AI Governance Frameworks You Can Implement Today

The board wants AI controls. Somebody has been told to go find a framework. And the honest answer inside most organizations is that nobody is sure which one, or what adopting one actually commits them to.

That confusion is reasonable, because the word “framework” is doing four different jobs in this conversation. One of these is a certifiable standard. One is a risk methodology. One is an audit and governance structure. One is a meta-framework designed to connect the others. They are not competing products and choosing between them is not a coin flip — it depends on a single question about your organization that most teams have never been asked.

This is the reference guide: what each one is, who it fits, who it does not, and where to find it. Every framework here is published and available to start implementing this week.

Before You Choose: What Comes First

None of this starts with the framework, and adopting one before you do the groundwork means retrofitting a standard onto an environment you never mapped.

Three things come first, in this order:

• The AI inventory — every AI system in the environment, who is using it, what data it touches, and what decisions it influences. This includes the AI now embedded in software you already bought and already approved. You cannot put a standard around something you have not found.
• One accountable owner — a single named person responsible for AI governance across the whole organization, not per department. Everyone has an opinion about what AI should be used for and what is too risky; accountability has to sit somewhere.
• The risk register — a documented risk assessment against what the inventory found, with each risk assigned an owner.

Only then does the framework question become answerable, because only then do you know what you are scoping it against.

ISO/IEC 42001 — The Certifiable Standard

What It Is

ISO/IEC 42001:2023, published in December 2023, is the first international standard for an Artificial Intelligence Management System (AIMS). It follows the Plan-Do-Check-Act model common to modern ISO management-system standards and is structured to align with ISO 27001 and ISO 9001.

It governs how you govern. Context and scope, leadership commitment and AI policy, risk assessment and AI impact assessment, operational control across the AI lifecycle, performance evaluation through internal audit and management review, and continual improvement.

The defining characteristic: it is certifiable by accredited third-party audit, with annual surveillance audits to maintain certification across the three-year cycle.

Reference: Published by the International Organization for Standardization. Available through the ISO catalogue at iso.org — search “ISO/IEC 42001:2023.”

Who It’s For

Organizations that need to prove governance to someone outside the building. If AI governance questions are showing up in your due diligence questionnaires, vendor security reviews, or enterprise contract negotiations, this is the one being asked for by name.

It is also the natural choice for organizations already holding ISO 27001, because the management-system scaffolding, internal audit function, and management review cadence already exist. You are extending a system rather than building one, which is a materially smaller project than it looks from outside.

Who It’s Not For

Organizations with no external pressure to demonstrate anything yet. Certification is a project with real cost and a recurring audit commitment, and pursuing it before you have an inventory and an owner produces an expensive certificate scoped to a fraction of your actual AI footprint.

It is also the wrong starting point if your immediate problem is that nobody knows how to assess AI risk. ISO 42001 tells you to have a risk process. It is less prescriptive about how to run one.

NIST AI RMF — The Risk Methodology

What It Is

The NIST AI Risk Management Framework 1.0, released January 26, 2023 by the National Institute of Standards and Technology within the US Department of Commerce, is a voluntary framework with no certification scheme attached.

It is organized around four functions: Govern, Map, Measure, Manage. Govern establishes accountability and runs through everything else. Map is understanding context and where AI is actually being used. Measure is assessing and tracking risk. Manage is acting on findings and prioritizing by risk.

NIST released it alongside a Playbook, a Roadmap, and a Crosswalk to other frameworks, and treats it as a living document.

Reference: nist.gov/itl/ai-risk-management-framework

Who It’s For

Organizations managing internal risk where nobody is auditing them yet, and organizations that need a method rather than a certificate. It gives teams unfamiliar with risk management a structure to follow instead of improvising, which is the most common real-world starting condition.

It carries particular weight for federal contractors and federal-adjacent organizations. NIST standards get referenced into government requirements and sector guidance, so alignment is frequently assumed even where it is not formally demanded. If your organization already runs NIST RMF for security, adding the AI RMF is incremental rather than new.

Who It’s Not For

Organizations that need external attestation. There is no certificate, no accredited assessor, and nothing to hand a procurement team. If a customer is asking you to prove AI governance, the AI RMF gives you good practice and shared vocabulary but no artifact that satisfies them.

COBIT and the ISACA Credentials

COBIT 2019 — The Audit and Governance Wrapper

COBIT 2019, from ISACA, is an enterprise IT governance framework rather than an AI-specific one. Auditors use it as the structure for assessing governance across domains, which makes it the lens through which your AI governance is likely to be examined whether or not you adopt it.

