Skovos guide

AI governance committee playbook for health systems

By Will Haver. Published and updated November 9, 2026.

What does a hospital AI governance committee do?

A hospital AI governance committee oversees how AI is selected, deployed, monitored, and retired. It assigns an owner to every tool, sets conditions for use, reviews performance and safety events, and reports to the board. Its decisions count only when they change an access right, a contract term, or a monitoring plan.

This playbook shows how to stand one up in 90 days, or fix one that only meets. It builds on our definition of a hospital AI governance committee. The 90-day plan, risk tiers, and meeting cadence below are Skovos recommendations. Guidance does not require them.

Why committees stall

Most committees do not fail at the first meeting. They fail later, in three ways.

Federal data point to the second problem. In 2024, 74% of hospitals reported that multiple entities were accountable for evaluating predictive AI, according to the ASTP/ONC data brief. The same brief reports that 71% of hospitals used predictive AI integrated with their EHR in 2024, up from 66% in 2023. It also reports that 82% evaluated predictive AI for accuracy, 74% for bias, and 79% did post-implementation evaluation or monitoring.

Who sits on the committee

The Joint Commission and Coalition for Health AI (CHAI) guidance on the Responsible Use of AI in Healthcare (RUAIH) says the team "could include" people with the expertise below. It does not require a standalone team. It does call for "a designated individual(s) with appropriate technology expertise, ideally in AI if available." Read the full text in the RUAIH guidance.

RoleWhy they are neededSource
Executive sponsorGives the committee authority and a path to the board.RUAIH element 1: executive leadership
Clinical chair or CMIOJudges clinical risk and use in care.RUAIH element 1: clinical and operational expertise; designated lead
Nursing leaderSees how tools change bedside work.RUAIH element 1: clinical and operational expertise
Quality and patient safetyOwns event review and monitoring.RUAIH element 1: safety and incident reporting
IT and integrationKnows where each tool connects and who can reach it.RUAIH element 1: information technology
CybersecurityReviews access, data flows, and vendor security.RUAIH element 1: cybersecurity and data privacy
PrivacyChecks data use and patient disclosure.RUAIH element 1: cybersecurity and data privacy
Compliance and legalMaps rules to each tool and reviews contracts.RUAIH element 1: regulatory and ethical compliance
ProcurementPuts conditions into purchase terms and renewals.Skovos recommendation
Data science or analyticsTests accuracy, bias, and drift.RUAIH element 1: technology expertise
Patient or community voiceSpeaks for people the tools affect.RUAIH element 1: impacted populations

Keep the full committee small enough to decide. Move day-to-day intake and monitoring to a working group.

Five decisions the committee owns

A committee decision counts only when it changes an access right, a contract term, or a monitoring plan. Each decision below needs evidence and a place where it lands.

DecisionQuestion it answersEvidence it needsWhere it must land
IntakeShould we adopt this tool, and at which risk tier?Intended use, vendor responses, data flows, tier assignmentRegistry entry with a tier
ConditionsWhat data, actions, and approvals apply?Approved scopes, human-approval points, contract termsAccess scopes and contract
OwnershipWho answers for this tool?Named business and technical owners, backupsRegistry owner field
MonitoringHow often do we recheck it, and what triggers a review?Baseline metrics, review dates, vendor change noticesMonitoring plan
RetirementWhen do we restrict, pause, or retire it?Stop criteria, safety event history, owner requestAccess revoked, contract notice

For vendor tools, the Health Sector Coordinating Council third-party AI risk guide (April 2026) includes appendices on a governance policy, inventory management, and a RACI matrix. Use them as a starting point for ownership and inventory fields. For intake detail, see AI vendor intake and review in hospitals.

Risk tiers for intake and monitoring

Not every tool needs the same scrutiny. RUAIH element 4, Ongoing Quality Monitoring, says: "Tools that inform or drive clinical decisions should be checked more often; tools that simply help with administrative or documentation tasks can be checked periodically." It adds that organizations should use structures they already have, such as quality, patient safety, and compliance, where possible.

TierReview depthSign-offRevalidation triggerMonitoring cadence
Tier 1: informs or drives clinical decisionsFull review: clinical, safety, privacy, security, biasCommittee voteAny vendor update, model change, or safety eventMonthly
Tier 2: clinical support and documentationStandard review by the working groupWorking group, reported to committeeMaterial vendor update or workflow changeQuarterly
Tier 3: administrative and operationalLight review: data, access, contractNamed owner, logged by the working groupChange in data or scopeTwice a year

The review depths, sign-offs, triggers, and cadences in this table are Skovos recommendations, not requirements. Adjust them to your size and risk appetite.

The first 90 days

WindowMilestoneWhat to do
Days 1 to 30Charter, chair, members, inventory startAdopt the charter. Name the chair and members. Start the inventory with every tool clinicians, staff, and vendors already use.
Days 31 to 60Intake form, risk tiers, first two reviewsPublish an intake form and the three tiers. Review one clinical tool and one administrative tool end to end.
Days 61 to 90Monitoring owners, board update, safety event pathName a monitoring owner for each live tool. Send the first board update. Agree how staff report an AI safety event.

Start the inventory early, because it surfaces shadow AI. The NIST AI Risk Management Framework, released January 26, 2023, makes the same point in GOVERN 1.6: "Mechanisms are in place to inventory AI systems and are resourced according to organizational risk priorities." See the NIST AI RMF page and the framework text. A registry holds the result. Our AI agent registry guide lists the fields.

Charter template

Copy this text into a document and edit the brackets. It is a starting point, not a legal document.

