NIST AI Risk Management Framework (AI RMF) v1.0

Why does it matter?

NIST says the risks of AI systems are in many ways unique: models can be trained on data that changes over time, failures are hard to detect, and systems are socio-technical, shaped by how people use them. The AI RMF gives organizations a common structure for that work. Section 1.2.3 states a hard limit: where an AI system presents unacceptable negative risk, such as imminent significant harm or harms actually occurring, development and deployment should cease in a safe manner until the risks can be managed. Section 4 lists the benefits NIST expects, including explicit go or no-go decisions on commissioning and deployment.

Key benefits include:

• Seven trustworthiness characteristics

Section 3 names them and says that creating trustworthy AI requires balancing them for the system's context of use; valid and reliable is the necessary base.

• Shared responsibility across AI actors

Section 2 describes AI actors across the lifecycle, from developers to deployers and the people affected, and says all of them manage risk in the systems they develop, deploy or use.

• Third-party AI is in scope

GOVERN 6, MAP 4 and MANAGE 3 cover risks from third-party software, data, pre-trained models and other supply chain issues.

• Profiles for use cases

Section 6 describes use-case profiles, such as a hiring or fair housing profile, plus Current and Target temporal profiles and cross-sectoral profiles.

• Residual risk is documented

Section 1.2.3 says to document residual risk after treatment, so end users and affected communities know the remaining impacts.

How does it work?

GOVERN has six categories and 19 subcategories. It covers policies and procedures, accountability structures, workforce diversity, equity, inclusion and accessibility, a risk culture, engagement with relevant AI actors, and third-party and supply chain policies. Section 5.1 calls it a cross-cutting function infused throughout the others, and senior leadership sets the tone for risk management. GOVERN is continual and intrinsic for as long as a system is in use.

MAP (five categories, 18 subcategories) establishes context, categorizes the AI system, compares capabilities and expected benefits and costs with benchmarks, maps risks and benefits for all components including third-party software and data, and characterizes impacts. MEASURE (four categories, 22 subcategories) selects methods and metrics, evaluates trustworthy characteristics, tracks risks over time and gathers feedback on measurement. MANAGE (four categories, 13 subcategories) prioritizes and responds to risks, plans strategies to maximize benefits, manages third-party risk and documents response, recovery and communication plans.

Section 5 says the actions are not a checklist and not necessarily ordered. After putting GOVERN outcomes in place, most users start with MAP and continue to MEASURE or MANAGE, iterating and cross-referencing between functions. The online AI RMF Playbook suggests tactical actions for each outcome, and users can create tailored guidance from it.

Key Elements

• GOVERN (5.1)

Six categories on policies, accountability, workforce, culture, engagement with AI actors, and third-party software and data.

• MAP (5.2)

Five categories that establish context, categorize the system, compare benefits and costs, map component risks and characterize impacts.

• MEASURE (5.3)

Four categories for methods and metrics, evaluation against trustworthy characteristics, tracking risks over time and feedback.

• MANAGE (5.4)

Four categories for prioritizing risks, planning benefit and impact strategies, managing third-party risk, and documenting treatments and communication plans.

• Trustworthy AI characteristics (section 3)

Valid and reliable, safe, secure and resilient, accountable and transparent, explainable and interpretable, privacy-enhanced, and fair with harmful bias managed.

• Profiles (section 6)

Use-case, temporal (Current and Target) and cross-sectoral profiles, with no prescribed template.

What does it cover?

The framework is for AI actors, the organizations and individuals who design, develop, deploy, evaluate or use AI systems. The executive summary defines an AI system as an engineered or machine-based system that generates outputs such as predictions, recommendations or decisions influencing real or virtual environments. Section 2 organizes the lifecycle by four dimensions, Application Context, Data and Input, AI Model, and Task and Output, with a People and Planet dimension for those affected, and it highlights testing, evaluation, verification and validation across them. NIST describes it as voluntary, rights-preserving and suitable for organizations of all sizes.

What are its objectives?

The executive summary states the goal: to offer a resource to organizations designing, developing, deploying or using AI systems.

Help organizations manage the many risks of AI.

Promote trustworthy and responsible development and use of AI systems.

Improve the ability to build the seven trustworthiness characteristics into design, development, use and evaluation.

Integrate AI risks with other enterprise risks such as cybersecurity and privacy (section 1.2.4).

Support go or no-go decisions, clearer accountability and better information sharing about risks, as listed in section 4.

Framework in Context

The AI RMF is a voluntary risk management framework, not a law. Section 1.2.4 says AI risks should be treated along with other critical risks such as cybersecurity and privacy, and that the AI RMF may be used with related frameworks, which is why it is often paired with the NIST Cybersecurity Framework and the NIST Privacy Framework. The document's update schedule says NIST will review it regularly, with a formal review including input from the AI community expected no later than 2028. On April 7, 2026 NIST released a concept note for an AI RMF Profile on Trustworthy AI in Critical Infrastructure.

Common Framework Mappings

NIST publishes an AI RMF Crosswalk and a Playbook, and organizations commonly use the AI RMF beside the frameworks and laws below to show how one risk process supports several expectations.

Mapped frameworks include:

EU AI Act

ISO 27001:2022

ISO 27701

ISO 42001

NIST CSF 2.0

NIST Privacy Framework v1.0

NIST 800-53 Rev.5

OECD AI Principles