Filed under — NIST AI RMF · Govern Function · AI Governance · Risk Management · Compliance
Prefer this source on Google →NIST AI RMF Govern Function: Practical Implementation Guide
The NIST AI RMF Govern function establishes accountability and oversight for AI systems. Learn how to implement six of its outcomes (GOVERN 1.1, 1.2, 1.6, 2.1, 4.2 and 5.1) with practical examples and templates.
Free check · no account
Check your own system against what this article describes.
NIST AI RMF Govern Function: Practical Implementation Guide
Updated 2 October 2026 — correction. This guide called GOVERN 1.1 to 1.6 the Govern function's six categories, gave the wrong wording for GOVERN 1.3, 1.4 and 1.5, dated CV-screening obligations to 2 August 2026, cited Article 29 for deployers and presented a voluntary framework as binding. It now uses the subcategory numbers and wording of NIST AI 100-1 (impact documentation is GOVERN 4.2, roles GOVERN 2.1, external feedback GOVERN 5.1), dates Annex III systems to 2 December 2027, cites the adopted Regulation (EU) 2024/1689 and updates the Colorado row.
The NIST AI Risk Management Framework (AI RMF) organizes AI risk management into four functions: Govern, Map, Measure, and Manage. Of these, Govern is the foundation. NIST describes it as a cross-cutting function that is infused throughout AI risk management and enables the other three: it establishes the organizational structures, policies, and accountability mechanisms that all other risk management activities rely on.
If you're implementing the NIST AI RMF — whether to satisfy customer requirements, prepare for regulatory compliance, or establish defensible AI governance — start with Govern. This guide explains what the Govern function asks for, provides practical implementation steps, and includes templates you can use immediately.
What the NIST AI RMF Govern Function Actually Says
The Govern function is organized into six categories, GOVERN 1 to GOVERN 6, which hold 19 subcategories between them. This guide works through six of those subcategories:
- GOVERN 1.1: Legal and regulatory requirements involving AI are understood, managed, and documented
- GOVERN 1.2: The characteristics of trustworthy AI are integrated into organizational policies, processes, procedures, and practices
- GOVERN 4.2: Organizational teams document the risks and potential impacts of the AI technology they design, develop, deploy, evaluate, and use, and they communicate about the impacts more broadly
- GOVERN 2.1: Roles and responsibilities and lines of communication related to mapping, measuring, and managing AI risks are documented and are clear to individuals and teams throughout the organization
- GOVERN 5.1: Organizational policies and practices are in place to collect, consider, prioritize, and integrate feedback from those external to the team that developed or deployed the AI system regarding the potential individual and societal impacts related to AI risks
- GOVERN 1.6: Mechanisms are in place to inventory AI systems and are resourced according to organizational risk priorities
GOVERN 1.3 to 1.5 are about something else: deciding the needed level of risk management from the organization's risk tolerance (1.3), establishing the risk management process and its outcomes through transparent policies (1.4), and monitoring and periodically reviewing that process (1.5).
The framework is voluntary. These outcomes are not legal duties, but they are concrete organizational capabilities that you build and document if you adopt it.
Why Govern Is Harder Than It Looks
Most organizations assume they already have "governance" because they have an AI ethics policy or a responsible AI committee. But the NIST AI RMF asks for something more rigorous: documented processes, assigned accountability, and continuous risk tracking.
Here's what breaks down in practice:
- No legal/regulatory tracking: You know the EU AI Act exists, but you haven't assigned anyone to track new AI regulations or assess their impact on your systems.
- No trustworthy AI definition: You talk about "responsible AI," but you haven't defined what that means for your organization or integrated it into product development processes.
- No impact assessment process: You deploy AI systems, but you've never documented their impact on users, communities, or society.
- No dedicated AI risk team: AI risk management is "everyone's responsibility," which means no one is actually accountable.
- No external feedback mechanism: You don't have a process to collect feedback from affected communities, civil society, or domain experts.
- No AI system inventory: You don't have a centralized list of all AI systems in production, their risk levels, or their compliance status.
The NIST AI RMF Govern function asks you to close all of these gaps — and to be able to show that you've closed them.
GOVERN 1.1: Legal and Regulatory Requirements
What it asks for:
Your organization identifies, understands, and tracks legal and regulatory requirements that apply to your AI systems. This includes sector-specific regulations (e.g., healthcare, finance) and horizontal AI regulations (e.g., EU AI Act, state-level AI laws).
Practical implementation:
- Assign ownership: Designate a Legal/Compliance lead responsible for tracking AI regulations.
- Create a regulatory tracker: Maintain a living document that lists applicable regulations, their enforcement dates, and their impact on your AI systems.
