AI governanceUSA

NIST AI RMF 1.0

Turn responsible AI goals into accountable work

Keep AI risk decisions, review tasks and supporting evidence connected throughout the lifecycle.

Full title
Artificial Intelligence Risk Management Framework (AI RMF 1.0), NIST AI 100-1
Issued by
National Institute of Standards and Technology (NIST), US Department of Commerce
Published
26 January 2023. NIST has stated that a revised version is in progress, with no date announced.
Applies to
Any organisation that designs, develops, deploys, uses or evaluates AI systems, in any sector
Nature
Voluntary, non-sector-specific framework. NIST does not certify against it.
Structure
Core of 4 Functions (Govern, Map, Measure, Manage), 19 Categories and 72 Subcategories, with a companion Playbook

Overview

What the AI RMF is

The NIST AI Risk Management Framework helps organisations manage the risks that AI systems pose to individuals, organisations and society, and build AI that is more trustworthy. It was developed through an open, consensus-driven process and released on 26 January 2023 as NIST AI 100-1. It is voluntary, not specific to any sector or technology, and intended to adapt as AI and its uses change.

Part 1 of the framework explains how AI risk differs from traditional software risk, and sets out seven characteristics of trustworthy AI: valid and reliable; safe; secure and resilient; accountable and transparent; explainable and interpretable; privacy-enhanced; and fair with harmful bias managed. NIST describes validity and reliability as the base, and accountability and transparency as relating to all the others. The characteristics involve trade-offs that the organisation has to weigh in context.

Part 2 is the Core. Govern cultivates a risk management culture and applies across the whole life cycle. Map establishes the context and identifies risks. Measure analyses, assesses and tracks them. Manage prioritises and acts on them. The Functions are not a one-time sequence: they are performed iteratively. The AI RMF Playbook offers suggested actions, references and documentation for each Subcategory, and NIST’s Trustworthy and Responsible AI Resource Center hosts the framework and related material.

NIST has published profiles that apply the framework to particular uses. The Generative Artificial Intelligence Profile, NIST AI 600-1, was released in July 2024 and describes 12 risks that are unique to or made worse by generative AI, with suggested actions mapped to the AI RMF Subcategories. NIST states that AI RMF 1.0 is being revised as part of the White House AI Action Plan, and in April 2026 it released a concept note for a profile on trustworthy AI in critical infrastructure.

Who it applies to

The AI RMF is intended for all AI actors across the AI life cycle, including:

  • Organisations building AI models or AI-enabled products
  • Organisations deploying third-party AI, including generative AI services, in their operations
  • Banks, insurers and healthcare providers using AI in decisions that affect people
  • Government bodies and their technology partners procuring AI systems
  • Risk, compliance, legal and internal audit teams overseeing AI use
  • Test, evaluation, verification and validation teams

The framework is voluntary. Organisations often use it alongside ISO/IEC 42001 and, where it applies, the EU AI Act.

Structure

How the framework is organised

Part 1 frames AI risk and trustworthiness. Part 2 is the Core: four Functions, 19 Categories and 72 Subcategories. The number of Subcategories in each Category is shown in brackets.

Part 1

Characteristics of trustworthy AI

Valid and reliable
The system does what it is intended to do, accurately and robustly, over time. NIST treats this as the base for the other characteristics.
Safe
The system does not, under defined conditions, endanger human life, health, property or the environment.
Secure and resilient
The system withstands adverse events and attacks, and can degrade safely and recover.
Accountable and transparent
Information about the system and its outputs is available to those who interact with it, and accountability is clear.
Explainable and interpretable
How the system works, and the meaning of its outputs, can be understood in context.
Privacy-enhanced
Human autonomy, identity and dignity are safeguarded through privacy values and practices.
Fair, with harmful bias managed
Systemic, computational and statistical, and human-cognitive bias are identified and managed.
GOVERN

Govern (6 Categories, 19 Subcategories)

A cross-cutting Function that applies throughout AI risk management.

