An AI governance framework is the structure a company uses to decide who can approve AI use, how AI risks are assessed, and how those decisions are monitored and reported. For most leadership teams, the right starting point is the NIST AI Risk Management Framework, because it is publicly available, voluntary and flexible. Build it so it can mature into ISO/IEC 42001 certification if customers, investors or federal buyers later ask for independent proof.
This article compares the two frameworks, explains when certification is worth the effort, and gives you a one-page template you can adopt. It focuses on framework selection and structure. Policy wording, such as what employees may paste into a chatbot, belongs in a separate document; our AI acceptable use policy template covers that.
What an AI governance framework is and why leadership needs one
A policy tells people what to do. A framework tells the company how decisions about AI get made, by whom, and how you know the decisions are holding up. It answers four questions:
- Who is accountable for AI risk, and who has the authority to say no?
- What AI do we use, build or buy, and where does it touch data, customers or employees?
- How do we judge whether a given use is low risk or high risk?
- How do we keep watching once it is live?
Leadership teams need one for practical reasons. AI arrives through many doors at once: features switched on inside existing software, tools staff adopt on their own, vendors embedding models in their products, and internal projects. Without a shared structure, each decision is made in isolation, and nobody can tell the board, an insurer or a customer what the company's AI exposure actually is. When a questionnaire or diligence request asks how you govern AI, a framework is what lets you answer clearly instead of assembling an answer under deadline.
The NIST AI risk management framework in practice
The NIST AI RMF is published by the National Institute of Standards and Technology. It is voluntary guidance, not a regulation, and there is no certification against it. It is written to apply to any organization that designs, develops, deploys or uses AI, in any sector. NIST also publishes a companion Playbook of suggested actions and a Generative AI Profile that applies the framework to generative models.
The framework describes the characteristics of trustworthy AI: valid and reliable; safe; secure and resilient; accountable and transparent; explainable and interpretable; privacy-enhanced; and fair, with harmful bias managed. Its core is organized into four functions.
Govern
Govern is cross-cutting and shapes the other three. In practice it means a named executive owner, a written statement of how much AI risk the company will accept, defined roles, policies, training, and a process for AI that comes from third parties. It is also about culture: people need to know that raising a concern about an AI system is expected, not resented.
Map
Map establishes context. For each AI system, you record its purpose, users, data, the people it could affect and what happens if it gets something wrong. You cannot manage what you have not mapped, and most companies find AI they did not know they had.
Measure
Measure assesses the risks you mapped. For systems you build, that means testing outputs for accuracy, bias and security weaknesses such as prompt injection (crafted inputs that trick a model into ignoring its instructions). For tools you buy, it means reviewing vendor evidence and running your own spot checks. Results are documented, not just discussed.
Manage
Manage is where you act. You prioritize risks, decide whether to accept, reduce, transfer or avoid each one, plan how to respond when an AI system fails, and retire systems that no longer earn their place.
ISO/IEC 42001 and how certification works
ISO/IEC 42001 is the international standard for an AI management system, often shortened to AIMS. It is a management system standard, which means it specifies requirements for establishing, running, maintaining and continually improving how an organization governs AI. It uses the same high-level structure as ISO/IEC 27001: clauses on context, leadership, planning, support, operation, performance evaluation and improvement. It adds a set of AI-specific controls, a statement of which controls apply to you and why, an AI risk assessment and an AI system impact assessment.
Certification is performed by an independent certification body. The typical path looks like this:
- Define the scope: which business units, products and AI systems the management system covers.
- Build the management system: AI policy, objectives, risk and impact assessments, selected controls and documentation.
- Operate it long enough to produce evidence, including an internal audit and a management review.
- Undergo the external audit: a review of your documentation, then an audit of how the system works in practice.
- Maintain the certificate through periodic surveillance audits and recertification.
Two points matter for leaders. Choose an accredited certification body, because an unaccredited certificate carries little weight with sophisticated buyers. And a certificate covers only its stated scope, so customers will read the scope statement.
