Skip to content

NIST AI RMF from Govern to Manage

The NIST AI Risk Management Framework is voluntary U.S. guidance: four functions organize the work, and seven characteristics define trustworthy AI. Modulos maps it to controls on each AI system and assesses them on the schedule you set.

Framework
AI RMF 1.0 (NIST AI 100-1), January 2023
Generative AI Profile
NIST AI 600-1, July 2024
Core
Govern, Map, Measure and Manage, with 72 subcategories in 19 categories
Status
Voluntary U.S. guidance, and NIST issues no certification

A voluntary framework that U.S. buyers expect

The framework is voluntary, but some U.S. laws, agency guidance and procurement contexts reference or accept it as the reference for AI risk practice.

It also describes a loop. Map, Measure and Manage run on each AI system and feed back into Govern, so a risk file written once stops matching the system after the next retraining or the next new use case.

For the announced US terminology, our super intelligence governance guide explains how NIST AI RMF relates to existing laws, standards and enterprise governance.

Govern, Map, Measure and Manage

The framework organizes AI risk work in four core functions, with 19 categories and 72 subcategories between them. Govern is cross-cutting and applies to the whole organization, while Map, Measure and Manage run for each AI system.

Govern

6 categories
19 subcategories

Cross-cutting

Govern cultivates a culture of risk management and establishes the policies, accountability and oversight that make the other three functions repeatable. It is cross-cutting: it sits above Map, Measure and Manage and is the only function that spans the entire organization.

Govern in the docs

Practices

  • Policies, processes and procedures for AI risk management
  • Accountability structures: defined roles, responsibilities and lines of communication
  • Workforce competence, diversity and oversight of human-AI configurations
  • Risk culture, transparency and engagement with relevant AI actors
  • Third-party and supply-chain risk policies

Map

5 categories
18 subcategories

Map establishes the context and identifies the risks for a specific AI system. It captures intended use, stakeholders, lifecycle dimensions and the trustworthiness characteristics that apply to the system. The output is a documented system context that the rest of the framework builds on.

Map in the docs

Practices

  • Documented system context: purpose, intended use and deployment setting
  • Categorization of the AI system as a classifier, generative model, recommender or other type
  • Mapped risks and benefits across all components, including third-party data and software
  • Impacts on individuals, groups, communities and society identified and prioritized

Measure

4 categories
22 subcategories

Measure analyzes and tracks the identified risks with quantitative and qualitative methods. It puts the metrics, evaluation processes and monitoring in place that turn the Map output into evidence the organization can act on.

Measure in the docs

Practices

  • Approaches and metrics selected for the most significant AI risks
  • AI systems evaluated for the trustworthy AI characteristics, with test sets and TEVV documentation
  • Mechanisms to track existing, unanticipated and emergent risks over time
  • Feedback on how well the measurement works, gathered and integrated

Manage

4 categories
13 subcategories

Manage allocates resources to treat risks, documents residual risk, responds to incidents and feeds the findings back into Govern and Map. It closes the loop and keeps the framework running as a continuous risk-management process.

Manage in the docs

Practices

  • Risk treatment decisions: proceed, modify, defer or terminate
  • Strategies to maximize AI benefits and reduce negative impacts
  • Third-party risk monitoring and risk-control application
  • Post-deployment monitoring, incident response and recovery plans

Job

NIST AI RMF · Claims triage assistant

Control assessment · Weekly · Monday 09:00

Monday 21 Sep, 09:00
  1. GVGovern16 controlsAssessed
  2. MPMap14 controlsAssessed
  3. MSMeasure18 controlsRunning
  4. MGManage10 controlsQueued

30 of 58 control steps done in this run, for one AI system. Illustrative data.

Assess Govern, Map, Measure and Manage on a schedule

The NIST AI RMF framework in library 1.1.0 breaks the functions into 105 requirements and maps them to 80 controls, each with an owner, a status and its evidence. One job assesses every control in scope for an AI system on the schedule you choose, and the platform records the date of each run.

NIST AI RMF in Modulos, in the docs

Characteristics of trustworthy AI

Section 3 of AI RMF 1.0 defines seven characteristics. Map records which of them apply to a system, and Measure evaluates the system against them. They often trade off against one another, but managing those trade-offs is part of the risk work.

Trustworthy AI in the docs

Valid and reliable
Objective evidence that the system meets its requirements and operates as intended in its deployment context.
Safe
No endangerment of human life, health, property or the environment under defined conditions.
Secure and resilient
The system withstands and recovers from adverse events, adversarial attacks included.
Accountable and transparent
Information about design, training data, intended use and outputs is available to those affected.
Explainable and interpretable
How an output was produced, and what it means in context.
Privacy-enhanced
Personal data and contextual privacy expectations are protected by design and in operation.
Fair, with harmful bias managed
Systemic, computational and human-cognitive biases are identified and treated for the use case.

