GOVERN 1 Policies, processes and procedures
The base layer of AI governance: legal requirements, trustworthy AI policy, risk tolerance, transparent documentation, periodic review, the AI inventory and decommissioning.
NIST AI RMF
NIST AI Risk Management Framework 1.0 (NIST AI 100-1), a voluntary framework with no certification, organized into Govern, Map, Measure and Manage. The pillar covers its 19 categories and 72 subcategories, with Playbook suggestions and Generative AI Profile actions for each.
GOVERN 2 Accountability structures
Who is accountable for AI risk: documented roles and lines of communication, AI risk management training, and executive responsibility for AI risk decisions.
GOVERN 3 Workforce diversity, equity, inclusion and accessibility
Team composition for AI risk decisions and the definition of human roles in human-AI configurations and oversight. America's AI Action Plan recommends removing references to diversity, equity and inclusion in the pending revision.
GOVERN 4 Organizational risk culture
Risk culture in AI teams: a safety-first and critical thinking mindset, documenting and communicating impacts, and practices that enable testing, incident identification and information sharing.
GOVERN 5 Engagement with relevant AI actors
How organizations collect feedback from people outside the AI team and feed adjudicated feedback from relevant AI actors back into system design.
GOVERN 6 Third-party software, data and supply chain
Governance of third-party AI: policies for vendor systems, data and intellectual property risks, and contingency plans for failures in high-risk third-party AI.
MAP 1 Context
Establishing context for an AI system: intended purpose, team composition, mission and business value, risk tolerances and system requirements.
MAP 2 Categorization of the AI system
Categorizing the AI system: its tasks and methods, its knowledge limits and human use of its output, and the scientific integrity and TEVV considerations behind it.
MAP 3 Capabilities, usage, benefits and costs
Understanding what the system will do and for whom: benefits, costs of errors, targeted scope, operator proficiency and human oversight processes.
MAP 4 Risks and benefits of components
Mapping technology, legal and intellectual property risks for every component of the AI system, including third-party data, software and models, and the internal controls that cover them.
MAP 5 Impact characterization
Characterizing impacts: the likelihood and magnitude of each identified impact and the engagement practices that capture feedback on positive, negative and unanticipated impacts.
MEASURE 1 Methods and metrics
Choosing and maintaining measurement: selecting metrics for the most significant mapped risks, reviewing metric and control effectiveness, and involving independent assessors.
MEASURE 2 Evaluation of trustworthy characteristics
The largest category in the AI RMF: TEVV documentation, human subject evaluations, deployment-like testing, production monitoring and evaluation of each trustworthiness characteristic.
MEASURE 3 Tracking AI risks over time
Tracking AI risks after deployment: existing and emergent risks, risks that current methods cannot measure, and end user problem reporting and appeals.
MEASURE 4 Feedback on measurement efficacy
Checking that measurement works in context: deployment-connected measurement approaches, expert validation of trustworthiness results, and measurable performance changes.
MANAGE 1 Risk prioritization and response
Turning MAP and MEASURE results into decisions: the go or no-go determination, risk treatment priorities, responses to high-priority risks and residual risk documentation.
MANAGE 2 Maximizing benefits and minimizing negative impacts
Sustaining deployed AI: resources and non-AI alternatives, mechanisms that sustain value, responses to newly identified risks, and the ability to supersede or deactivate systems.
MANAGE 3 Third-party AI risks and benefits
Operational management of third-party AI: monitoring third-party resources and controls, and monitoring pre-trained models used in development.
MANAGE 4 Risk treatment, response, recovery and communication
After deployment: monitoring plans, continual improvement and communication of incidents and errors to relevant AI actors, including affected communities.
NIST AI RMF and AI laws
Guides to the US state laws that name the NIST AI RMF or its Generative AI Profile, what each one requires or rewards, what changed when Colorado replaced its AI Act, and how the planned NIST revision affects laws that cite the latest version.
NIST AI RMF implementation
Guides for putting the NIST AI RMF into practice: where to start, what drives cost and effort, what to outsource, how to use the Playbook, what documentation to keep, and how organizations that buy AI rather than build it apply the framework.
NIST AI RMF generative AI risks
Guides to the generative AI risks and primary considerations in NIST AI 600-1 that matter most to organizations running or buying generative AI: what each risk covers, which suggested actions address it and what evidence a program shows for it.
NIST AI RMF and AI security
Guides to the NIST publications that turn the AI RMF's secure and resilient characteristic into security work: the adversarial machine learning taxonomy, prompt injection, supply chain and agent guidance, SP 800-53 overlays, the Cyber AI Profile, secure AI development and NIST's open-source test platform.
NIST AI RMF by industry
Guides to applying the NIST AI RMF where requirements differ by sector: the Treasury financial services adaptation, critical infrastructure, hiring and HR technology, state and local government, and AI vendors answering enterprise customers.
NIST AI RMF guides
Questions organizations ask when a customer, contract or regulator brings up the NIST AI RMF: how to assess against it without a certificate, which profiles exist, what the Generative AI Profile adds, and how it compares with ISO/IEC 42001.