GOVERN 1 (7)
Policies, processes, procedures and practices for mapping, measuring and managing AI risks are in place, transparent and implemented effectively.
GOVERN 2 (3)
Accountability structures empower and train the right teams and individuals.
GOVERN 3 (2)
Workforce diversity, equity, inclusion and accessibility processes are prioritised in AI risk management.
GOVERN 4 (3)
Teams are committed to a culture that considers and communicates AI risk.
GOVERN 5 (2)
Processes for robust engagement with relevant AI actors.
GOVERN 6 (2)
Policies and procedures address risks and benefits from third-party software, data and other supply chain issues.
MAP

Map (5 Categories, 18 Subcategories)

Establishes the context in which risks are identified.

MAP 1 (6)
Context is established and understood, including intended purposes, settings, users and legal requirements.
MAP 2 (3)
The AI system is categorised, including its tasks, methods and knowledge limits.
MAP 3 (5)
AI capabilities, targeted usage, goals and expected benefits and costs are understood against benchmarks.
MAP 4 (2)
Risks and benefits are mapped for all components, including third-party software and data.
MAP 5 (2)
Impacts on individuals, groups, communities, organisations and society are characterised.
MEASURE

Measure (4 Categories, 22 Subcategories)

Uses quantitative, qualitative or mixed methods to analyse, assess, benchmark and monitor AI risk.

MEASURE 1 (3)
Appropriate methods and metrics are identified and applied.
MEASURE 2 (13)
AI systems are evaluated for each trustworthy characteristic, including in deployment conditions.
MEASURE 3 (3)
Mechanisms for tracking identified AI risks over time are in place.
MEASURE 4 (3)
Feedback about the efficacy of measurement is gathered and assessed.
MANAGE

Manage (4 Categories, 13 Subcategories)

Allocates resources to the risks that have been mapped and measured.

MANAGE 1 (4)
Risks from Map and Measure are prioritised, responded to and managed.
MANAGE 2 (4)
Strategies to maximise benefits and minimise negative impacts are planned, documented and informed by relevant AI actors.
MANAGE 3 (2)
Risks and benefits from third-party entities are managed.
MANAGE 4 (3)
Risk treatments, including response, recovery and communication plans, are documented and monitored regularly.
AI 600-1

Generative AI Profile: 12 risks

NIST AI 600-1, July 2024. Suggested actions are organised by AI RMF Function and Subcategory.

Information and content risks
CBRN information or capabilities; confabulation; dangerous, violent or hateful content; information integrity; obscene, degrading and/or abusive content.
Privacy, rights and fairness
Data privacy; harmful bias or homogenisation; intellectual property.
Security and operations
Information security; value chain and component integration; human-AI configuration; environmental impacts.

Implementation roadmap

How to implement it, step by step

A practical sequence for putting the AI RMF into operation. NIST expects the Functions to be repeated as systems and contexts change.

  1. 1

    Inventory AI use

    List the AI systems the organisation builds, buys or uses, including generative AI services, with owners and purposes.

    OutputsAI inventorySystem owners
  2. 2

    Set governance

    Approve an AI policy, define roles and accountability, set risk tolerance, and decide how third-party AI and data will be assessed.

    OutputsAI policyRoles and accountabilityRisk tolerance statement
  3. 3

    Map each system’s context

    Record intended purpose, users, setting, legal requirements, benefits and potential impacts on people and the organisation.

    OutputsContext recordImpact assessment
  4. 4

    Choose measures

    Select methods and metrics for the trustworthy characteristics that matter for each system, and plan evaluation before and after deployment.

    OutputsMeasurement planEvaluation criteria
  5. 5

    Evaluate and record

    Run the evaluations, document results and limitations, and track risks over time.

    OutputsEvaluation resultsRisk records
  6. 6

    Decide and treat

    Prioritise risks, decide whether to deploy, change or stop, and document treatments, responses and communication plans.

    OutputsDeployment decisionTreatment plan
  7. 7

    Monitor and improve

    Monitor systems in use, handle incidents and feedback, and revisit the context, measures and decisions when anything changes.