Measure and manage the system in production

Measure asks you to track risks over time, and Manage asks for post-deployment monitoring. In Modulos, runtime tests read live values from your monitoring. When a test fails, each control linked to it moves back to not executed in every framework it serves, the reason goes into the activity log, and the control owner is notified.

MCF-67 Production Data Drift MonitoringMetric test · daily 06:00
  1. Mon 31 AugExecutedEvidence attached and locked
  2. Tue 22 SepTest failedDrift above the threshold on the new channel
  3. Tue 22 SepNot executedControl owner and test assignee notified
  4. Thu 24 SepEvidence refreshedRetrained model, new baseline
  5. Fri 25 SepExecutedSet by the control owner after the 06:00 pass

Comments and LogsTest Drift metric failed. Status updated from executed to not executed in control MCF-67.

Reopened in EU AI Act, NIST AI RMF. Illustrative data.

See how compliance runs in the platform

The Generative AI Profile

NIST AI 600-1, published in July 2024, layers twelve risks specific to generative AI onto the four functions and suggests actions for each.

The profile builds on AI RMF 1.0 without replacing it. Organizations that govern generative AI start with the core framework and add the profile for the GenAI-specific risks and their suggested actions. Sectoral profiles apply the whole framework to one industry, and cross-sectoral profiles like AI 600-1 apply it to one technology class across all sectors.

  • CBRN information or capabilities
  • Confabulation
  • Dangerous, violent, or hateful content
  • Data privacy
  • Environmental impacts
  • Harmful bias and homogenization
  • Human-AI configuration
  • Information integrity
  • Information security
  • Intellectual property
  • Obscene, degrading, or abusive content
  • Value chain and component integration

The Generative AI Profile in the docs Source: NIST AI 600-1, Generative AI Profile (PDF, nist.gov)

Implement NIST AI RMF

The time it takes depends on scope and on how mature Govern already is. A small organization with a tight scope and existing governance can stand up a working program in weeks, but a large one with several AI portfolios and no prior governance typically works in quarters.

  1. Define the scope

    Decide which AI systems and which part of the organization the program covers first. A tight first scope keeps the work finite and gives you a result to build on.

  2. Establish Govern first

    Set the policies, accountability and oversight roles before you work on single AI systems. Govern makes Map, Measure and Manage repeatable.

  3. Run the loop on a pilot system

    Pick one moderate-risk AI system and work the loop end to end: map its context and risks, measure them with suitable metrics and record the treatment decisions. Use the lessons to scale.

  4. Build a profile

    A profile tailors AI RMF to your context. If you operate generative AI, start with the Generative AI Profile. In a regulated industry, a sectoral profile aligns AI RMF with the expectations of that sector.

  5. Operate a continuous loop

    Map, Measure and Manage feed back into Govern, so the framework keeps running after the first pass. Treating it as a one-time checklist is the most common failure mode.

EU AI ActCBUAE Consumer AISaudi AI Risk (SDAIA)FINMA AIMAS FEATUAE AI EthicsSingapore MGFColorado SB 26-189CCPA ADMTNYC Local Law 144ISO/IEC 42001NIST AI RMFprEN 18282EN 18286prEN 18228IEEE 7003GDPRUAE PDPLISO/IEC 27701ISO/IEC 27001NIS2DORACRAOWASP LLM Top 10OWASP Agentic Top 10Microsoft Supplier DPR
Framework library 1.1.0: 26 frameworks and 929 controls

Reuse your NIST AI RMF controls for the EU AI Act

79 of the 80 controls in the NIST AI RMF framework also map to at least one other framework in the library, and 58 of them serve the EU AI Act. If the Act also applies to you, part of that work is done once these controls have their evidence.

Controls shared with NIST AI RMF, of 80 · library 1.1.0

See all frameworks in the library

Certification and standards contributions

Kevin Schawinski at the AI Safety Institute Consortium First Plenary Meeting, December 2024
Kevin Schawinski at the AI Safety Institute Consortium First Plenary Meeting, December 2024
See all contributions
2024
Member of the NIST AI Safety Institute ConsortiumModulos is a member of the AI Safety Institute Consortium (AISIC), now under the Center for AI Standards and Innovation (CAISI).Participation does not imply NIST evaluation or endorsement of Modulos products.nist.gov/aisi
2024
ISO/IEC 42001 product conformity certificationThe Modulos AI Governance Platform holds the first product conformity certification for an AI governance platform against ISO/IEC 42001:2023, issued by the accredited body CertX (certificate 213-001/24).The ISO/IEC 42001 page
JTC 21
Experts in the committee that drafts the European AI standardsKevin Schawinski and Andrea Basso serve as experts in CEN-CENELEC JTC 21 Working Group 3, Engineering aspects, and Working Group 4, Foundational and societal aspects.
What is the NIST AI RMF?

