Actions

Difference between revisions of "NIST AI Risk Management Framework"

(Created page with "'''NIST AI Risk Management Framework''' ('''NIST AI RMF''') is a voluntary framework published by the United States National Institute of Standards and Technology (NIST) t...")
 
m
 
Line 1: Line 1:
−
'''NIST AI Risk Management Framework''' ('''NIST AI RMF''') is a voluntary framework published by the United States [[National Institute of Standards and Technology]] (NIST) to help organizations identify, assess, prioritize, and manage risks associated with [[Artificial Intelligence (AI)|artificial intelligence]] (AI) systems. Published as ''Artificial Intelligence Risk Management Framework (AI RMF 1.0)'', NIST AI 100-1, in January 2023, it provides a common structure for incorporating considerations of trustworthiness into the design, development, deployment, use, and evaluation of AI systems.<ref name="rmf">National Institute of Standards and Technology, ''Artificial Intelligence Risk Management Framework (AI RMF 1.0)'', NIST AI 100-1, January 2023, [https://doi.org/10.6028/NIST.AI.100-1 DOI:10.6028/NIST.AI.100-1].</ref>
+
'''NIST AI Risk Management Framework''' ('''NIST AI RMF''') is a voluntary framework published by the United States [[National Institute of Standards and Technology (NIST)]] to help organizations identify, assess, prioritize, and manage risks associated with [[Artificial Intelligence (AI)|artificial intelligence]] (AI) systems. Published as ''Artificial Intelligence Risk Management Framework (AI RMF 1.0)'', NIST AI 100-1, in January 2023, it provides a common structure for incorporating considerations of trustworthiness into the design, development, deployment, use, and evaluation of AI systems.<ref name="rmf">National Institute of Standards and Technology, ''Artificial Intelligence Risk Management Framework (AI RMF 1.0)'', NIST AI 100-1, January 2023, [https://doi.org/10.6028/NIST.AI.100-1 DOI:10.6028/NIST.AI.100-1].</ref>
  
 
The framework organizes risk-management outcomes into four interconnected functions: '''Govern''', '''Map''', '''Measure''', and '''Manage'''. Govern provides a cross-cutting organizational foundation; the other functions help establish the context of an AI system, assess its risks, and determine responses. These functions are intended to operate iteratively across the [[AI system lifecycle]], rather than as a one-time certification procedure or mandatory sequence of steps.<ref name="rmf" />
 
The framework organizes risk-management outcomes into four interconnected functions: '''Govern''', '''Map''', '''Measure''', and '''Manage'''. Govern provides a cross-cutting organizational foundation; the other functions help establish the context of an AI system, assess its risks, and determine responses. These functions are intended to operate iteratively across the [[AI system lifecycle]], rather than as a one-time certification procedure or mandatory sequence of steps.<ref name="rmf" />

Latest revision as of 17:59, 18 November 2026

NIST AI Risk Management Framework (NIST AI RMF) is a voluntary framework published by the United States National Institute of Standards and Technology (NIST) to help organizations identify, assess, prioritize, and manage risks associated with artificial intelligence (AI) systems. Published as Artificial Intelligence Risk Management Framework (AI RMF 1.0), NIST AI 100-1, in January 2023, it provides a common structure for incorporating considerations of trustworthiness into the design, development, deployment, use, and evaluation of AI systems.[1]

The framework organizes risk-management outcomes into four interconnected functions: Govern, Map, Measure, and Manage. Govern provides a cross-cutting organizational foundation; the other functions help establish the context of an AI system, assess its risks, and determine responses. These functions are intended to operate iteratively across the AI system lifecycle, rather than as a one-time certification procedure or mandatory sequence of steps.[1]

NIST AI RMF is a specific published framework within the broader discipline of AI risk management. It can support AI governance and responsible AI programs, but using it does not by itself establish legal compliance, prove that an AI system is safe, or confer NIST certification.

Purpose and scope

The AI RMF was developed for organizations and individuals that design, develop, deploy, procure, operate, or evaluate AI systems. Its purpose is to help them understand risks to people, organizations, and society, including harms arising from both technical system behavior and the conditions in which a system is used. NIST describes the framework as voluntary, rights-preserving, non-sector-specific, and adaptable to different applications and organizational capacities.[1]

AI risks may be difficult to anticipate because performance can depend on training data, deployment conditions, interactions with people, system updates, and changing operating environments. The framework therefore considers AI to be a socio-technical system and places attention on affected individuals and groups alongside organizational objectives. Its guidance applies to systems developed internally and systems acquired from third parties, although organizations must adapt the controls to their own roles and access to information.

