Skip to content
TRANSMISSION ACTIVE
// FREQ: GOVCON EPISODE: 033 STATUS: SECURE

033 The Race for an AI Governance Framework: ISO 42001, NIST AI RMF, and How to Choose

The board wants AI controls, procurement is writing AI clauses into contracts, and nobody in the building is sure which framework to commit to. This transmission walks the four options actually on the table — and why none of them is where you start.

JUMP POINTS //

01:18

Pressure From Every Direction At Once

The board wants market advantage. Customers want features the moment they think of them. Procurement is reading AI clauses into contracts. And states are writing their own AI disclosure rules while federal policy runs behind the technology.

04:38

ISO 42001 — The One Showing Up in Contracts

Published in 2023 as the first AI management system standard. Why its alignment with ISO 27001 makes it a building block rather than a new program, and what “certifiable” actually buys you.

07:03

NIST AI RMF — A Foundation Instead of Guessing

Not a certification and not mandatory. Govern, Map, Measure, Manage — and why our CISO says it matters most for organizations that have never run a risk program at all.

08:52

COBIT and the ISACA Credentials

Why COBIT is the lens auditors bring rather than something you adopt for AI, and how the ISACA AI certifications map onto the security, audit, and risk roles you already have.

11:11

GRAICE — The Framework That Includes Ethics

The Global Council for Responsible AI’s meta-framework. It maps the other standards together so you don’t have to, and it covers the thing none of the others do: whether a use of AI is one you should be making.

13:05

Don't Start With the Framework

Our CISO’s direct answer to “which one do I pick.” Inventory, accountable owner, risk register — in that order — before the framework question is even answerable.

17:58

What Happens When You Just Turn Copilot On

A person can only see so much data. AI sees all of it at once. Why organizations flip on AI, discover staff reading highly sensitive information, and have to switch it back off — plus whether the AI spend is returning anything at all.

23:02

The Marching Orders

Three moves: the inventory and the owner, the question of who outside your company will care, and only then the framework — with training before you widen access.

// INCOMING SITREP

The episode covers how to think about choosing. The SITREP is the reference guide — each framework, who it fits, who it doesn't, and where to find it.

ACCESS THE BRIEF »

TRANSMISSION LOG //

Every organization is somewhere in the same scramble. How do we use AI securely. Is there a process for doing this correctly. Is there a framework, or a certification, or something we are supposed to point at.

And inside most of these companies the honest answer is that nobody is sure — or nobody is sure which one to commit to.

This week we laid the options out. What is actually on the table, what each one does, and how to choose between them.

The Pressure Is Coming From Everywhere

Our CISO’s read on why this landed on every desk at once: the pressure is not coming from the security team. It is arriving from four directions simultaneously.

The board and executives want market advantage, efficiency, and lower operating costs. Customers no longer wait for new features — they want them as soon as they think of them. Procurement is reviewing vendor contracts and finding AI in products that did not have it last year.

And the regulators are moving unevenly. States are already publishing their own AI policies, including requirements that if AI was used in any way to produce a deliverable, that has to be acknowledged in contracts and disclosures. Procurement teams are now checking for exactly that language.

Federal policy is being written as fast as it can be written, and it is still behind. AI is already in the products.

“It is right now the wild, wild west.”

One structural observation worth sitting with: our CISO is seeing AI land on the CEO’s desk, because nobody is certain where else to put it. It is a business decision with significant security and privacy consequences attached, which means it does not fit cleanly into any existing owner.

What Is Actually on the Table

ISO 42001 — The Certifiable Standard

Published in 2023, ISO/IEC 42001 was the first artificial intelligence management system standard. It follows the Plan-Do-Check-Act model and is deliberately structured to align with ISO 27001 and ISO 9001.

That alignment is the practical point. An organization already holding 27001 — common for anyone doing international business — is not starting over.

“It makes it really easy to go from one certification to the next, because you’re only looking for the difference.”

This is also the one showing up in contracts right now.

NIST AI RMF — The Method

On the US side, the NIST AI Risk Management Framework is not a certification and not mandatory. It comes out of the Department of Commerce and it is a framework you follow rather than one you pass.

Our CISO walked the sequence: governance first, because governance is where accountability lives. Then map where AI is being used — that is your inventory. Then run a risk assessment against it. Then manage and prioritize what you found.

“Even without being able to get NIST certified, it gives you a foundation — something to follow versus just blindly trying to figure things out.”

For government contractors that last point carries extra weight. NIST frameworks get referenced into federal requirements as a matter of course, and organizations already running NIST RMF for security will find the AI RMF incremental rather than new.

COBIT and the ISACA Credentials

COBIT, from ISACA, is a governance and audit framework rather than an AI-specific one. Auditors use it as their structure, which means it is the lens through which your AI governance is likely to get examined whether you adopt it or not.

Our CISO was blunt about the implication:

“You can’t adopt COBIT for AI. That’s not how it works. COBIT is an auditing framework — you need to understand it, because when auditors come in, that’s where they’re going to look for risk.”

On the credentials, she drew the line clearly: these are extensions of the security and audit certifications your people already hold. AAISM — Advanced in AI Security Management is the AI counterpart to CISM, for the security leader. AAIA — Advanced in AI Audit is the counterpart to CISA, for the auditor who now has to audit AI systems. ISACA has since added a third, AAIR — Advanced in AI Risk, for the risk side.

The underlying argument: a framework nobody in the building is qualified to operate does not survive its first year.

GRAICE — The Meta-Framework

The newest entry, and the one our CISO has been working with directly through the Global Council for Responsible AI.

