AI governance
What AI governance program does a large health system need?
A program-level view for health system leaders. NIST wording is quoted as read on 12 November 2026.
A large health system needs one program that covers six parts: an inventory of every AI system, risk tiers that set how closely each system is reviewed, a committee with decision rights, a review of vendor AI, monitoring after go live, and evidence that the work happened. NIST's AI Risk Management Framework Playbook, read on 12 November 2026, says in GOVERN 1.6: "Mechanisms are in place to inventory AI systems and are resourced according to organizational risk priorities." This page gives the program view. Each part that has its own page links to it. Kriv AI uses NIST as a reference and does not claim to conform to it.
plain answer
What does the program cover?
Think of the program as six parts that depend on each other. The inventory says what exists. Risk tiers say how closely each system is reviewed. The committee decides. Vendor review covers AI that arrives inside software you already run. Monitoring watches systems after go live. Evidence shows an outsider that all of this happened.
Three of the parts have their own page on this site: the committee, vendor review and evidence. The sections below say in a sentence or two what each one is, then link to it. Inventory, risk tiers and monitoring are covered here in more detail.
capabilities
Inventory
The inventory is the list of every AI system the health system builds, buys or switches on, and each entry names who answers for it. It should include AI features inside software you already run, not only tools bought as AI products.
NIST's GOVERN 1.6 says "Mechanisms are in place to inventory AI systems and are resourced according to organizational risk priorities." The second half matters: the effort spent on the list follows risk, so a system that touches patient care gets more attention than a scheduling aid.
Start with what you can find. Ask each department, check purchase records and list models your own team runs. Add a step so a new system cannot go live without an entry. Our page on shadow AI in a health system covers tools staff adopt unannounced.
Sources: NIST AI RMF Playbook, Govern (https://airc.nist.gov/AI_RMF_Knowledge_Base/Playbook/Govern), read 2026-11-12.
framework
Risk tiers
Risk tiers sort the inventory by the harm a wrong output could do. A tool that suggests a care decision belongs in a higher tier than one that drafts meeting notes. The tier then sets the review depth, the sign-off chain and how often the system is checked again.
NIST's MAP 1.5 says "Organizational risk tolerances are determined and documented." Tiers are one way to write that down. The health system decides what it will accept for each kind of use, records it, and applies the same rules to every system in that tier.
Keep the scheme small enough that staff use it. A short list of tiers with a plain test for each is more likely to be applied than a long scoring sheet.
Sources: NIST AI RMF Playbook, Map (https://airc.nist.gov/AI_RMF_Knowledge_Base/Playbook/Map), read 2026-11-12.
ai gap
Committee
The committee is the standing group that decides which AI systems may be used, under what conditions and who is told. It needs clinical, technical, legal and compliance voices, and it needs real decision rights, not only advice.
Our committee page goes through how health systems build one, who should be in the room and what the published examples did. Read it for membership, decision rights and the review process from intake to retirement.
For this program, the committee has one job above the others: it reads the inventory and the tiers, and it is the place where a high tier system gets approved or refused.
capabilities
Vendor review
Vendor review covers AI that comes inside software from a supplier, such as an EHR vendor. It is easy to switch on without the review a standalone AI product would get, so the program needs an intake step that catches it.
Our vendor page explains the split of duties. The vendor has obligations for the models it supplies, and the health system still owns local validation, configuration sign-off, monitoring and escalation for each model it turns on.
NIST's MAP 4 category says "Risks and benefits are mapped for all components of the AI system including third-party software and data." In program terms, a vendor model goes on the inventory and into a tier like any other.
Sources: NIST AI RMF Playbook, Map (https://airc.nist.gov/AI_RMF_Knowledge_Base/Playbook/Map), read 2026-11-12.
framework
Monitoring
Monitoring is the check on a system after it goes live. Approval at launch tells you little about how a system behaves months later, after the patient mix, the way staff write notes or the model behind a feature has changed.
NIST's MEASURE 2.4 says the system's functionality and behavior "are monitored when in production." The program should say who looks at the monitoring output, how often, and what happens when it shows a problem. Monitoring output that nobody reads does not protect a patient.
Higher tiers get checked more often and by more senior reviewers. Write down the cadence for each tier, and record each check so the committee can see the history.
Sources: NIST AI RMF Playbook, Measure (https://airc.nist.gov/AI_RMF_Knowledge_Base/Playbook/Measure), read 2026-11-12.
ai gap
Evidence
Evidence is what a board, an auditor or a regulator can be shown to prove the program runs: the inventory, tier decisions, committee minutes, review records and monitoring checks, each dated and attributed.
A program that exists only as a policy document is hard to defend. Our case studies page looks at two health systems that have published how they structure oversight, and it is a useful reference for what documented evidence looks like. It describes their programs, not Kriv AI client work.
Build the records as a by-product of the work. If keeping them is a separate chore, they will lapse.
engagement
How Kriv AI helps
We help a health system stand the program up alongside its own compliance, informatics and IT staff. If the health system runs its AI in AWS, AWS AI governance consulting for healthcare sets out the account-level questions that change a quote. We start by reading what already exists, then draft the inventory format, the tier scheme and the review routine, and fit them into committees you already have. Your team keeps and runs the program. We do not replace your committee or your clinical reviewers.
Kriv AI scopes and prices work after a short discovery call. We do not publish rates on this page.
Related resources
Continue exploring
Straight answers
Frequently asked questions about What AI governance program does a large health system need?
What does an AI governance program for a health system include?
Six parts: an inventory, risk tiers, a committee, vendor review, monitoring and evidence. This page explains how they fit together.
Does NIST require an AI inventory?
NIST says the AI RMF is intended for voluntary use (read 2026-11-12). GOVERN 1.6 says "Mechanisms are in place to inventory AI systems and are resourced according to organizational risk priorities" (read 2026-11-12).
What are risk tiers for?
They set how closely each system is reviewed. NIST's MAP 1.5 says "Organizational risk tolerances are determined and documented" (read 2026-11-12).
Who owns the program?
A standing committee with decision rights, with a named owner for each AI system. Our committee page covers how health systems build one.
Is vendor AI inside the EHR part of the program?
Yes. It goes on the inventory and into a tier like any other model. Our vendor governance page covers the split of duties.
How often should systems be checked after go live?
NIST's MEASURE 2.4 says a system's behavior is "monitored when in production" (read 2026-11-12). Set the cadence by tier and write it down.
What does this cost with Kriv AI?
Kriv AI scopes and prices work after a short discovery call. We do not publish rates on this page.
Talk to the team that would do the work
Bring your requirements to a working session with the person who'll actually deliver.
Book a Discovery Call