NIST AI RMF vs ISO 42001: key differences
| | NIST AI RMF | ISO/IEC 42001 | |---|---|---| | Purpose | Guidance on identifying and managing AI risk | Requirements for an auditable AI management system | | Type | Voluntary framework | International standard | | Structure | Four functions: Govern, Map, Measure, Manage | Management system clauses plus AI-specific controls | | Certifiable | No | Yes, by accredited certification bodies | | Effort | Scales with how far you take it | Sustained documentation, internal audit and external audits | | Cost drivers | Internal time | Purchasing the standard, internal time and audit costs | | Who tends to ask | US federal buyers, boards and regulators familiar with NIST | Enterprise procurement, international customers, organizations with EU exposure |
The frameworks are complementary rather than competing. NIST describes what good AI risk practice looks like. ISO describes how to run that practice as a system an auditor can verify. The four NIST functions map naturally onto ISO's clauses, so work done under one is rarely wasted under the other.
Which framework to start with
For most companies, start with the NIST AI RMF. It lets you put useful structure in place quickly, it speaks the same language as the NIST Cybersecurity Framework many security programs already use, and it does not commit you to audit cycles before you know whether anyone will ask for them.
When ISO 42001 certification becomes worth it
Certification becomes worth the effort when one or more of these is true:
- A material customer or prospect requires it in a contract or procurement process.
- You sell AI-enabled products internationally, particularly into the EU. The EU AI Act places obligations on providers and deployers of certain AI systems. Certification does not equal legal compliance, but it is evidence of organized, auditable governance.
- You already hold ISO 27001, so the shared structure lowers the additional effort.
- Investors or acquirers want independent assurance rather than self-attestation.
- A federal solicitation or prime contractor asks for it. Federal buyers generally speak NIST's language, so read each solicitation rather than assuming.
If none of these apply, a well-documented framework aligned to the NIST AI RMF is enough. Do not certify for the logo.
A lightweight AI governance framework for a growing company
The lightweight version uses the four NIST functions for substance and borrows ISO's discipline for structure: a named owner, a defined scope, a consistent risk method and kept records. That combination is what lets it mature into certification later.
Roles
- Accountable executive. Usually the CEO, COO or CISO. Owns risk appetite and signs off on high-risk uses.
- AI governance lead. Runs the framework day to day. Often the security or risk leader.
- Review group. Legal or privacy, security, the relevant business owner, and engineering where you build AI. Keep it small enough to meet quickly.
- System owners. Every AI system has one named person responsible for it.
AI inventory
The inventory is the backbone. Record, for each system: name, vendor or internal, purpose, owner, data types involved, users and affected parties, whether it influences decisions about people, risk tier, last review date and status. Build the first version from single sign-on logs, expense reports, your vendor list and short conversations with each team.
Risk tiers
- Low. Productivity use with no sensitive data, where a person reviews every output.
- Moderate. Touches confidential internal data, or produces customer-facing content that a person reviews before release.
- High. Makes or materially influences decisions about people (hiring, credit, health, eligibility), processes regulated data, acts autonomously, or faces customers without human review.
Review gates
- Intake. Nothing goes into use until it is in the inventory.
- Pre-deployment. Effort matches the tier: self-attestation for low, security and privacy review for moderate, full risk assessment and executive sign-off for high.
- Change. A new model, new data source or new use case triggers re-review.
- Retirement. Access removed, data handled, inventory updated.
Monitoring
Set review frequency by tier, with high-risk systems reviewed most often. Keep an incident log for AI failures and near misses, watch vendor change notices, and track a small set of metrics you can report upward.
How to run an AI risk assessment under the framework
A good AI risk assessment framework fits on a few pages per system. Run it the same way every time:
- Describe the system. What it does, who uses it, what data goes in and what comes out.
- Identify who is affected. Employees, customers, applicants, the public.
- List the risks. Data exposure, inaccurate or fabricated output, bias, security weaknesses such as prompt injection or poisoned training data, intellectual property and legal issues, over-reliance on a single vendor, and reputational harm.