- Conduct quarterly reviews: Review the tracker quarterly and update it with new regulations or guidance.
- Integrate into product development: Require that every new AI system undergo a regulatory compliance check before deployment.
Example regulatory tracker:
| Regulation | Jurisdiction | Enforcement Date | Applicable Systems | Compliance Status |
|---|---|---|---|---|
| EU AI Act, Article 50 | EU | Aug 2, 2026 | Customer chatbot (transparency) | In progress |
| EU AI Act, Annex III | EU | Dec 2, 2027 | CV screening AI (high-risk) | In progress |
| Colorado SB26-189 (repeals and re-enacts SB24-205) | Colorado, USA | Jan 1, 2027 (first developer duties) | Automated decision-making in consequential decisions | Not started |
| NYC Local Law 144 | New York City | Jul 5, 2023 | HR AI tools | Compliant |
| GDPR Article 22 | EU | May 25, 2018 | All automated decision-making | Compliant |
Deliverable: A regulatory compliance tracker, updated quarterly, with assigned ownership.
GOVERN 1.2: Trustworthy AI Characteristics
What it asks for:
Your organization defines what "trustworthy AI" means and integrates those characteristics into organizational policies, procedures, and practices.
The NIST AI RMF identifies seven characteristics of trustworthy AI:
- Valid and reliable: The system performs as intended.
- Safe: The system does not cause unacceptable harm.
- Secure and resilient: The system withstands attacks and recovers from adverse events.
- Accountable and transparent: Information about the system and its outputs is available to the people who interact with it, and someone answers for it.
- Explainable and interpretable: Stakeholders can understand how the system works and what its outputs mean.
- Privacy-enhanced: The system protects personal data.
- Fair – with harmful bias managed: The system does not produce discriminatory outcomes.
Practical implementation:
- Adopt or adapt the NIST characteristics: Use the seven NIST characteristics as a starting point, or customize them for your organization.
- Document in an AI policy: Create or update your AI governance policy to explicitly reference these characteristics.
- Integrate into product development: Require that every AI system design document address how it satisfies each characteristic.
- Create acceptance criteria: Define measurable acceptance criteria for each characteristic (e.g., for "Fair", a stated maximum gap in outcomes between groups, with the threshold and its rationale written down).
Example policy language:
All AI systems developed or deployed by [Company Name] must satisfy the following trustworthy AI characteristics: validity, safety, security, accountability, explainability, privacy, and fairness. Each AI system design document must include a section titled "Trustworthy AI Assessment" that addresses how the system satisfies each characteristic.
Deliverable: An AI governance policy that defines trustworthy AI characteristics and integrates them into product development.
GOVERN 4.2: Impact Assessment Process
What it asks for:
Your teams document the risks and potential impacts of the AI systems they design, develop, deploy, evaluate, and use, and communicate about those impacts. (The MAP function then asks for the likelihood and magnitude of each impact to be identified and documented, in MAP 5.1.)
Practical implementation:
- Create an impact assessment template: Develop a structured template that prompts teams to consider impacts across multiple dimensions (individual, group, societal).
- Require impact assessments for high-risk systems: Mandate that all high-risk AI systems (e.g., those affecting employment, credit, or essential services) undergo an impact assessment before deployment.
- Involve diverse stakeholders: Include legal, ethics, product, and domain experts in the assessment process.
- Document and review: Store completed impact assessments in a centralized repository and review them annually.
Example impact assessment template:
| Impact Dimension | Questions to Consider | Assessment | Mitigation |
|---|---|---|---|
| Individual | Could this system harm individual users? Could it affect their rights or opportunities? | Medium risk: System may deny loan applications | Human review for all denials |
| Group | Could this system disproportionately affect a protected group (race, gender, age, disability)? | Low risk: Bias testing shows no disparate impact | Ongoing bias monitoring |
| Community | Could this system affect community cohesion, trust, or access to resources? | Low risk: System used only for internal credit scoring | N/A |
| Organizational | Could this system create reputational, legal, or operational risk for the organization? | Medium risk: Regulatory scrutiny likely | Compliance audit before deployment |
| Societal | Could this system contribute to broader societal harms (e.g., surveillance, inequality)? | Low risk: System not used for surveillance | N/A |
Deliverable: An impact assessment template and a repository of completed assessments.
GOVERN 2.1: AI Risk Management Teams
What it asks for:
Your organization documents roles, responsibilities, and lines of communication for AI risk management, so that they are clear to individuals and teams. (GOVERN 2.3 adds that executive leadership takes responsibility for decisions about AI risks.)
Practical implementation:
- Define roles: Identify who is responsible for AI risk management activities (e.g., AI Risk Lead, Legal/Compliance Lead, Product Owners, Data Scientists).
