← back to the archiveCover illustration for “NIST AI RMF: a framework for managing AI risk”
TERM7mo ago·Published · Updated ·by Andy Padia

NIST AI RMF: a framework for managing AI risk

In short: The NIST AI Risk Management Framework is a voluntary US framework that organises AI risk work into four functions: Govern, Map, Measure and Manage.

The NIST AI Risk Management Framework (AI RMF) is a voluntary framework from the US National Institute of Standards and Technology, released on 26 January 2023, to help organisations manage risks from AI systems. It organises the work into four functions: Govern, Map, Measure and Manage. NIST added a Generative AI Profile on 26 July 2024, and its page says version 1.0 is now being revised.

NIST AI RMF in plain words

A checklist-style playbook for asking the right questions about an AI system before, during and after it goes live, without being a law.

How it works

  1. Govern: policies, roles, accountability and culture for AI risk.
  2. Map: understand the context, intended use, affected people and possible harms.
  3. Measure: assess risks with tests, metrics and monitoring.
  4. Manage: prioritise, treat and track risks over the system's life.

The framework document describes trustworthy AI as valid and reliable, safe, secure and resilient, accountable and transparent, explainable and interpretable, privacy-enhanced, and fair with harmful bias managed. It offers suggested actions, with no mandatory controls. The Generative AI Profile adds 12 risks that generative AI creates or worsens, such as confabulation, data privacy and information security.

Example

As an illustration, before launching a loan-assistance chatbot, a bank maps who could be harmed (applicants misled about eligibility), measures accuracy and groundedness with evals, assigns a risk owner, and sets a monitoring plan with thresholds for rollback.

Where you'll meet it

  • Enterprise AI governance policies and risk registers
  • Vendor security and AI questionnaires
  • NIST's own playbook and crosswalk documents that map it to other frameworks
  • Public-sector AI procurement

What it changes for you

For leaders, it provides a shared vocabulary between engineering, risk and legal teams. For engineers, the Measure function is where evals, red-teaming and monitoring become governance evidence. My view: for services firms, aligning delivery to it makes client audits simpler, because the questions arrive in a familiar order.

NIST AI RMF vs ISO/IEC 42001

The NIST AI RMF is voluntary guidance with no certificate attached. ISO/IEC 42001 is a management-system standard that an organisation can be audited against. My view: use NIST to shape the practice, and add ISO when a customer asks for a certificate.

Common mistake

My view: producing a framework document and stopping there. The value is in the Measure and Manage functions running continuously, with real evals and owners, not in the policy PDF.

AI governance · ISO/IEC 42001 · EU AI Act · Red teaming · Evals

From the archive: Voluntary AI frameworks are regulatory signals

The framework pays off when Measure and Manage run continuously, with real evals and named owners.

#field-guide#governance#risk#compliance#frameworks

Sources

  1. nist.gov/itl/ai-risk-management-framework
  2. nvlpubs.nist.gov/nistpubs/ai/NIST.AI.100-1.pdf
  3. nvlpubs.nist.gov/nistpubs/ai/NIST.AI.600-1.pdf
← older drop
ISO/IEC 42001: the certifiable AI management standard
newer drop →
BM25: the keyword ranking formula behind search

related drops

explore all 557 drops →
← back to the archiveday 175