What Is the NIST AI Risk Management Framework?

The NIST AI Risk Management Framework (AI RMF) is a voluntary framework, published by the U.S. National Institute of Standards and Technology, for managing the risks of artificial intelligence across its lifecycle. Released as AI RMF 1.0 on January 26, 2023, it gives organizations a common, technology-neutral way to identify, assess, and reduce AI risk while building the qualities of trustworthy AI: validity, reliability, safety, security, accountability, transparency, explainability, privacy, and fairness.

Unlike the EU AI Act, the AI RMF is not law and carries no penalties. It is a framework you adopt because it works: a structured, repeatable practice for governing AI that has become a de facto reference point for AI risk programs worldwide, often cited in procurement, board oversight, and regulatory readiness even outside the United States.

NIST AI RMF Defined

The AI RMF was developed through an open, consensus-driven process (multiple public drafts, workshops, and comment periods) in response to a Congressional direction in the National Artificial Intelligence Initiative Act of 2020. The result is deliberately broad: it applies to any organization, any sector, and any kind of AI system, and it can be used by the teams that design, develop, deploy, or evaluate AI.

Its defining characteristics:

• Voluntary and non-prescriptive. It describes outcomes to achieve, not a fixed checklist of controls, so organizations adapt it to their own risk profile and maturity.
• Risk-based and outcome-focused. It centers on the qualities of trustworthy AI and on harms to people and society, not only on risk to the organization.
• Lifecycle-oriented. It treats AI risk as a continuous practice across design, development, deployment, and monitoring, not a one-time gate.
• Technology-neutral. The same four functions apply to a credit model, a chatbot, or an autonomous agent.

NIST also ships companion resources: the AI RMF Playbook (suggested actions for each function), a Roadmap, a Crosswalk mapping the framework to other standards, and a set of profiles for specific use cases. The framework is currently undergoing revision, and in April 2026 NIST released a concept note for an AI RMF profile on trustworthy AI in critical infrastructure, a sign of how the framework keeps extending into higher-stakes domains.

The Four Core Functions

The heart of the AI RMF is four functions, each broken into categories and subcategories. Three of them (Map, Measure, Manage) run as an iterative cycle, while the fourth, Govern, is a cross-cutting function that surrounds and informs the other three.

• Govern. The cross-cutting function. It builds a culture of AI risk management: policies, accountable roles, oversight structures, and processes that connect the other three functions to the organization's values and legal obligations. Govern is what makes Map, Measure, and Manage durable rather than ad hoc.
• Map. Establish the context and frame the risks. Who are the stakeholders, what is the intended purpose, what could go wrong, and how does this AI system fit into a wider business process? Map produces the shared understanding the other functions depend on.
• Measure. Analyze, assess, and track the risks identified in Map, using quantitative and qualitative methods (metrics, evaluations, red-teaming, monitoring). Measure turns "this could be biased" into evidence you can act on.
• Manage. Prioritize and respond. Allocate resources to the risks that matter most, decide whether to treat, transfer, avoid, or accept each one, and plan for response and recovery when something goes wrong.

The functions are not a strict sequence. Map, Measure, and Manage are revisited continuously as a system changes, new data arrives, or new risks emerge, while Govern runs throughout.

Generative AI & Agents

On July 26, 2024, NIST released the Generative AI Profile (NIST AI 600-1), a companion that names twelve risks specific to generative AI, including confabulation (hallucination), data privacy, harmful bias, intellectual-property exposure, information integrity, and value chain and component integration, the untraceable upstream data and components feeding a model. For each, it offers suggested actions tagged to the four functions (Govern, Map, Measure, Manage); a profile does not replace the AI RMF, it tailors the same actions to a specific setting. Several of those actions call directly for tracking the provenance and history of training data and for vetting third-party suppliers, which is catalog, lineage, and classification work by another name.

This matters as organizations move from single models to agentic systems that plan, call tools, and act with growing autonomy. Agents widen the risk surface: they retrieve data from many sources, take actions in live systems, and chain decisions in ways that are hard to trace. The AI RMF stays applicable because its functions are technology-neutral, but applying them well to agents demands far more rigorous context: an agent can only be governed if you know what data and tools it can reach, what each means, and who is accountable for them. The framework names the discipline; making it real for agents is a data and context problem.

AI RMF vs EU AI Act & ISO 42001

These three are the reference points of modern AI governance, and they are complementary rather than competing:

• NIST AI RMF is a voluntary framework. It tells you how to think about and operate AI risk management, with maximum flexibility and no certification or legal force.
• The EU AI Act is binding law. It tells you what you must do for AI placed on the EU market, with risk tiers and penalties. Many of its obligations (risk management, data governance, documentation, human oversight) map closely onto AI RMF functions, so an AI RMF practice is strong preparation for it.
• ISO 42001 is a certifiable management-system standard. It tells you how to run an organization-wide AI management system and lets an accredited body audit and certify it.

A useful way to hold them together: NIST AI RMF gives you the practice of managing AI risk, ISO 42001 gives you a certifiable system to run that practice, and the EU AI Act sets the legal floor the practice has to clear. Many organizations use the AI RMF as the conceptual backbone and then formalize it through ISO 42001 and map it to AI Act obligations. NIST's own Crosswalk and the framework's shared concepts (a risk assessment discipline, lifecycle thinking, documentation) make moving between them far less work.

Governed Context as the Foundation

The AI RMF describes what good AI risk management looks like. It is largely silent on the substrate every one of its functions runs on: trustworthy, well-documented data and a clear, shared understanding of what that data means. In practice, that substrate is data governance.

Each function quietly depends on it:

• Map cannot establish context or identify risk without knowing what data an AI system uses, where it came from, and how sensitive it is. That is the job of a data catalog, a business glossary, and data classification.
• Measure cannot assess bias, drift, or quality without trusted inputs and the data lineage to trace a result back to its sources.
• Manage cannot assign and act on risk without clear ownership and accountability for the data and models involved.
• Govern cannot evidence oversight without the documented record that catalog, glossary, and lineage produce as a by-product of normal work.

This is where Dawiso fits. Its AI governance approach treats governed context as a layer the whole organization maintains once and reuses everywhere: a catalog of every data asset, a glossary that fixes what each term means, lineage that traces how data flows, and classification that flags what is sensitive, with an accountable owner for each. That governed context is exactly what the AI RMF's Map, Measure, and Manage functions need to be real rather than aspirational, and what an AI agent needs in order to be governable at all. The framework is the discipline; governed context is what makes it operable.

Conclusion

The NIST AI Risk Management Framework has become the common language for talking about AI risk: voluntary, flexible, and technology-neutral, with a clear core of Govern, Map, Measure, and Manage. It pairs naturally with the certifiable ISO 42001 and the binding EU AI Act, giving organizations a coherent path from practice to system to legal compliance. But like every governance framework, it rests on a single capability that it does not itself provide: knowing, documenting, and trusting the data and context behind your AI. Get that foundation right and the four functions have something solid to stand on.