Skip to content

NIST AI Risk Management Framework Crosswalk

This document maps the FSI Agent Governance Framework controls to the NIST AI Risk Management Framework (AI RMF 1.0). Financial services stakeholders have expressed support for voluntary NIST AI RMF adoption, and Treasury has committed to working with financial regulators to clarify its applicability and develop financial sector-specific guidance (December 2024 AI in Financial Services report).

Overview

The NIST AI RMF provides a structured approach to managing AI risks through four core functions:

Function Purpose FSI Framework Alignment
GOVERN Establish AI governance structures and policies Pillar 2 (Management), Framework Layer
MAP Identify and categorize AI systems and risks Pillar 3 (Reporting), Zone Classification
MEASURE Assess and analyze AI risks Pillar 1 (Security), Pillar 2 (Testing)
MANAGE Prioritize and treat AI risks All Pillars, Playbooks Layer

GOVERN Function Mapping

Cultivate and implement a culture of risk management within organizations designing, developing, deploying, evaluating, or acquiring AI systems.

GOVERN 1: Policies, processes, procedures, and practices across the organization

NIST AI RMF Category FSI Controls Coverage
GOVERN 1.1: Legal/regulatory requirements identified 2.12, 3.3 Full
GOVERN 1.2: Processes to assess compliance 2.6, 2.12 Full
GOVERN 1.3: Processes for oversight of third-party AI 2.7 Full
GOVERN 1.4: Risk management integrated with enterprise Framework Layer, 2.6 Full
GOVERN 1.5: Ongoing monitoring processes established 3.2, 3.4 Full
GOVERN 1.6: Mechanisms for inventory of AI systems 3.1 Full
GOVERN 1.7: Processes for decommissioning AI systems Agent Decommissioning Playbook Full

GOVERN 2: Accountability structures established

NIST AI RMF Category FSI Controls Coverage
GOVERN 2.1: Roles and responsibilities defined Operating Model, RACI Templates Full
GOVERN 2.2: Personnel trained in AI risk 2.14 Full
GOVERN 2.3: Executive leadership oversight Executive Summary, Governance Cadence Full

GOVERN 3: Workforce diversity and AI literacy

NIST AI RMF Category FSI Controls Coverage
GOVERN 3.1: Decision-making informed by diverse team 2.12, RACI Partial
GOVERN 3.2: AI literacy appropriate to roles 2.14 Full

GOVERN 4: Organizational culture of AI risk awareness

NIST AI RMF Category FSI Controls Coverage
GOVERN 4.1: Risk culture embedded in organization Framework Layer, Adoption Roadmap Full
GOVERN 4.2: Feedback mechanisms for AI concerns 3.10, 3.4 Full
GOVERN 4.3: Risk management activities documented 2.13 Full
NIST AI RMF Category FSI Controls Coverage
GOVERN 5.1: Legal compliance integrated Regulatory Framework, Regulatory Mappings Full
GOVERN 5.2: Ongoing monitoring of legal landscape Governance Cadence, Compliance Reporting Full

GOVERN 6: External stakeholder risk management

NIST AI RMF Category FSI Controls Coverage
GOVERN 6.1: Policies for AI-related external risks 2.7, 1.4 Full
GOVERN 6.2: Processes for third-party due diligence 2.7 Full

MAP Function Mapping

Establish context to frame risks related to an AI system.

MAP 1: Context established and documented

NIST AI RMF Category FSI Controls Coverage
MAP 1.1: Intended purpose documented 3.1, Agent Inventory Entry Full
MAP 1.2: Interdependencies identified 2.17, [Related Controls sections] Full
MAP 1.3: Technical specifications documented Agent metadata, Control documentation Full
MAP 1.4: Deployment context documented Zones and Tiers, Per-Agent Data Policy Full
MAP 1.5: Expected benefits and costs articulated 3.5, Business justification Full
MAP 1.6: Scientific and technical limits known 3.10, Model documentation Full

MAP 2: AI system categorized

NIST AI RMF Category FSI Controls Coverage
MAP 2.1: AI system risk categorized Zones and Tiers, Zone 1/2/3 classification Full
MAP 2.2: Risk tolerance established Zone requirements, 2.6 Full
MAP 2.3: Specific risks identified AI Risk Assessment Template Full

MAP 3: AI capabilities, targeted usage, and potential misuse documented

NIST AI RMF Category FSI Controls Coverage
MAP 3.1: Expected and potential uses documented Agent Inventory Entry, Per-Agent Data Policy Full
MAP 3.2: Potential misuse identified 1.8, 2.20 Full
MAP 3.3: Trustworthiness requirements identified Zone requirements, 2.6 Full

MAP 4: Risks associated with third-party entities identified

NIST AI RMF Category FSI Controls Coverage
MAP 4.1: Third-party components inventoried 2.7, Connector inventory Full
MAP 4.2: Third-party risks assessed 2.7 Full

MAP 5: Impacts characterized

NIST AI RMF Category FSI Controls Coverage
MAP 5.1: Benefits and harms to individuals characterized 2.11, 2.19 Full
MAP 5.2: Environmental impact considered Out of scope (not primary FSI concern) N/A

MEASURE Function Mapping

Employ quantitative, qualitative, or mixed-method tools, techniques, and methodologies to analyze, assess, benchmark, and monitor AI risk.