Structure of the framework

NIST AI RMF 1.0 has two principal parts. Foundational Information explains AI risk, the intended audience, the AI lifecycle, and characteristics of trustworthy AI. Core and Profiles describes four risk-management functions and a method for adapting them to particular use cases or organizations.[1]

The four core functions

Function Role in the framework Typical questions and evidence
Govern Establish an organizational culture, policies, accountability, roles, and oversight mechanisms for AI risk management. Who is accountable? What risk tolerances, policies, escalation paths, training, and review records apply?
Map Establish the AI system's purpose, context, stakeholders, intended uses, impacts, and plausible sources of risk. Where and by whom is the system used? Who could be harmed? What assumptions, dependencies, and use restrictions matter?
Measure Assess, analyze, and track identified risks and trustworthiness characteristics using appropriate methods and evidence. What testing, monitoring, impact assessment, uncertainty, and evaluation results support the risk judgment?
Manage Prioritize risks and select, implement, document, and monitor responses throughout the AI lifecycle. Which risks are accepted, reduced, avoided, or otherwise addressed? Who owns the response and how is its effectiveness reviewed?

The functions contain categories and subcategories describing desired outcomes. Their detailed statements are not a universal checklist and need not be undertaken in a prescribed order. NIST positions Govern as applying across the other three functions and the full AI lifecycle.[1]

Govern

Govern addresses the structures that allow AI risk decisions to be made consistently. It encompasses documented policies and procedures, responsibilities, legal and regulatory considerations, workforce competence, communication, risk tolerance, and arrangements for monitoring and review. The function also addresses how relevant external parties and affected stakeholders inform risk management.

In an enterprise, governance may involve a business owner for the use case, technical and data teams, information security, legal and privacy functions, procurement, and independent assurance. The particular allocation of authority is an organizational decision; the AI RMF does not prescribe a single committee or reporting line.

Map

Map develops the contextual understanding required for credible risk assessment. It considers intended purpose, deployment environment, users, affected people, data and model dependencies, potential benefits, foreseeable misuse, and circumstances in which the system should not be used. An AI application that is appropriate for drafting internal documents may require a different evaluation when used to make consequential decisions about people.

The output can include a use-case description, stakeholder and impact analysis, system boundaries, assumptions, and a preliminary AI risk assessment. Mapping remains relevant when a system's role or operating context changes.

Measure

Measure addresses the evaluation of risks using qualitative and quantitative methods suited to the application. Evidence can include validity and reliability testing, performance under relevant conditions, security assessments, privacy and fairness analysis, human-factors evaluation, incident trends, and post-deployment monitoring. Uncertainty, measurement limitations, and the representativeness of test data affect what conclusions can be drawn.

A benchmark score or a successful technical test is not a complete measure of trustworthiness. The significance of a result depends on how the AI system will be used, the potential severity of error, and the people affected.

Manage

Manage connects risk findings to organizational action. Its outcomes concern prioritization, response selection, resource allocation, incident and change management, and continuing review of risks and benefits. Responses may involve changing the use case, improving system controls, adding human oversight, limiting deployment, selecting an alternative, or discontinuing the system.

This function makes the link between evaluation evidence and operational decisions explicit. Documentation may record the residual risk, assigned owner, approval or escalation decision, conditions for continued use, and the events that require reassessment.

Trustworthiness characteristics

NIST AI RMF 1.0 identifies several interrelated characteristics of trustworthy AI systems. They include:

  • Valid and reliable: the system performs as intended within defined conditions and with understood limitations.
  • Safe: risks of physical or other harm are considered and addressed for the use context.
  • Secure and resilient: the system can withstand, respond to, and recover from relevant threats or adverse events.
  • Accountable and transparent: responsibility and information about system operation and decisions are appropriate to the context.
  • Explainable and interpretable: relevant people can obtain meaningful explanations or understanding of system outputs and behavior.
  • Privacy-enhanced: privacy risks across data and system activities are managed.
  • Fair, with harmful bias managed: harmful bias and disparate impacts are assessed and addressed in relevant contexts.