If you already run COBIT, AI governance becomes a domain inside a structure you already have. If you do not, understand it anyway — when auditors arrive, it shapes where they look for risk.

Who it’s not for: anyone considering adopting COBIT for AI. That is not what it is, and you will spend months building enterprise governance scaffolding that answers none of the AI-specific questions.

Reference: ISACA, at isaca.org — search “COBIT.”

AAISM, AAIA, and AAIR — The People Layer

A framework with nobody qualified to operate it is shelfware. ISACA now offers three AI credentials, each mapped to a different function:

• AAISM — Advanced in AI Security Management. Launched August 19, 2025; the first AI-centric security management certification. Requires an active CISM or CISSP. Three domains: AI Governance and Program Management, AI Risk Management, and AI Technologies and Controls. Exam is 90 questions in 150 minutes. Built on the NIST AI RMF, ISO/IEC 42001, the OWASP Top 10 for LLM Applications, and MITRE ATLAS. — isaca.org/credentialing/aaism
• AAIA — Advanced in AI Audit. For audit professionals holding a CISA or equivalent. The audit-side companion. — isaca.org/credentialing/aaia
• AAIR — Advanced in AI Risk. Introduced in 2026 for IT risk professionals. Three practice areas: AI risk governance and framework integration, AI lifecycle risk management, and AI risk program management. — isaca.org/credentialing/aair

Who these are for: the person you just named as accountable owner, and whoever will run the assessment cycle. Who they’re not for: organizations treating a credential as a substitute for a program. The certification qualifies a person; it does not govern anything by itself.

Reference: isaca.org/resources/artificial-intelligence

GRAICE — The Meta-Framework

What It Is

GRAICE™ — the Global Responsible AI Compliance & Ethics Framework — comes from the Global Council for Responsible AI (GCRAI), an independent international body operating across 77 countries with more than 1,000 Global Ambassadors. GRAICE was unveiled at the World Economic Forum Annual Meeting in Davos in 2026.

Its architecture runs on a single repeatable logic: Values → Pillars → Assurance → Governance.

• Six Foundational Values — Human Dignity and Autonomy; Accountability and Governance; Fairness and Justice; Transparency and Explainability; Reliability and Security; Inclusivity and Societal Benefit.
• Seven Pillars that translate those values into operational requirements — Ethical Leadership; Purpose-Driven Innovation; Human-Centric Use; Responsible Implementation; AI Literacy and Workforce Readiness; Data Governance and Integrity; Transparency and Auditability. Each pillar carries expectations, leadership questions, implementation pathways scaled by organization size, and the evidence artifacts assurance requires.
• A three-tier assurance model — Tier 1 Organizational Readiness, Tier 2 System-Level Assurance, Tier 3 Independent Assessment. It is lifecycle-based and continuous, with reassessment triggered by system updates, data changes, incidents, or shifts in risk or context.
• Three levels of governance — organizational, system-level, and independent or external — with explicitly defined roles including Responsible AI Officer, System Owner, Data Governance Lead, Human Oversight Lead, and Independent Reviewer, and defined decision rights for deployment approval, risk acceptance, system pause or shutdown, escalation, and external disclosure.

Reference: gcrai.ai — the framework’s executive overview is published at gcrai.ai/wp-content/uploads/2026/02/GRAICE-Framework-Overview.pdf

Why We Recommend It

Two reasons, and the second is the one that matters most.

The first is structural. GRAICE is designed as a global meta-framework that does not replace existing regulations or standards but connects them through a harmonized structure. Anyone who has done compliance work knows the tax: you end up mapping every framework and standard against every other one yourself, maintaining crosswalks by hand, and reproducing the same evidence in three formats. GRAICE does that mapping up front, which lets you reuse evidence across multiple regulatory and audit requirements and operate consistently across jurisdictions.

The second is the gap it closes. Look at the other frameworks on this page and notice what is missing: ethics. They govern security, risk, and management process. They do not ask whether a use of AI is one you should be making.

That gap matters because AI does not have ethics. It does not second-guess itself, does not ask whether it should, and does not stop to consider consequence. Somebody has to hold that line, and it has to be a person — which is why GRAICE builds the human-oversight role into the governance architecture as a named position with defined authority to pause or shut down a system, rather than leaving “human in the loop” as a principle nobody owns.

It also insists on evidence over assertion. Many frameworks articulate principles. GRAICE defines behaviors and requires artifacts, which is the difference between what an organization says about AI and what it can actually demonstrate.

