Key Takeaways
- A healthcare AI governance checklist doesn't need to be long, six or seven questions cover almost every real risk
- The single most important question is whether the vendor will sign a Business Associate Agreement, not whether their marketing page says "HIPAA compliant"
- Ask about data training, retention, and offboarding before a tool goes live, not after
- Every checklist item should take a few minutes to answer, if it doesn't, the vendor probably can't answer it clearly
- Run this checklist on every tool, including the ones staff already started using without asking
Why a Checklist Beats a Policy Document
Most practices that try to build AI governance start by writing a policy. The policy sits in a shared drive, gets read once, and gets forgotten the next time someone finds a new tool. A checklist works differently because it attaches to the actual moment of decision, right when someone is about to sign up for something. It's short enough to run in five minutes and specific enough that a yes or no answer actually tells you something.
Our longer piece on what AI governance actually means covers the bigger picture, ownership, review cadence, staff policy. This one is meant to sit next to it as the thing you actually pull up when a new tool lands on your desk.
Question 1: Will the Vendor Sign a BAA?
Not "are they HIPAA compliant." Not "does their website mention HIPAA." Will they sign a Business Associate Agreement, in writing, before any patient data touches their system. A surprising number of tools that market themselves as healthcare-friendly don't actually offer a BAA on their standard plan, only on an enterprise tier that costs significantly more. Ask this question first, because if the answer is no, none of the other questions matter.
Question 2: Does the Tool Train on Your Data?
Some AI tools use whatever gets typed into them to improve their models. If patient information ends up in that training data, it can resurface in ways nobody can fully control later. A compliant vendor will have a clear, specific answer here, usually something like "enterprise accounts are excluded from training by default" or "we offer a contractual guarantee your data is never used for training." A vague answer, or one that takes three follow-up emails to pin down, is itself the answer.
Question 3: Where Does the Data Actually Live?
Ask where the data is stored, whether it's encrypted at rest, and whether the servers are in a jurisdiction that matters for your compliance obligations. This connects directly to what we cover in our HIPAA compliant AI guide, encryption in transit and at rest isn't optional, and neither is knowing which cloud provider is actually holding the data.
Question 4: What Happens If You Leave?
Before you're locked in, ask what offboarding looks like. Can you export your data? Will the vendor delete it on request, and how quickly? This sounds like a question for later, but it's much easier to get a straight answer before you sign than after you've been a customer for a year and want out.
Question 5: Who at Your Practice Actually Approved This?
This one isn't a question for the vendor, it's a question for you. If nobody can name the person who reviewed and approved a specific tool, that's the gap. It doesn't need to be a committee. In a small practice it's often just the owner. In a larger group it might be a fractional AI officer or a designated point person. The point is that someone's name should be attached to the decision, not "the team decided" or, worse, nobody decided and it just happened.
Question 6: When Was This Last Reviewed?
Vendors change their terms of service. Companies get acquired. A tool that had a clean BAA and a solid privacy policy a year ago might look different today. Set a recurring date, quarterly works for most practices, to run back through the tools already in use and confirm nothing has quietly changed underneath you.
What to Do When a Tool Fails the Checklist
Sometimes the answer isn't yes or no, it's "we don't know yet." If a vendor can't answer the BAA question clearly, or takes more than a day or two to get back to you with a straight answer, treat that as information. A vendor who is genuinely built for healthcare usually has these answers memorized, because they get asked constantly. Hesitation is a signal worth paying attention to, not something to work around.
If a tool is already in use and fails the checklist, the fix isn't necessarily to rip it out immediately. It's to pause new patient data from going into it while you sort out whether there's a compliant version of the same tool, a different vendor entirely, or a way to get the missing pieces (a signed BAA, a training opt-out) in writing.
Building This Into How Your Practice Actually Works
The checklist only helps if it gets used at the actual moment someone wants to try a new tool, not buried in an employee handbook nobody opens after their first week. Some practices post it somewhere staff will actually see it, a shared note, a printed card near the front desk computer. Others fold it into whatever approval step already exists for other purchases. What matters isn't where it lives, it's whether someone actually pulls it out the next time a new AI tool shows up.
Get Help Running This Across Your Practice
If you're not sure how many AI tools are currently in use across your practice, or whether any of them have actually been through a review like this, that's exactly the gap our free AI readiness assessment is built to find. It takes about an hour and gives you a clear picture of what's already in place versus what still needs a second look. For a broader rollout across a multi-location group, our healthcare AI consulting team can build this into a standing process instead of a one-time check.
Frequently Asked Questions
What is a healthcare AI governance checklist?
It's a short, repeatable set of questions a practice runs before approving any new AI tool, covering things like BAA status, data training policies, and who's responsible for the decision.
How long should reviewing a new tool against this checklist take?
Usually 10 to 15 minutes if the vendor has clear answers ready. Longer than that is often a sign the vendor doesn't have those answers well documented.
Does this checklist replace a full HIPAA compliance review?
No. It's a fast first pass to catch obvious problems early. Tools that pass still need the fuller compliance review before handling real patient data at scale.
What if a tool staff are already using fails the checklist?
Pause new patient data from going into it while you sort out whether a compliant version exists, rather than removing it immediately without a plan.
How often should existing tools be re-checked against this list?
Quarterly works well for most practices. Vendor terms and ownership can change without much notice, so a recurring check catches problems before they sit unnoticed for months.