    OutputsMonitoring recordsIncident logReview notes

Audit and evidence guide

What auditors look for, and the evidence to keep

NIST does not certify AI RMF conformance. Organisations show their approach through internal review, internal audit, customer due diligence or independent assessment that uses the framework as a reference.

What reviewers look for

  • An inventory of AI systems with accountable owners
  • An approved AI policy, risk tolerance and defined roles
  • Context and impact documentation for each significant system
  • Evaluation methods and results for the relevant trustworthy characteristics
  • Assessment of third-party AI, models and data
  • Records of deployment decisions, including decisions not to proceed
  • Monitoring in production and incident handling
  • Human oversight and how outputs are reviewed
  • Generative AI risks considered where relevant, using AI 600-1

Key dates and cycles

AI RMF 1.0 released26 January 2023
Generative AI Profile (AI 600-1)July 2024
Critical infrastructure profile concept noteApril 2026
AI RMF revisionIn progress, no date
Risk tracking (MEASURE 3)Ongoing
Treatment monitoring (MANAGE 4)Regular, as defined

Evidence to keep, by area

Examples to guide your evidence library. Your auditor or regulator decides what is sufficient.

Govern

  • AI policy
  • Roles and accountability
  • Risk tolerance
  • Training records
  • Third-party AI requirements

Map

  • AI inventory
  • Intended purpose and context
  • Legal requirements
  • Impact assessments

Measure

  • Metrics and methods
  • Evaluation and test results
  • Bias and robustness testing
  • Known limitations

Manage

  • Risk prioritisation
  • Deployment decisions
  • Treatment plans
  • Incident and response records

Third parties

  • Supplier assessments
  • Model and data documentation
  • Contract terms

Oversight

  • Human review procedures
  • Monitoring reports
  • Leadership reporting

Areas that often need attention

  • No complete inventory of AI in use, especially generative AI services
  • AI policy approved but not linked to actual systems
  • Evaluations done once before launch and never repeated
  • Third-party models and data accepted without assessment
  • No record of why a system was approved for deployment
  • Impacts on people outside the organisation not considered
  • Incidents and user feedback not fed back into risk records

How Netra helps

Where each part of the work lives in Netra

Netra loads the AI RMF as requirements, tasks and policies, so AI governance work, evaluations and decisions sit on one record.

Subcategory outcomes owned

Task Engine

Outcomes arrive as tasks with owners and due dates, and recurring tasks cover re-evaluation and monitoring.

Status for each Function

Compliance Centre

Status is derived from completed tasks and evidence, never typed in, and one task can count towards ISO/IEC 42001 where the work is the same.

AI policy and procedures (GOVERN 1)

Policy Hub

A policy library generated at onboarding, with a policy score that rises as the work behind each policy is done.

AI risk register

Risk Manager

A risk register with inherent and residual heatmaps, with risks reduced as linked tasks complete.

Third-party AI and data (GOVERN 6, MANAGE 3)

Third-Party Risk

Finds vendors from what Accounts and Purchase pay, including AI services, assesses them through a vendor portal and records decisions with ratings.

AI systems inventory

Asset Management

AI systems recorded as assets and linked to the tasks that govern them.

Internal review of AI governance

Audit Centre

Run internal audits end to end, with findings and corrective actions, using evidence in a tamper-evident vault.

Drafting and evidence suggestions

Netra AI

Works from the organisation’s own records, suggests missing evidence and drafts SOPs. It is a co-pilot: it never changes status on its own.

Netra organises AI governance work and evidence. It does not test, evaluate or monitor AI models, and it does not by itself establish that an AI system is trustworthy or that the organisation conforms to the AI RMF.

Common questions

More questions about Netra

Related frameworks

View all frameworks
AI governance

ISO 42001:2023

Build an AI management system with clear ownership

AI governance

EU AI ACT 2024

Keep AI governance decisions supported by evidence

Security

NIST-SSDF

Make secure development practices easier to evidence

See NIST AI RMF 1.0 in Netra

See how Netra connects your framework requirements with tasks, evidence and audit follow-up.