These characteristics interact and can require tradeoffs; they are not a single composite score or a guarantee of zero risk. The appropriate evidence and degree of assurance depend on the purpose and consequences of the system.[1]

AI RMF Profiles

An AI RMF Profile is a tailoring mechanism that connects framework outcomes to a particular organizational setting, sector, technology, or application. A profile helps organizations select relevant outcomes and express priorities rather than treating every category as equally significant in every context. NIST distinguishes the general Core from Profiles that make its application more specific.[1]

NIST published the Generative Artificial Intelligence Profile (NIST AI 600-1) in July 2024 as a companion to AI RMF 1.0. It addresses risks and suggested actions relevant to generative AI, including risks associated with generated content and other generative-system characteristics.[2]

In April 2026, NIST announced a concept note for a profile concerning trustworthy AI in critical infrastructure. A concept note is a development-stage resource, not an indication that a finalized profile has been issued.[3]

Supporting resources

NIST maintains several resources to help readers apply and interpret the AI RMF:

  • AI RMF Playbook: suggested practices aligned with the Core's categories and subcategories; NIST specifies that it is neither a complete checklist nor a mandatory set of steps.[4]
  • Profiles: adaptations of framework outcomes for defined contexts, technologies, or sectors.
  • Crosswalks: mappings between AI RMF elements and other standards or guidance. A listed third-party crosswalk does not necessarily represent NIST endorsement or complete equivalence between resources.[5]
  • AI Resource Center: a NIST-maintained entry point for framework information, supporting materials, and examples of use.

These resources supplement the AI RMF rather than replacing the primary publication or independently establishing compliance.

Organizational use and evidence

Organizations can use the Core to connect an AI inventory with explicit risk ownership and measurable review practices. For example, a company considering an AI-enabled customer-support assistant can use Map to describe intended use, foreseeable harmful outputs, user populations, and vendor dependencies. Measure can examine test results and operational monitoring. Manage can set escalation rules, usage restrictions, and actions for serious incidents. Govern makes sure responsibilities, approvals, and periodic reviews remain clear.

Typical artifacts can include a documented AI use case, system inventory entry, risk and impact assessment, testing and validation results, monitoring records, risk-treatment decisions, incident records, and management review evidence. These are illustrative operational artifacts, not a NIST-issued mandatory document list.

An organization should also distinguish a framework mapping from evidence of functioning controls. Merely assigning a Core subcategory to a policy does not demonstrate that the policy is implemented, effective, or suitable for a particular risk.

Limitations and relationship to other standards

NIST AI RMF provides a flexible basis for managing AI risks, but it does not prescribe a universal risk appetite, fixed scoring method, specific technical control set, or outcome-based certification. Its voluntary status does not remove any obligations that may arise under applicable laws, contracts, or sector requirements.

The framework is related to, but distinct from, several other concepts:

  • AI Risk Management is the broader practice of identifying and addressing AI-related risks; NIST AI RMF is one published framework used to organize that practice.
  • AI Governance establishes organizational authority and accountability for AI decisions; the AI RMF's Govern function supplies risk-management outcomes within that larger governance activity.
  • AI Security focuses on protecting AI systems and their supporting assets from security threats; security and resilience are among the trustworthiness considerations covered by the AI RMF.
  • ISO/IEC 42001 specifies requirements for an AI management system, whereas NIST AI RMF is voluntary risk-management guidance rather than a certifiable management-system requirements standard.
  • ISO/IEC 23894 provides AI-specific risk-management guidance that organizations may consider alongside NIST AI RMF.

No single mapping among these resources should be assumed to establish complete coverage of another standard or a jurisdiction's legal requirements.

Version and current status

NIST published AI RMF 1.0 on January 26, 2023. As of November 18, 2026, NIST's official AI RMF pages continue to identify AI RMF 1.0 as the available framework and state that a revised version is in progress. NIST also indicates that its Playbook will be updated after the framework revision.[3][4] The exact applicable edition and companion-resource status should be checked against NIST's official publications when making a formal organizational adoption or compliance decision.

See also

References

  1. ↑ 1.0 1.1 1.2 1.3 1.4 1.5 1.6 National Institute of Standards and Technology, Artificial Intelligence Risk Management Framework (AI RMF 1.0), NIST AI 100-1, January 2023, DOI:10.6028/NIST.AI.100-1.
  2. ↑ National Institute of Standards and Technology, Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile, NIST AI 600-1, July 26, 2024, DOI:10.6028/NIST.AI.600-1.
  3. ↑ 3.0 3.1 National Institute of Standards and Technology, AI Risk Management Framework, official framework and revision-status page, accessed November 18, 2026.
  4. ↑ 4.0 4.1 NIST AI Resource Center, NIST AI RMF Playbook, accessed November 18, 2026.
  5. ↑ NIST AI Resource Center, Crosswalk Documents, accessed November 18, 2026.