MEASURE 1: Appropriate methods and metrics identified

NIST AI RMF Category FSI Controls Coverage
MEASURE 1.1: Approaches for measurement identified 2.5, 2.6 Full
MEASURE 1.2: Metrics appropriate to risk 3.2, 3.10 Full
MEASURE 1.3: Internal/external evaluations conducted 2.5, 2.20 Full

MEASURE 2: AI systems evaluated for trustworthiness

NIST AI RMF Category FSI Controls Coverage
MEASURE 2.1: Tested against trustworthiness characteristics 2.5, 2.6 Full
MEASURE 2.2: Safety evaluated 1.8, 2.20 Full
MEASURE 2.3: Security and resilience evaluated 1.8, Pillar 1 Security Full
MEASURE 2.4: Explainability evaluated Zone 1 Explainability Partial
MEASURE 2.5: Privacy evaluated 1.5, 1.6, 1.14 Full
MEASURE 2.6: Fairness evaluated 2.11 Full
MEASURE 2.7: Human-AI interaction evaluated Human-in-the-Loop, 2.12 Full
MEASURE 2.8: Transparency claims verified 2.19, 2.21 Full
MEASURE 2.9: Environmental impact evaluated Out of scope (not primary FSI concern) N/A
MEASURE 2.10: Validity and reliability evaluated 2.5, 2.6 Full
MEASURE 2.11: Third-party evaluated 2.7 Full

MEASURE 3: Mechanisms for tracking identified AI risks

NIST AI RMF Category FSI Controls Coverage
MEASURE 3.1: Risks tracked over time 3.4, AI Risk Assessment Template Full
MEASURE 3.2: Feedback mechanisms implemented 3.10 Full
MEASURE 3.3: Risk assessment updates documented 3.3, Quarterly reviews Full

MEASURE 4: Measurement feedback incorporated

NIST AI RMF Category FSI Controls Coverage
MEASURE 4.1: Feedback integrated into system improvements 3.10, 2.3 Full
MEASURE 4.2: Measurement approaches reviewed Governance Cadence, Quarterly assessments Full

MANAGE Function Mapping

Allocate risk resources to mapped and measured risks on a regular basis and as defined by the GOVERN function.

MANAGE 1: AI risks based on assessments and priorities treated

NIST AI RMF Category FSI Controls Coverage
MANAGE 1.1: Prioritized risks addressed Zone classification, AI Risk Assessment Full
MANAGE 1.2: Treatment plans implemented Remediation Tracking, Control implementation Full
MANAGE 1.3: Risk tolerance thresholds acted upon Zone escalation, 3.4 Full
MANAGE 1.4: Negative impacts minimized Runtime protection, DLP, Access controls Full

MANAGE 2: Strategies to maximize benefits and minimize harms

NIST AI RMF Category FSI Controls Coverage
MANAGE 2.1: Resources allocated to manage risks Governance Cadence, RACI assignments Full
MANAGE 2.2: Benefit/harm considerations in deployment Zone classification, Per-Agent Data Policy Full
MANAGE 2.3: Decisions documented 2.13, Decision logs Full
MANAGE 2.4: Mechanisms for appeal or recourse Human-in-the-Loop, Escalation matrix Full

MANAGE 3: AI risks managed throughout lifecycle

NIST AI RMF Category FSI Controls Coverage
MANAGE 3.1: Pre-deployment risk management 2.5, Agent Promotion Checklist Full
MANAGE 3.2: Post-deployment monitoring 3.2, 3.10 Full

MANAGE 4: Risk treatments monitored and response actions taken

NIST AI RMF Category FSI Controls Coverage
MANAGE 4.1: Post-deployment monitoring implemented 3.2, 1.7 Full
MANAGE 4.2: Mechanisms to capture emergent risks 3.4, 3.10 Full
MANAGE 4.3: Incident response mechanisms in place AI Incident Response Playbook Full

Coverage Summary

Methodology

This crosswalk maps FSI Agent Governance Framework controls to the NIST AI RMF 1.0 subcategories. The table below reflects the subcategories explicitly addressed in this document. The official NIST AI RMF 1.0 contains 72 subcategories; this crosswalk addresses 67 subcategories that are most relevant to Microsoft 365 AI agent governance in financial services.