- Rate likelihood and impact, then assign the tier.
- Choose treatments. Controls such as human review, data restrictions, logging, contract terms or a decision not to proceed.
- Record the decision. Owner, residual risk and who approved it.
- Set the next review date and the events that would trigger an earlier one.
This record later serves as the basis for ISO 42001's risk and impact assessments.
A one-page AI governance framework template
Copy this outline into a document and fill in each line. The bracketed notes show which NIST function each part supports.
- Purpose and scope. Which AI uses, business units and products the framework covers. [Govern]
- Risk appetite. Uses we welcome, uses that need approval, uses we prohibit. [Govern]
- Roles. Accountable executive, governance lead, review group, system owners. [Govern]
- AI inventory. Where it lives, who maintains it, required fields. [Map]
- Risk tiers. Criteria for low, moderate and high. [Map]
- Review gates. Intake, pre-deployment, change and retirement steps by tier. [Manage]
- Risk assessment method. The steps above and the rating scale. [Measure]
- Third-party AI. Vendor questions, contract terms, review of embedded AI features. [Govern, Map]
- Testing and evidence. What is tested, how often, where results are kept. [Measure]
- Monitoring and incidents. Review cadence, incident logging and escalation. [Manage]
- Training. Who is trained on what, and when. [Govern]
- Reporting. Metrics for leadership and the board. [Govern]
- Framework review. Who reviews this document and how often. [Govern]
Who should own it and how long implementation takes
Ownership works best split by role. An accountable executive owns the risk decisions. The operational owner is usually the CISO, because AI risk overlaps heavily with data protection, vendor risk and access control. Legal owns interpretation of laws and contracts. A product leader owns AI that the company builds. Avoid handing the whole thing to legal alone, where it tends to become a contract exercise, or to product alone, where speed tends to win every trade-off. If you do not have a security leader, fractional CISO services are one option to consider.
On timing, a first working version is a matter of weeks of focused effort, not a long program. The inventory and tiering take the most time because they require talking to every team. Run the framework through a few review cycles before calling it mature. ISO 42001 readiness takes considerably longer, because certification requires evidence that the system has operated, including an internal audit and a management review.
How the framework connects to NIST CSF, SOC 2 and board reporting
Treat AI governance as part of your existing security and risk program, not a parallel one.
- NIST CSF. The current version of the Cybersecurity Framework includes a Govern function. Add AI systems to your asset inventory and AI risks to the same risk register.
- SOC 2. Existing controls for vendor management, change management, access and data classification extend naturally to AI. If AI features sit inside a service covered by your SOC 2 report, expect auditors and customers to ask about them.
- ISO 27001. If you hold it, build ISO 42001 readiness on the same management system.
- Board reporting. Give the board a short, consistent view each quarter: AI systems by tier, high-risk approvals, incidents, open exceptions and relevant regulatory developments.
If you want outside help connecting this to the rest of your security program, see our work on AI governance and AI policy.
Build the structure now, earn the certificate when it pays
Start with a lightweight framework aligned to the NIST AI RMF: a named owner, an inventory, three risk tiers, clear review gates and a short board report. Keep records the way an auditor would expect to see them. That gives you defensible AI governance today and a direct path to ISO 42001 if a customer, investor or regulator makes certification worth the effort.
If you want help assessing your exposure, start with the free AI risk assessment. It takes five minutes and gives you a baseline score right away.
For the field guide version of what I publish here each week, pick up a copy of Exposed: Inside Risks and the New Architecture of AI Defense on Amazon.
NightFortress works with executives, founders and leadership teams in Northern Virginia and the DC metro area to assess exposure, govern risk, and build security programs that match the actual threat landscape. Contact us to start a conversation.
The information in this article is for educational and informational purposes only. It is not intended as legal, compliance, or professional cybersecurity advice for any specific organization. Consult qualified professionals before making security or compliance decisions.

