁷³
The EU AI Act’s requirements for high-risk systems, which apply from December 2, 2027 for Annex III systems.³²

The mapping is the author’s and indicative, so confirm clause-level mappings against the current texts with the compliance team.

Control-plane control	ISO/IEC 42001 (Annex A)	NIST AI RMF 1.0	EU AI Act (high-risk systems)
Accountability: executive owner, governance council, business owners and agent sponsors	A.2 Policies related to AI A.3 Internal organization	GOVERN 2.1 (roles and responsibilities) GOVERN 3.2 (roles for human-AI configurations and oversight)	Art. 26 (obligations of deployers)
Inventory and lifecycle: registry, owner, tier, recertification and retirement	A.6 AI system life cycle	GOVERN 1.6 (inventory) GOVERN 1.7 (decommissioning)	Art. 9 (risk management throughout the life cycle)
Risk tiers, with an impact assessment before promotion	A.5 Assessing impacts of AI systems (guidance in ISO/IEC 42005)	MAP 1.1 (context and intended use) MAP 5.1 (likelihood and magnitude of impacts)	Art. 9 (risk management) Art. 27 (fundamental rights impact assessment, where required)
One policy point: gateway policy, agent identity and least privilege	A.6 AI system life cycle A.9 Use of AI systems	MEASURE 2.7 (security and resilience)	Art. 15 (accuracy, robustness and cybersecurity)
Human oversight: approvals bound to actions, drills and takeover	A.9 Use of AI systems	GOVERN 3.2 MANAGE 2.4 (supersede, disengage or deactivate)	Art. 14 (human oversight) Art. 26 (competent, trained people assigned to oversight)
Evaluation and red-teaming	A.6 AI system life cycle	MEASURE 2.3 (performance in conditions like deployment) MEASURE 2.7	Art. 9 (testing) Art. 15
Monitoring, logging and audit trails	A.6 AI system life cycle	MEASURE 3.1 (tracking emergent risks) MANAGE 4.1 (post-deployment monitoring)	Art. 12 (record-keeping) Art. 26 (monitoring and keeping logs)
Kill switch and incident response	A.6 AI system life cycle A.9 Use of AI systems	MANAGE 2.4 MANAGE 4.3 (incidents tracked and communicated)	Art. 26 (suspend use and inform the provider) Art. 73 (serious incident reporting)
Trusted data and context	A.7 Data for AI systems	MAP 2.3 (data selection) MAP 4.1 (third-party data and software)	Art. 10 (data and data governance)
Third-party agents, models and MCP servers	A.10 Third-party and customer relationships	GOVERN 6.1 (third-party risks) MANAGE 3.1 (third-party resources monitored)	Art. 25 (responsibilities along the AI value chain)

Singapore’s Model AI Governance Framework for Agentic AI covers the same ground in four dimensions: assess and bound the risks up front, make humans meaningfully accountable, implement technical controls and processes, and enable end-user responsibility.⁷⁵ The Cloud Security Alliance’s AI Controls Matrix maps its control objectives to ISO/IEC 42001, the NIST AI RMF and the EU AI Act, and splits responsibilities among model providers, orchestrated service providers, application providers, AI customers and cloud providers, so an enterprise that runs its own control plane will usually read the matrix’s guidance for AI customers and for orchestrated service providers.⁷⁴