NIST AI RMF (AI Risk Management Framework)
A voluntary US framework from NIST (version 1.0, January 2023) for managing AI risk through 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 for managing the risks of AI systems. Version 1.0 was released on January 26, 2023. It organizes the work into four functions: Govern, Map, Measure and Manage.
Govern covers policies, roles and culture and applies across the other three. Map means understanding the context: what the system is for, who it affects, what could go wrong. Measure means testing and tracking those risks, for example with evals and red teaming. Manage means deciding what to do about them and responding to incidents. NIST also publishes a Playbook with suggested actions and, on July 26, 2024, released the Generative AI Profile (NIST AI 600-1), which lists risks specific to generative AI, such as confabulation and information security, with actions for each.
It is not a law and there is no certificate. That is the difference from its neighbors: the EU AI Act is binding law with fines, and ISO/IEC 42001 is a management system standard you can be audited and certified against. US enterprise buyers and their security questionnaires often ask whether a vendor aligns with the AI RMF. A credible answer maps your existing AI governance to the four functions.
Sources
- NIST: AI Risk Management Framework nist.gov
- NIST: Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile nist.gov
- NIST: AI Risk Management Framework 1.0 (NIST AI 100-1) doi.org
Related terms
- ISO/IEC 42001 (AI management system): The international standard for an AI management system: how a company governs, assesses and monitors the AI it builds or uses. It can be certified.
- EU AI Act: The EU regulation on AI systems: bans some uses, regulates high-risk ones, and sets transparency and AI literacy duties.
- AI governance: The rules and roles that decide which AI a company uses, on what data, with what checks: policy, inventory, risk tiers, approvals and monitoring.
- Security questionnaire (vendor risk assessment, SIG, CAIQ): A list of security and privacy questions a company sends a vendor before buying, to judge the risk of trusting it with data and access.
- AI red teaming: Deliberately attacking your own AI feature with harmful, tricky and injected inputs to find failures before users or attackers do.
- AI system inventory (AI register): A maintained list of every AI tool and system a company uses or builds, with owner, purpose, data, vendor and risk level for each.
