Who Actually Has to Approve an AI Governance Platform

An AI governance platform rarely gets approved by one person. In our work across GRC accounts, the buying committee commonly includes the CISO, General Counsel, a data or AI lead, the audit function, and whoever controls the budget, and each one tends to judge the platform by a different standard. A pitch that only speaks to one of them is usually the one that stalls.

Key Takeaways

• An AI governance platform purchase typically draws in a cross-functional committee, not a single buyer: the CISO, General Counsel, a data or AI lead, the audit function, and the budget owner, increasingly the CFO.
• These roles aren’t rubber-stamping the same pitch. Each one is scoring the deal against a completely different risk.
• General Counsel stalls more deals than any other role, and it’s rarely personal. Minimizing exposure is the job.
• Budget for AI governance doesn’t live anywhere yet. It’s shifting from innovation funds to scrutinized operating expenses.
• Audit is often the easiest “yes” in the room if the vendor shows up with one specific artifact: a risk-classified use-case inventory.

Who actually sits on this committee, and why isn’t it a one-person decision?

A long list of names sounds like a lot for a software purchase. But an AI governance platform isn’t ordinary SaaS. It asks an organization to hand off a piece of itself, a security judgment, a legal position, a financial control, to something that can now act on its own. McKinsey partner Rich Isenberg put it more bluntly than most vendors would like: “Agency isn’t a feature; it’s a transfer of decision rights.” Once a system can act instead of generate text, everyone who owns a piece of that decision gets a vote before it goes live.

Here’s who typically ends up in the room, and what each one is really weighing:

PwC’s risk leadership team has found that the clearer these responsibilities are drawn, the faster approvals move. Where they’re blurry, every sign-off turns into its own negotiation. The two hardest yeses to earn, and the ones vendor pitches get wrong most often, belong to the CISO and General Counsel.

Why does a CISO’s yes not mean the General Counsel’s yes?

Ask a CISO what would make them comfortable, and they’ll describe a threat model: how the system behaves under attack, what happens when someone feeds it a poisoned prompt, and whether the guardrails actually hold. Ask a General Counsel the same question, and you get a different list entirely: is this company a “provider” or a “deployer” under the EU AI Act, and can the accountability framework prove it.

Neither list is wrong. They’re just not the same list, and that’s easy to miss when a vendor builds one polished packet and assumes it covers everyone in the room.

The law doesn’t make splitting the difference any easier, because it was never written around job titles in the first place. Obligations under the EU AI Act attach to what an organization does, and the Act names no CISO anywhere. Article 17 asks providers of high-risk systems to maintain an accountability framework, not to appoint a single accountable person, which just pushes the question back to the org chart: Is this company even a provider, and what proves it? Answering that belongs to legal; proving the system survives an adversarial test belongs to security. Gartner’s December 2025 CISO guidance draws the same line: CISOs should influence AI governance broadly, but lead only the cybersecurity assurance piece, not sign off on the whole program.

Which is exactly why the deals that move are the ones where a vendor shows up with both packets ready, not just the one the sales team is proudest of.

Where does the money actually come from?

Increasingly, the CFO’s desk. Not because finance suddenly develops opinions about responsible AI, but because the money itself has moved. Early pilots got funded out of innovation and transformation budgets, the kind nobody scrutinizes too closely because they’re meant to be exploratory. Once a pilot becomes permanent infrastructure, the spending migrates to ordinary operating expense, and ordinary operating expense gets audited like everything else.

The Controllers Council’s reporting makes the shift plain: finance leaders are becoming the primary gatekeepers of AI spend simply because no one else is positioned to ask whether a given tool is earning its keep. That’s a hard room for a vague pitch to survive. A slide about “enterprise risk reduction” won’t clear it, but a specific number will, hours of audit time saved, an incident avoided on one named process, something the CFO can actually hold onto.

The awkward part for sellers: there’s no established budget line for AI governance yet, unlike for security tools or sales software. Until a company decides where that spending lives, every pitch to the CFO starts from zero.

So, which role actually kills the most deals?

General Counsel blocks more deals than any other role, and it’s rarely personal. Legal’s job is to minimize exposure, and caution is just what that looks like from the outside. What turns caution into a bottleneck is timing: in an area still being written in real time, full legal review stops being an occasional extra step. It becomes the default gate that everything has to pass through.

The pattern holds across GRC accounts here at Purplepatch Services, too. Deals clear fastest when legal answers the provider-or-deployer question on day one, not after every other approval is already done.

PwC’s research on board oversight backs that up from a different angle: governance works better with a seat at the table from the start than a review of a decision that’s already been made.

What is the one document that makes audit an easy yes?

Of the roles involved, audit might be the simplest signature to get, because it only wants one thing: a use-case inventory that ties every AI system to a risk classification and flags whether it touches anything Sarbanes-Oxley cares about.

PwC’s guidance on AI and audit readiness is specific about what “proving” a framework looks like in practice: model risk assessments, reviewer sign-offs, exception logs, and evidence that a human actually reviewed the output before it reached a financial report. Not a policy that says this should happen, but proof that it did.

Almost nobody builds this in from the start. Most governance platforms treat it as a custom report request instead of a default feature, which is strange given how much leverage it carries with the one stakeholder who can otherwise sit on a deal indefinitely.

Frequently Asked Questions

Does a smaller company really need every one of these approvals?

Not necessarily as separate individuals for each role. Smaller companies compress roles instead of skipping them; an AI lead’s job might fold into IT, but the same underlying questions still get asked by somebody.

What happens if the CFO only gets looped in at the end?

The pitch usually restarts because a CFO’s first question is about the return on a specific process, not enterprise-wide value.

Does a lightweight tool, like an internal chatbot, need the same level of sign-off?

Not at the same intensity. Risk frameworks like the EU AI Act scale scrutiny to the level of risk, so a low-risk tool clears committee faster than a high-risk decision-making agent. 

What’s the fastest way to get the committee aligned?

Bring legal and audit in before the CISO finishes the technical review, not after. 

Who actually holds the final signature?

Usually, no one person does. It’s a consensus across the committee rather than a single approval.

Most GRC vendor content is written for the CISO, because the CISO is the easiest buyer to picture. But the CISO’s yes is rarely the one holding up the deal. At Purplepatch Services, we build content and sales assets aimed at whichever role is actually in the room: security, legal, the data team, audit, or finance, so a deal doesn’t stall on a stakeholder nobody bothered to write for. If that sounds like the gap in your funnel, let’s talk.