Organizations deploying AI systems face growing governance obligations. The NIST AI Risk Management Framework provides the most practical structure for managing AI risk. Here is how to use it.
Artificial intelligence is no longer a future consideration for most organizations — it is a present operational reality. AI systems are making credit decisions, triaging medical records, screening job applicants, detecting fraud, and generating customer-facing content at scale.
With that deployment comes risk. And with that risk comes governance obligation.
The NIST AI Risk Management Framework (AI RMF 1.0), released in January 2023, is the most comprehensive and widely adopted structure for managing AI risk in the United States. Understanding what it requires — and how to implement it — is increasingly essential for any organization deploying AI systems.
Why AI Governance Is Different From Traditional IT Risk
Traditional IT risk management focuses on confidentiality, integrity, and availability. AI risk management requires a broader lens.
AI systems can fail in ways that traditional software does not. They can produce biased outputs that discriminate against protected classes. They can be manipulated through adversarial inputs. They can drift over time as the data they were trained on becomes less representative of current reality. They can make consequential decisions that are difficult or impossible to explain.
These failure modes require governance structures that go beyond standard security controls. They require oversight of model development, deployment, monitoring, and retirement — and accountability for outcomes, not just processes.
The Structure of the NIST AI RMF
The AI RMF is organized around two components: the Frame and the Core.
The Frame establishes the context for AI risk management — the organizational culture, roles, responsibilities, and processes that enable effective governance. It emphasizes that AI risk management is not a one-time assessment but an ongoing organizational capability.
The Core organizes AI risk management activities into four functions:
Govern. Establishes the organizational policies, processes, and accountability structures for AI risk management. This includes defining risk tolerance, assigning roles and responsibilities, establishing governance committees, and integrating AI risk into enterprise risk management.
Map. Identifies and categorizes AI risks in context. This involves understanding the intended use of the AI system, the stakeholders affected, the potential harms, and the organizational risk tolerance for those harms.
Measure. Analyzes and assesses AI risks using qualitative and quantitative methods. This includes evaluating bias, fairness, robustness, explainability, and security properties of AI systems.
Manage. Prioritizes and addresses AI risks based on the measurement results. This includes risk treatment decisions, monitoring, incident response, and continuous improvement.
The Seven Properties of Trustworthy AI
The AI RMF defines seven properties that characterize trustworthy AI systems. These properties are not binary — they exist on a spectrum and must be balanced against each other and against business objectives.
Accountable and transparent. Organizations can explain how AI systems make decisions and are responsible for their outcomes.
Explainable and interpretable. AI outputs can be understood by the people affected by them and the people responsible for them.
Fair with bias managed. AI systems do not produce discriminatory outcomes across demographic groups, and bias is actively identified and mitigated.
Privacy-enhanced. AI systems respect privacy rights and comply with applicable privacy regulations.
Reliable and accurate. AI systems perform as intended across the range of conditions they will encounter in deployment.
Resilient. AI systems maintain acceptable performance under adversarial conditions and unexpected inputs.
Safe. AI systems do not cause unacceptable harm to people or the environment.
Practical Implementation: Where to Start
For most organizations, the most practical starting point is the Govern function — establishing the organizational foundation before diving into technical risk assessment.
Define your AI inventory. You cannot govern what you cannot see. Catalog every AI system your organization uses or develops, including third-party AI tools embedded in your software stack.
Assign accountability. For each AI system, identify who is responsible for its governance. This is often more complex than it sounds — responsibility may be distributed across product, engineering, legal, and business teams.
Define risk tolerance. What level of AI risk is acceptable for your organization? This varies by use case. An AI system making credit decisions has a different risk profile than one generating marketing copy.
Establish a review process. Before deploying a new AI system, what review does it go through? Who approves it? What documentation is required?
ISO/IEC 42001 and the EU AI Act
The NIST AI RMF is not the only governance structure organizations need to understand. ISO/IEC 42001, published in 2023, provides an auditable management system standard for AI — analogous to ISO 27001 for information security. Organizations seeking third-party certification of their AI governance program will increasingly pursue ISO 42001.
The EU AI Act, which entered into force in 2024, imposes mandatory requirements on AI systems deployed in the European Union, with obligations scaled to risk level. High-risk AI systems — those used in employment, credit, healthcare, law enforcement, and other sensitive domains — face the most stringent requirements.
Organizations operating globally need to understand how these frameworks interact and where their obligations lie. Our resource article on ISO/IEC 42001 and the EU AI Act for small and mid-sized companies covers the practical overlap in detail. For a direct comparison of ISO/IEC 42001 and the NIST AI RMF — including how to decide which one belongs at the center of your program — see our post on building an AI governance program: ISO/IEC 42001 and NIST AI RMF compared.
The GRC2 Approach to AI Governance
At GRC2, we help organizations build AI governance programs that are practical, auditable, and aligned to the frameworks that matter to their regulators and customers. We start with your AI inventory, assess your current governance posture against the NIST AI RMF and ISO 42001, and build a program that addresses your specific risk profile.
AI governance is not a compliance checkbox. It is a risk management discipline that protects your organization, your customers, and the people affected by your AI systems.
One of the most underappreciated dimensions of AI governance is the risk that lives in your vendor stack — the AI tools your teams are already using that were never formally reviewed. Our post on AI supply chain risk covers how to surface that exposure. And once your governance program is in place, our post on how boards should think about AI security covers how to bring leadership along — including the four questions every board should be asking their security team.
Explore Topics
Written by
GRC2 Advisory Team
Content creator and writer sharing insights and stories.