GRAICE takes the existing frameworks and standards — including physical security and safety frameworks — and maps them into a single unified structure. Anyone who has done compliance work knows why that matters. You always end up mapping every framework against every other one by hand, maintaining the crosswalks yourself, and producing the same evidence in three formats. GRAICE does that work up front.

But the reason it exists is the second thing it adds:

“If you look at all of the other ones, there’s no ethics. It’s all about security. AI doesn’t have ethics. So somebody has to have ethics.”

That is the gap. The other frameworks govern security, risk, and management process. None of them asks whether a use of AI is one you should be making. GRAICE builds the human-oversight role into the governance architecture as a defined position rather than leaving “human in the loop” as a principle nobody owns.

Disclosure: our CISO works with the Global Council for Responsible AI. We would rather say that plainly than have you find it out.

Don’t Start With the Framework

This was the most useful ninety seconds of the episode, and it was the answer to a simple question: given all these options, which one do I pick?

Her answer was that this is the wrong first question.

“You can’t protect what you don’t know. You can’t put a standard or framework around something you don’t know.”

Three things come first, in order:

  • The AI inventory — what AI is in your environment, who is using it, who has access, and what they are using it for. Including third parties, because vendors have been quietly embedding AI into software you already bought.
  • One accountable owner — a single person responsible across the whole organization, not per department, because everyone has their own opinion about what AI should be used for and what is too risky.
  • The risk register — a documented risk assessment against what the inventory found, with owners assigned.

Only then does the framework question become answerable. As she put it, one framework might be overkill for your organization and another might not be enough — and you cannot judge that until you know what you are scoping it against.

This is where the episode hands off. We cover how to think about choosing here; the companion SITREP is the reference guide. AI Governance Frameworks You Can Implement Today breaks down each framework in detail — what it is, who it’s for, who it’s explicitly not for, and where to find the official source — including the full GRAICE architecture and our reasoning for recommending it.

The Part Nobody Wants to Say Out Loud

Actual asked the CISO to step back from the technical specifics and just react to the whole thing. Her answer went somewhere more uncomfortable than expected.

“The board says that. But the boots on the ground don’t say that. The question really is — is your organization actually getting benefit from AI?”

She has watched organizations spend hundreds of thousands, sometimes millions, on AI proof-of-concept work that returns a fraction of what was projected. Somebody has a good idea, the pressure is to get to market before a competitor does, the money gets spent, and the gain turns out to be modest.

Actual compared it to the era when every company decided it needed a website because every company had a website. The CISO extended the comparison and then broke it: a badly built website was embarrassing. It did not expose anything.

What Happens When You Just Turn It On

This is the sharpest operational point in the episode, and it is the one we would put in front of any government contractor.

“A person can only see so much data. AI sees everything at once.”

An employee with too much access still has to know a document exists and go looking for it. That friction was never a control, but it functioned as one. Turn on an assistant that reads everything a user can reach, and the friction is gone.

She has watched multiple organizations enable Copilot to make their teams more effective, and then discover staff surfacing highly sensitive information — because the permissions were never right. People change roles and keep old access. Accounts are over-permissioned. Orphaned accounts sit on systems after the person has left.

The result is predictable: turn Copilot back off, write policies, review everything, start over.

And the second failure mode she named is worse than exposure. AI does not second-guess itself. It does not stop to ask whether it should. She referenced the incidents where AI agents have deleted production code bases — not maliciously, but because the system decided the task needed doing and had no mechanism for doubt.

Train Before You Widen Access

Her prescription was direct. If the plan is to hand AI to everyone so they can be more effective, train them first. Publish what AI may be used for and what it may not. Then widen access.

“Instead of just handing them — it’s like giving a bunch of people construction equipment and saying, hey, go do something with it.”

Marching Orders

1. Build the AI Inventory and Name the Owner

Every AI system in the environment, who uses it, what it touches — including the AI your vendors embedded into software you already run. Then name one person accountable across the whole organization.

2. Find Out Who Outside Your Company Cares

Customers, procurement, regulators, insurers. Know what those requirements are and on what timeline, because in a contracting environment the answer is frequently “sooner than you think” and it arrives as contract language rather than a warning.

3. Pick the Framework and Train Your People

Only once you know what you have and who is asking. Then train before you widen access — the policy has to exist before the capability does.

Execute the Standard

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

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.

And none of it starts with the framework. It starts with knowing what you are already running.

If you need an outside read on your AI governance posture, help 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.

Mission success starts with the inventory.

Execute the standard.

// DECODED TRANSCRIPT

Access the full text logs of this transmission for compliance and review purposes.

SECURE FEED //

APPLE PODCASTS SPOTIFY YOUTUBE

JOIN "THE WATCH" //

Receive critical SITREPs, Industry Alerts, and Threat Indicators sent directly to your inbox.

By submitting this form, you agree to our Terms & Conditions and Privacy Policy.

RELATED INTEL //

024 The CMMC Phase II Suspension: What Defense Contractors Need to Do Now

ACCESS TRANSMISSION »

023 Ransomware-as-a-Service, AI Phishing, and Everything You Need to Know About CMMC Level 3

ACCESS TRANSMISSION »

022 OT Attacks, AI Risk, and Data Security in the Oil and Gas Industry

ACCESS TRANSMISSION »

HOW WE WATCHUR6 //

AUDIT READINESS ASSESSMENT REMEDIATION LIAISON

SILENCE THE NOISE. AMPLIFY THE SIGNAL.

INTELLIGENCE IS USELESS IF YOU AREN'T LISTENING.

Join The Watch to receive New Episode Alerts, Strategic Breakdowns, and Guest Intel delivered to your inbox.