- Create a RACI matrix: Document who is Responsible, Accountable, Consulted, and Informed for each AI risk management activity.
- Establish a cross-functional AI governance committee: Convene a committee that meets quarterly to review AI risks, compliance status, and policy updates.
- Assign accountability: Ensure that every AI system has a named owner who is accountable for its risk management.
Example RACI matrix:
| Activity | AI Risk Lead | Legal/Compliance | Product Owner | Data Scientist |
|---|---|---|---|---|
| Regulatory tracking | I | A/R | I | I |
| Impact assessment | C | C | A/R | C |
| Bias testing | C | I | C | A/R |
| Incident response | A/R | C | C | C |
| Policy updates | A/R | C | I | I |
Key: A = Accountable, R = Responsible, C = Consulted, I = Informed
Deliverable: A RACI matrix and a charter for the AI governance committee.
GOVERN 5.1: External Feedback Mechanisms
What it asks for:
Your organization has policies and practices to collect, consider, prioritize, and integrate feedback from those outside the team that developed or deployed the AI system (users, affected communities, civil society, domain experts).
Practical implementation:
- Establish feedback channels: Create mechanisms for external stakeholders to provide feedback (e.g., a dedicated email address, a feedback form, public consultations).
- Document feedback: Log all external feedback in a centralized tracker.
- Review and prioritize: Review feedback quarterly and prioritize items for action.
- Close the loop: Communicate back to stakeholders how their feedback was considered and what actions were taken.
Example feedback tracker:
| Date | Source | Feedback Summary | Priority | Action Taken | Status |
|---|---|---|---|---|---|
| Jan 15, 2026 | User email | CV screening AI rejected qualified candidate | High | Reviewed case; updated training data | Closed |
| Feb 3, 2026 | Civil society org | Request for bias testing results | Medium | Published summary of bias testing methodology | Closed |
| Mar 10, 2026 | Domain expert | Suggested improvement to explainability | Low | Added to product roadmap for Q3 | Open |
Deliverable: A feedback tracker and a documented process for external feedback collection and review.
GOVERN 1.6: AI System Inventory
What it asks for:
Your organization has mechanisms to inventory its AI systems, resourced according to its risk priorities.
Practical implementation:
- Create an AI system registry: Develop a centralized database or spreadsheet that lists all AI systems in development or production.
- Capture key metadata: For each system, document: name, owner, intended purpose, risk level, compliance status, deployment date.
- Update regularly: Require that the registry is updated whenever a new AI system is deployed or an existing system is modified.
- Link to risk assessments: Ensure that each system in the registry links to its impact assessment, bias testing results, and compliance documentation.
Example AI system inventory:
| System Name | Owner | Intended Purpose | Risk Level | Compliance Status | Deployment Date |
|---|---|---|---|---|---|
| CV Screening AI | HR Tech Lead | Automate candidate screening | High-risk (EU AI Act Annex III, from Dec 2, 2027) | In progress | Q3 2026 |
| Fraud Detection AI | Payments Lead | Detect fraudulent transactions | Not high-risk (Annex III point 5(b) excludes fraud detection) | No Article 50 duty identified | Jan 2024 |
| Chatbot | Customer Support Lead | Answer customer questions | Not high-risk | Article 50(1) AI disclosure in place since Aug 2, 2026 | Mar 2025 |
Deliverable: An AI system inventory with links to risk assessments and compliance documentation.
How the Govern Function Connects to EU AI Act Compliance
If you're preparing for EU AI Act compliance, the NIST AI RMF Govern function provides a structured approach that supports many of the regulation's requirements:
- GOVERN 1.1 → Tracks the EU AI Act and other regulations
- GOVERN 1.2 → Integrates trustworthy AI characteristics that correspond to the high-risk requirements (Articles 9, 10, 13, 14, 15)
- GOVERN 4.2 → Supports the risk management system (Article 9) and the fundamental rights impact assessment that certain deployers must carry out (Article 27)
- GOVERN 2.1 → Establishes accountability for the obligations of providers (Article 16) and deployers (Article 26)
- GOVERN 5.1 → Collects feedback from affected communities; the Act has no general duty to consult them, but providers' post-market monitoring collects and analyses data on system performance (Article 72), and any person may complain to a market surveillance authority (Article 85)
- GOVERN 1.6 → Maintains an AI system inventory (not a requirement of the Act as such, but classifying each system under Article 6 starts from knowing which systems you have)
Implementing the NIST AI RMF Govern function is not a substitute for EU AI Act compliance, but it provides the organizational foundation you need.
This article is for informational purposes only and does not constitute legal advice. Consult a qualified AI governance expert or attorney for guidance specific to your organization.
Related dispatches