The NIST AI Risk Management Framework 1.0 is voluntary guidance from the U.S. National Institute of Standards and Technology, published in January 2023 (NIST AI 100-1). It helps organizations identify, assess, and manage AI risks across the AI system lifecycle. It is organized around four core functions, Govern, Map, Measure, and Manage, and seven characteristics of trustworthy AI.

Is the NIST AI RMF mandatory?

No. NIST AI RMF is voluntary guidance with no force of law, but it is widely used by U.S. federal agencies, regulators, and enterprises as the de-facto reference for trustworthy AI, and some U.S. state AI laws and procurement contexts reference it explicitly, Texas HB 149 among them, where alignment supports a defense. For organizations selling into regulated U.S. buyers it functions as an expected baseline.

Does NIST AI RMF provide certification?

No. NIST does not certify organizations or products against the AI RMF. Any vendor or consultant claiming "NIST AI RMF certification" is referring to private training certifications; no NIST-issued attestation exists. The framework itself is voluntary guidance. Organizations seeking a certifiable AI management system typically use ISO/IEC 42001, which is internationally certifiable.

When was NIST AI RMF published?

AI RMF 1.0 (NIST AI 100-1) was published in January 2023. The Generative AI Profile (NIST AI 600-1), a cross-sectoral profile that applies AI RMF to generative AI specifically, followed in July 2024.

What are the four core functions of NIST AI RMF?

Govern, Map, Measure, and Manage. Govern is cross-cutting and sits above the other three: it establishes the policies, accountability, and culture that make risk management repeatable. Map establishes context and identifies risks for a specific AI system. Measure analyzes and tracks those risks using quantitative and qualitative methods. Manage allocates resources to treat them.

What are the seven characteristics of trustworthy AI?

Per NIST AI RMF 1.0 Section 3: Valid and Reliable; Safe; Secure and Resilient; Accountable and Transparent; Explainable and Interpretable; Privacy-Enhanced; and Fair, with Harmful Bias Managed. These characteristics are not independent boxes to tick. They often trade off against one another, and managing those trade-offs is part of what risk management is for.

What is the Generative AI Profile (NIST AI 600-1)?

Published in July 2024, the Generative AI Profile is a cross-sectoral profile that applies AI RMF to generative AI specifically. It maps GenAI-specific risks, including CBRN information, confabulation, data privacy, harmful bias, information integrity, and value chain risks, onto the four core functions and provides suggested actions for each.

How does NIST AI RMF compare to ISO/IEC 42001?

AI RMF is voluntary U.S. guidance; ISO/IEC 42001 is a certifiable international management system standard. They layer. Most mature programs use NIST AI RMF as the risk-management operating model inside an ISO 42001 AI management system, which is the certifiable wrapper that makes the program auditable and durable.

How does NIST AI RMF relate to the EU AI Act?

AI RMF is voluntary; the EU AI Act is binding regulation with legal obligations and penalties. Implementing AI RMF helps build the risk-management practices the Act expects of high-risk AI providers (Article 9 risk management), but it does not replace the Act’s specific compliance and conformity-assessment requirements. The two are complementary at different layers.

What is the difference between NIST AI RMF and the NIST Cybersecurity RMF?

Different frameworks. The Cybersecurity RMF (NIST SP 800-37) governs information system security risk for federal information systems. The AI RMF (NIST AI 100-1) governs AI-specific risks, including data quality, model behavior, and socio-technical harms that the Cybersecurity RMF does not address. Organizations operating both share governance processes where possible.

How do you implement NIST AI RMF?

Define scope, stand up Govern first, run Map, Measure, and Manage on a pilot AI system, build a profile (sectoral or cross-sectoral) tailored to your context, then operate it as a continuous loop. The framework is an operating model that keeps running after the first pass. Treating it as a one-time checklist is the most common failure mode.

How long does NIST AI RMF implementation take?

Implementation timelines vary widely with starting maturity and scope. Smaller organizations with a tight scope and existing governance can stand up a working AI RMF program in weeks; larger organizations with multiple AI portfolios and no prior governance typically work in quarters. The dominant variables are scope and the maturity of the Govern function.

How does Modulos help with NIST AI RMF implementation?

Modulos automates evidence collection and assessments across the four core functions; AI agents (Scout, evidence processing, and the Control Assessment Agent) reduce manual work; controls map across NIST AI RMF, the EU AI Act, and other frameworks simultaneously. The platform holds CertX product conformity certificate 213-001/24 against ISO/IEC 42001:2023, the management system standard most mature NIST AI RMF programs operate inside.

Can small organizations implement NIST AI RMF?

Yes. The framework is scalable. Smaller organizations typically scope tightly to one or two AI systems, use the AI RMF Playbook’s suggested actions as a starting point, and grow the program incrementally. The voluntary, the profile-based design lets you adapt the framework to your organization.

Walk one AI system through Govern, Map, Measure and Manage

Pick a system you plan to bring under NIST AI RMF. We set up its controls with you and show how assessments and runtime tests keep the Measure and Manage evidence current.