NIST AI 100‑1 AI Risk Management Framework

Document Background

The AI Risk Management Framework (AI RMF 1.0) was released on January 26, 2023, following a multi‑year development process that included three public drafts, a request for information, several workshops, and input from over 240 organizations and individuals. Congress directed NIST to develop the framework through the National AI Initiative Act of 2020 (NIST, 2023).

The AI RMF is designed to be voluntary, rights‑preserving, non‑sector‑specific, and use‑case agnostic. It applies to any organization that designs, develops, deploys, or uses AI systems, regardless of whether those systems are commercial products, government tools, or research prototypes. Unlike a compliance checklist, it is a flexible framework that organizations can adapt to their specific context and risk tolerance (NIST, 2023).

What the AI RMF Does

The AI RMF helps organizations think about AI risks at every stage of the AI lifecycle, from initial concept through design, development, deployment, operation, and decommissioning. It recognizes that AI risks are different from traditional software risks in important ways (NIST, 2023).

First, AI systems can behave unpredictably. Unlike conventional software where outputs follow deterministic logic, machine learning models can produce unexpected results when they encounter data that differs from their training distribution. Second, the harms from AI can extend beyond the technical realm to include societal impacts like discrimination, loss of privacy, and erosion of trust. Third, AI risks are often emergent, meaning they may not be apparent until a system is deployed at scale in real‑world conditions.

The framework introduces the concept of AI actors: the people and organizations involved at each stage of the AI lifecycle. These include data scientists, engineers, product managers, executives, end users, and affected communities. Understanding who the AI actors are and what decisions they make is essential to managing risk effectively (NIST, 2023).

The Four Functions

The AI RMF is organized around four core functions. If you are familiar with the NIST Cybersecurity Framework (CSF), the structure will feel familiar. Each function contains categories and subcategories that provide more specific guidance (NIST, 2023).

1. Govern

   The Govern function is cross‑cutting. It establishes the organizational structures, policies, and processes needed to manage AI risk. This includes defining roles and responsibilities, setting risk tolerances, establishing accountability mechanisms, and fostering a culture of responsible AI development. Govern applies to and informs all other functions.

   Key activities include: establishing AI risk management policies, defining organizational AI principles, creating oversight structures, engaging diverse stakeholders, and documenting decisions and rationale.

2. Map

   The Map function is about context. Before you can manage the risks of an AI system, you need to understand what the system does, who it affects, and where the risks might emerge. Mapping involves identifying the intended and potential unintended uses of the system, understanding the operational environment, and recognizing the stakeholders and communities that may be impacted.

   Key activities include: defining the AI system's purpose and scope, identifying affected stakeholders, cataloging data sources and their limitations, and assessing the potential for bias and other harms.

3. Measure

   The Measure function focuses on assessment. It involves developing and applying metrics, methodologies, and tools to evaluate AI risks and the trustworthiness of AI systems. Measurement should occur throughout the AI lifecycle, not just at deployment. This function also emphasizes the importance of tracking known limitations and documenting the conditions under which the system performs well or poorly.

   Key activities include: selecting appropriate metrics for the AI system's context, conducting bias testing, evaluating robustness and reliability, and documenting model performance and limitations.

4. Manage

   The Manage function is about action. Based on what has been mapped and measured, organizations take steps to treat, mitigate, or accept identified AI risks. This includes deploying controls, monitoring systems in production, establishing incident response procedures, and making decisions about whether to continue, modify, or discontinue an AI system.

   Key activities include: implementing risk treatments, monitoring deployed systems for performance degradation, establishing processes for user feedback and incident reporting, and maintaining the ability to decommission systems that pose unacceptable risks.

Trustworthy AI Characteristics

The AI RMF defines seven characteristics of trustworthy AI systems. These are not binary pass/fail criteria but rather dimensions along which organizations should evaluate their AI systems. Tensions can exist between these characteristics, and organizations must make informed trade‑offs (NIST, 2023).

Parallels with CSF 2.0

Students who have studied the NIST Cybersecurity Framework (CSF) will notice structural similarities with the AI RMF. This is intentional. NIST designed the AI RMF to be compatible with and complementary to the CSF (NIST, 2023).

The parallels are not exact, and the AI RMF addresses concerns (bias, explainability, societal impact) that fall outside the traditional scope of cybersecurity. But the structural similarity means that organizations already using the CSF have a mental model and governance infrastructure they can extend to cover AI risks (NIST, 2023).

Citing This Document (APA 7)

The AI RMF uses the AI 100 series designation rather than the SP 800 series, but it follows the same APA 7 citation format as other NIST publications: the agency as author, the report number in parentheses, and the parent department as publisher.