AI GOVERNANCE COMMITTEE CHARTER (template)

1. Purpose
The committee oversees how [Organization] selects, deploys, monitors, and retires AI tools, so that use is safe, lawful, and accountable.

2. Scope
- Clinical AI that informs or drives care decisions
- Administrative and operational AI
- AI embedded in vendor products and EHR features
- Autonomous AI agents and the systems they can reach

3. Membership
Chair: [name, role]. Executive sponsor: [name, role].
Members: clinical, nursing, quality and patient safety, IT, cybersecurity, privacy, compliance and legal, procurement, data science, patient or community voice.

4. Quorum
[Chair or delegate plus at least half of voting members, including one clinical and one compliance or privacy member.]

5. Decision rights
The committee decides intake, conditions, ownership, monitoring, and retirement. Each decision is recorded with its evidence and reflected in the registry, access scopes, contract terms, or monitoring plan.

6. Escalation
Any member or owner may stop a tool they believe puts patients at risk. The chair reviews every stop at the next meeting. Unresolved disputes go to [executive sponsor].

7. Board reporting
The committee reports to [board committee] at least [quarterly]: inventory by tier, decisions, open conditions, safety events, stops and retirements.

8. Review cycle
The committee reviews this charter every [12] months and after any serious AI safety event.

Meeting cadence and a standing agenda

Skovos recommends a monthly meeting of 60 minutes. Use the same agenda each time so evidence builds up.

  1. New intake requests and tier assignments
  2. Conditions to verify before go-live
  3. Monitoring exceptions
  4. Safety events and near misses
  5. Vendor changes and updates
  6. Retirements and stops

Reporting to the board

RUAIH element 1 states: "The fiduciary board of the healthcare organization should be regularly updated on AI use and its outcomes in healthcare." Keep the report to one page. Include:

Evidence to keep

Records turn a committee into a program you can show someone. This table maps each record to the RUAIH element and the NIST AI RMF subcategory it supports. The mapping is our reading, not an official crosswalk.

RecordRUAIH elementNIST AI RMF
Meeting minutes and attendanceElement 1GOVERN 2.1
AI inventoryElement 1GOVERN 1.6
Intake decisions and conditionsElement 1GOVERN 2.1
Monitoring results and review datesElement 4MANAGE 2.4
Safety events and stopsElement 5MANAGE 2.4
Training recordsElement 7GOVERN 2.1

NIST MANAGE 2.4 asks for mechanisms "to supersede, disengage, or deactivate AI systems" that perform badly. Plan that path before you need it. See how to stop an AI agent in a hospital.

For a view of how another organization structures this, see Duke Health's ABCDS Oversight program, and CHAI's AI Governance Framework, which has four core domains: AI Policy, Organizational Structures, Organizational Resources, and Organizational Processes. These are examples, not endorsements of this playbook.

Section 1557 duty

Under 45 CFR 92.210(b), a covered entity has "an ongoing duty to make reasonable efforts to identify" uses of patient care decision support tools that use input variables measuring race, color, national origin, sex, age, or disability. Under 92.210(c), it must "make reasonable efforts to mitigate the risk of discrimination" for each tool identified. The committee should own that list. Read the regulation text and our Section 1557 guide. Not legal advice. Ask counsel how the rule applies to you.

How Skovos supports a committee

Skovos turns committee decisions into enforced settings. The owner and scopes recorded for each agent drive its runtime permission checks, a recall stops it, and the audit trail gives the committee evidence of what each agent actually did. This guide does not depend on Skovos, and following it does not make an organization compliant or certified. For the voluntary certification, see Joint Commission responsible use of AI certification.

Frequently asked questions

Does a hospital need a separate AI governance committee?

Not necessarily. The Joint Commission and CHAI guidance says the governance structure does not need to be its own standalone team. Many hospitals extend an existing quality, patient safety, or clinical decision support committee, as long as AI has named owners and a standing agenda slot.

Who should chair a hospital AI governance committee?

The guidance does not name a chair. A chief medical information officer, chief quality officer, or chief data or AI officer is a common choice. The guidance does ask for a designated individual with appropriate technology expertise, ideally in AI if available, to lead implementation.

How often should an AI governance committee meet?

The guidance sets no required cadence. Skovos recommends a monthly full committee meeting for policy, exceptions, and board reporting, plus a smaller working group that handles intake requests as they arrive.

What should an AI governance committee report to the board?

The Joint Commission and CHAI guidance says the fiduciary board should be regularly updated on AI use and its outcomes. A useful report covers the inventory count by risk tier, decisions made, open conditions, safety events, and any tools stopped or retired.

Does the committee cover vendor AI embedded in the EHR?

Yes. The guidance says the governance structure covers third-party, internally developed, and AI-embedded tools. Embedded features that arrive through a vendor update belong in the inventory and should trigger a review.

What is the first thing a new committee should do?

Write the charter and start the inventory. A charter gives the group authority and decision rights. An inventory shows which AI tools are already in use, including ones no one approved. Both can start in the first 30 days.

Is this legal advice?

No. This guide is general information, not legal advice. Ask counsel how Section 1557 and other rules apply to your organization.

Sources

Related reading

Get a free AI governance assessment

Send a vendor list and an agreement index. We return a vendor-agreement gap register like the sample, free. No patient data needed.

Request a governance walkthrough   See a sample assessment

Start free

A map of every AI agent in your health system.

We map who owns each agent, what it can access, and what evidence you could show your board or a surveyor.

Free. Asking takes 20 seconds. Please do not send patient information.