Disclosure: WatchUr6’s CISO works with the Global Council for Responsible AI. We are telling you that up front because you should weigh it, and because we would rather disclose an affiliation than have you discover it.

Who It’s Not For

Organizations under immediate procurement pressure to produce a recognized certificate. GRAICE is newer than ISO/IEC 42001 and does not yet carry the same name recognition in enterprise due diligence questionnaires. If a customer contract is gating on a named certification this quarter, pursue ISO 42001 and use GRAICE as the governance architecture underneath it.

It is also not a shortcut. A meta-framework that spans seven pillars and three assurance tiers is more, not less, than a single standard. For an organization with three AI systems and no governance function at all, start with the inventory and the NIST AI RMF, then grow into it.

Reading Them Side by Side

The practical distinctions, in one pass:

• ISO/IEC 42001 — Certifiable. Management system. Produces an artifact third parties rely on. Choose when someone outside asks you to prove it.
• NIST AI RMF — Voluntary, not certifiable. Risk methodology. Produces internal practice and shared language. Choose when you need a method and nobody is auditing you yet.
• COBIT 2019 — Audit and governance structure, not AI-specific. Produces the lens auditors use. Understand it regardless; do not adopt it for AI.
• GRAICE™ — Meta-framework with a three-tier assurance model. Connects the others and adds the ethics and human-oversight layer none of them carry. Choose as the architecture, not as a replacement.
• AAISM / AAIA / AAIR — Individual credentials, not organizational frameworks. Choose based on whether the person is running security, audit, or risk.

And the point worth repeating: adopting one does not preclude another. The published crosswalks overlap substantially, and one well-built set of processes, policies, and documentation can satisfy more than one. The common sequence is ISO 42001’s management-system structure as the skeleton, the NIST AI RMF’s four functions as the risk methodology inside it, and GRAICE as the governance and ethics architecture connecting both to whatever regulation reaches you.

Where Audit Readiness Fits

Every framework on this page assumes you can produce evidence. That assumption is where most AI governance programs fail — not at the policy, but at the artifact.

That is the work of a structured Audit Readiness program. Assessment builds the inventory and establishes what your governance posture actually is, which is the prerequisite every framework on this page shares. Remediation closes gaps in the order that reduces exposure rather than the order a checklist outputs. Liaison is representation during the assessment itself — the keystone step, and the one that matters when an assessor starts asking how an AI system entered your environment.

Our GovCon industry brief covers the NIST 800-171 and CMMC posture these frameworks now sit alongside.

Marching Orders

1. Build the AI Inventory This Week

Every AI system, sanctioned or not, including features embedded in tools you already approved. Who uses it, what data it reaches, what decisions it influences, and which vendors have quietly added AI to software you already run. Nothing else on this list is possible without it.

2. Name One Accountable Owner

One person, across the whole organization. Put it in writing and in the org chart. This costs nothing and is the single highest-leverage move available to an organization with no AI governance today.

3. Answer the Prove-It Question

Ask whether anyone outside your company — customers, procurement, regulators, insurers — will require you to demonstrate AI governance, and on what timeline. That answer chooses your framework. Yes points to ISO/IEC 42001. Not yet points to the NIST AI RMF.

4. Build One Cross-Framework Register

One row per AI system, with the applicable requirements from each framework mapped against it. Not three parallel trackers maintained by three people, which diverge within a quarter. This is also the artifact GRAICE is designed to let you reuse across obligations.

5. Get Somebody Credentialed This Quarter

Match the credential to the role — AAISM for the security leader, AAIA for audit, AAIR for risk. A framework nobody in the building is qualified to operate does not survive its first year.

6. Train Your People Before You Widen Access

Publish what AI may be used for and what it may not, and train on it, before handing the organization a capability and hoping for the best. Handing out powerful tools with no instruction is not enablement.

Execute the Standard

Nobody is going to hand you a single rulebook for this. Regulation is arriving unevenly by jurisdiction, the standards are new, and the technology is moving faster than either.

That is not a reason to wait for clarity. It is the reason to pick a structure now. The organizations that choose a framework, run the inventory, and build the register will have an answer when a customer asks, when an auditor asks, and when the regulation finally does land. The ones waiting for certainty will start from zero on somebody else’s deadline.

If you need an outside read on your AI governance posture, help selecting and scoping a framework against what you actually run, or representation when the assessor arrives, that is the work we do. Verify your security posture or book a strategy call.

Build the inventory. Name the owner. Pick the framework and start the register.

Execute the standard.