An AI governance committee is the group that decides which AI uses your organization will approve, under what controls, and who is accountable when something goes wrong. Most enterprises now have copilots, chat assistants, and early agents in use. However, many still have no single forum that owns those decisions.
This guide explains how to charter an AI governance committee that moves quickly without becoming a bottleneck. It draws on the NIST AI Risk Management Framework, ISO/IEC 42001, and the federal model set by OMB.
Why an AI governance committee matters now
AI risk no longer sits inside one team. For example, a copilot touches data governance. An agent that files tickets touches security and identity. A model that scores applications touches legal and civil rights. As a result, no single department can approve AI use alone.
In fact, the frameworks agree on this point. The NIST AI RMF GOVERN function calls for documented roles, responsibilities, and lines of communication for managing AI risk. It also says executive leadership takes responsibility for decisions about AI risk. Likewise, ISO/IEC 42001 sets requirements for an AI management system, including leadership commitment and defined roles.
Federal agencies offer a useful model. For instance, OMB memorandum M-25-21, issued in April 2025, directed each CFO Act agency to convene an AI Governance Board within 90 days. In that model, the board is chaired at the Deputy Secretary level, with the Chief AI Officer as vice-chair. In addition, members come from IT, cybersecurity, data, budget, legal, privacy, and civil rights offices. Of course, private companies are not bound by that memo. Still, its structure is a sound template.
Who sits on the AI governance committee
In general, keep the core group small and senior enough to make decisions. A typical membership looks like this:
- Executive sponsor (chair): a C-level leader who can settle trade-offs between speed and risk
- AI lead (vice-chair): the person who runs the AI program day to day
- Chief Information Security Officer: identity, access, and threat modeling
- Data governance lead: data classification, quality, and lineage
- Legal and privacy: contracts, regulatory mapping, and privacy reviews
- Enterprise architecture or platform owner: approved tools, integration patterns, and cost
- PMO or portfolio lead: intake, prioritization, and tracking against business value
Typically, business unit leaders join when their use cases are on the agenda. In addition, HR and procurement should attend when AI affects workers or vendor contracts.
What the charter should decide
Above all, a charter keeps the committee focused. Specifically, it should answer five questions in writing.
- Scope: Which AI uses need review? Also include vendor features turned on inside existing software, not only custom builds.
- Risk tiers: How are uses classified? For example, low risk for drafting help, and high risk for anything that affects rights, safety, money, or regulated data.
- Decision rights: What can the AI lead approve alone, and what must come to the full committee?
- Minimum controls by tier: What testing, human oversight, logging, and monitoring each tier requires.
- Cadence and reporting: Finally, decide how often the committee meets and what it reports to leadership.
For high-risk uses, M-25-21 lists practices worth borrowing: pre-deployment testing, an impact assessment, ongoing monitoring, trained human oversight, a way to appeal outcomes, and feedback from affected users. Similarly, the NIST AI RMF calls for an inventory of AI systems resourced by risk.
A simple RACI for AI decisions
For an AI governance committee, clear roles prevent the two common failures. Otherwise, nobody approves anything, or everyone assumes someone else did.
- Propose a use case and success measures: the business owner is accountable, and the AI lead does the work.
- Assign the risk tier: the committee is accountable, with input from security and legal.
- Approve data access and identity scopes: security is accountable, while the AI lead configures access.
- Run pre-release testing and red-team reviews: security is accountable, and the AI lead runs the tests.
- Approve a high-risk use for production: the committee is accountable after the AI lead presents the evidence.
- Monitor incidents, drift, and cost: the AI lead is accountable, and security handles day-to-day monitoring.
- Map regulations and keep audit evidence: legal and risk are accountable, with support from the other roles.
Of course, adjust the roles to fit your organization. However, keep exactly one accountable party for each decision.
The first 90 days
Days 1 to 30: Inventory and charter
- First, inventory every AI tool, copilot, and agent in use, including features vendors enabled by default.
- Next, approve the charter, risk tiers, and decision rights.
- Finally, name an accountable business owner for each AI use.
Days 31 to 60: Controls and intake
- First, publish an AI acceptable use policy that staff can read in five minutes.
- Add an AI intake form to your existing portfolio process instead of building a separate one.
- Then define minimum controls for each tier, including identity, logging, and human approval steps.
Days 61 to 90: Operate and report
- First, review the first wave of high-risk use cases against the new controls.
- Next, run one incident drill with a realistic AI failure scenario.
- Finally, report value, cost, incidents, and coverage to the executive team.
Then move the AI governance committee to a steady cadence. Usually, monthly meetings work for most organizations, with a fast path for low-risk approvals between meetings.
Common mistakes to avoid
- Making the committee the only path to yes. Rather, delegate low-risk approvals so the group can focus on high-risk decisions.
- Reviewing models but not actions. Because AI agents act on systems, review the permissions and tools they use, not just the output.
- Running AI intake outside the portfolio. Instead, route AI requests through the same intake and prioritization as other investments.
- Skipping the inventory. After all, you cannot govern what you cannot see.
For agent-specific controls, see our article on AI agent regulations and the governance floor.
Frequently asked questions
How is an AI governance committee different from an AI center of excellence?
The committee makes risk and approval decisions. In contrast, a center of excellence builds skills, patterns, and shared platforms. Moreover, many organizations need both, and the AI lead often connects them.
How often should the committee meet?
In general, monthly is a good default. Meanwhile, a delegated fast path can handle low-risk requests in days rather than waiting for the next meeting.
Do small and mid-size organizations need one?
Yes, though it can be lighter. Even a group of four or five leaders with a one-page charter is better than no clear owner for AI risk.
How VTPMO can help
VTPMO helps organizations stand up AI governance that fits how they already plan and deliver work. We design charters, risk tiers, intake processes, and controls that connect to portfolio management and Microsoft 365. Explore our AI adoption and governance services and data governance services, or contact us to start your first 90 days.