Subcategory Coverage

NIST AI RMF Function Subcategories Addressed Full Coverage Partial Coverage Not Applicable
GOVERN 19 18 1 0
MAP 16 15 0 1
MEASURE 19 17 1 1
MANAGE 13 13 0 0
TOTAL 67 63 2 2

Coverage Calculation

  • Subcategories addressed: 67 of 72 NIST AI RMF subcategories (93%)
  • Full coverage: 63 of 67 addressed subcategories (94%)
  • Partial coverage: 2 subcategories (GOVERN 3.1, MEASURE 2.4)
  • Not applicable to FSI agent governance: 2 subcategories (MAP 5.2, MEASURE 2.9 - environmental impact)
  • Not explicitly addressed: 5 subcategories (MEASURE 2.12, 2.13, and others focused on large-scale AI system development not applicable to Microsoft 365 agents)

Effective Coverage of Applicable Subcategories: 97% (63 full + 2 partial of 65 applicable)

Alignment Gaps and Remediation

Partial Coverage Areas

Category Gap Remediation
GOVERN 3.1 (Diverse team) Framework does not mandate diversity requirements Organizational hiring/team practices; out of technical scope
MEASURE 2.4 (Explainability) Basic explainability guidance only Enhance Zone 1 explainability playbook for enterprise needs

Not Applicable Areas

Category Rationale
MAP 5.2 (Environmental) Environmental impact not primary FSI regulatory concern
MEASURE 2.9 (Environmental) Environmental impact not primary FSI regulatory concern

Subcategories Not Explicitly Addressed

The following NIST AI RMF subcategories are not explicitly addressed in this crosswalk because they pertain to large-scale AI system development, training, and data curation rather than governance of pre-built Microsoft 365 agents:

Category Description Rationale
MEASURE 2.12 Environmental impact quantified Not applicable - Microsoft manages infrastructure environmental impact
MEASURE 2.13 Effectiveness of risk mitigations verified Covered implicitly through testing controls (2.5, 2.6) but not separately tracked
MAP 3.4 Assumptions about data validated Microsoft manages Copilot training data; organization controls grounding data via Pillar 4
MAP 3.5 Data provenance documented Microsoft manages model data provenance; framework covers grounding data governance
GOVERN 4.4 Documentation of AI system development Not applicable - framework governs deployed agents, not AI development

Organizations developing custom AI models beyond Microsoft 365 agents should consult the full NIST AI RMF for these additional requirements.

Using This Crosswalk

For Compliance Officers

  1. Reference this document when responding to regulator inquiries about AI risk management frameworks
  2. Map specific control implementations to NIST AI RMF categories for examination documentation
  3. Use coverage summary to identify areas requiring additional organizational controls

For AI Governance Leads

  1. Use NIST AI RMF categories as checklist for new agent deployments
  2. Reference specific controls when designing governance procedures
  3. Identify gaps in current implementations vs. NIST expectations

For External Auditors

  1. Framework demonstrates substantive alignment with NIST AI RMF (93% of subcategories addressed, 97% effective coverage of applicable areas)
  2. Partial coverage areas and N/A designations are documented with rationale
  3. Control documentation provides implementation evidence
  4. Five NIST subcategories not explicitly addressed relate to large-scale AI development (not applicable to Microsoft 365 agent governance)

Complementary Framework: ISO/IEC 42001

ISO/IEC 42001 AI Management System Standard

Organizations seeking certification-based AI governance may consider ISO/IEC 42001:2023, the international standard for AI management systems. ISO/IEC 42001 is complementary to NIST AI RMF, not an alternative—the frameworks serve different purposes and can be implemented together for comprehensive AI governance. ISO/IEC 42001 is particularly valuable for organizations that:

  • Require third-party certification of AI governance
  • Operate in jurisdictions where ISO certification is expected
  • Already maintain ISO 27001 or other ISO management systems

How they complement each other:

Aspect NIST AI RMF ISO/IEC 42001
Nature Framework (voluntary) Standard (certifiable)
Structure 4 functions, 72 subcategories PDCA management system
Certification Not certifiable Third-party certifiable
Geography U.S.-focused International
Primary Role Risk identification and measurement Formal governance structures

Recommended approach: Begin with NIST AI RMF risk assessments (lower implementation barrier), formalize findings into ISO 42001 AIMS policies and controls, then pursue ISO 42001 certification. Use NIST for continuous risk monitoring between annual ISO 42001 surveillance audits.

This crosswalk can be adapted for ISO/IEC 42001 alignment by mapping FSI controls to ISO 42001 clauses (Annex A controls). Contact your compliance officer for guidance on implementing both frameworks together.

References

FSI Agent Governance Framework v1.2 | Updated: January 2026 | NIST AI RMF Crosswalk Last Verified: January 19, 2026