NIST AI Risk Management Framework

The Foundation for Trustworthy AI Systems

Version: AI RMF 1.0 (Jan 2023) GenAI Profile: July 2024 Status: Voluntary

Overview

The NIST Artificial Intelligence Risk Management Framework (AI RMF 1.0) is the foundational U.S. standard for managing risks associated with AI systems throughout their lifecycle. Published in January 2023, the framework provides a flexible, voluntary approach that organizations can adapt to their specific contexts, use cases, and risk tolerances.

Why NIST AI RMF Matters

The AI RMF has become the de facto standard for AI governance in the United States—cited by SEC, CFPB, federal agencies, and state regulators. Referenced by major consulting frameworks and corporate AI policies. Crosswalks available to EU AI Act, ISO 42001, and Singapore AI Verify.

Core Functions

The AI RMF organizes risk management into four interconnected functions:

GOVERN

Cultivate a culture of risk management. Establish policies, roles, accountability.

MAP

Understand context and identify AI system characteristics. Categorize by risk.

MEASURE

Analyze risks using quantitative and qualitative methods. Assess trustworthiness.

MANAGE

Address identified risks. Implement mitigation, monitoring, incident response.

Seven Characteristics of Trustworthy AI

CharacteristicDescription
Valid and ReliableSystem performs as intended under expected conditions
SafeDoes not endanger human life, health, property, or environment
Secure and ResilientMaintains confidentiality, integrity, and availability
Accountable and TransparentClear accountability and openness about operation
Explainable and InterpretableOutputs can be understood by stakeholders
Privacy-EnhancedProtects personal information appropriately
Fair with Bias ManagedTreats individuals and groups equitably

Generative AI Profile (AI 600-1)

Released July 2024, the GenAI Profile addresses risks unique to large language models and generative systems:

GAI RiskDescription
ConfabulationGeneration of false or misleading information
Dangerous ContentCBRN, weapons, or harmful instruction generation
Data PrivacyTraining data memorization and exposure
Information IntegrityMisinformation and manipulation
Information SecurityModel extraction and adversarial attacks
Intellectual PropertyCopyright and licensing concerns
Value ChainThird-party model and component risks

Industry-Specific Considerations

Financial Services

Insurance

Utilities

Related Frameworks

Implement NIST AI RMF?

KAiM helps mid-market organizations adopt the AI RMF with industry-specific integration.