November 19, 2026 · Jermaine Barker
Why Your Clinical Staff Are Already Using AI (And What to Do About It This Quarter)
Shadow AI is already in your health system, whether you approved it or not. Here is a practical way to find it, govern it, and turn it into a sanctioned use case within 90 days.

Last year I sat in a room with a hospital operations director who told me, with total confidence, that her organization did not use generative AI yet. Twenty minutes later, a nurse manager mentioned she had been pasting shift notes into a public chatbot to draft handoff summaries. Every week.
Nobody was being reckless. She was tired, the tool was free, and it worked.
That is the real state of healthcare AI right now. The question is not whether your staff are using it. The question is whether you know where.
The gap between policy and practice
Most healthcare organizations I work with have one of two things: no AI policy, or a policy written by legal that nobody has read. Neither one changes behavior.
Meanwhile, the tools are a browser tab away. Clinicians use them to summarize literature. Billing staff use them to draft appeal letters. Administrators use them to clean up spreadsheets. Some of those uses are harmless. Some put protected health information into systems with no business associate agreement.
You cannot govern what you cannot see. And a blanket ban does not make it visible. It pushes it underground.
Start with an inventory, not a prohibition
In every engagement, the first useful step is the same. Find out what is actually happening. Do it without blame.
Here is the approach that works:
- Ask, don't audit. A short, anonymous survey by department. What tools do you use? For what tasks? What data goes in?
- Check the network. Your IT team can see which AI services are being reached from your environment. This takes days, not months.
- Talk to the front line. Ten conversations with managers will tell you more than a hundred-page report.
You are looking for patterns. Where is the pain? Staff do not adopt tools for fun. They adopt them because a workflow is broken.
Sort what you find into three buckets
Once you have the inventory, classify each use:
- Stop now. Any use that puts PHI into a non-approved tool. This one is not negotiable. Fix it this week, and give people an alternative so they are not left with the broken workflow.
- Allow with guardrails. Low-risk uses with clear rules: no patient data, human review before anything leaves the building, approved tools only.
- Promote to a real project. Uses that show real value and repeat across teams. These are your candidates for a sanctioned, governed deployment.
The third bucket is where the opportunity lives. Your staff just ran a free pilot for you. They showed you which workflows are worth fixing.
Governance is what lets leadership say yes
I hear this a lot: governance will slow us down. I have seen the opposite.
When a CMO or compliance officer cannot answer basic questions, such as what data goes in, who reviews the output, and what happens when it is wrong, the answer is no. Or worse, silence. The project sits in review for six months.
When those answers are written down, approval gets faster. Guardrails are not the brake. They are what make a yes possible.
At minimum, every sanctioned use should have:
- A named owner who is accountable for outcomes
- A defined data boundary: what can and cannot be entered
- A human review step that is real, not a checkbox
- A way to log and audit what the system did
- A clear condition for pausing or shutting it down
That is a one-page document. Not a 40-page framework.
Ship one governed use case in 90 days
About 80% of AI pilots never reach production. In healthcare the reasons are consistent: unclear ownership, no data boundary, and no path from experiment to operations.
So pick one workflow from your third bucket. Choose something administrative and bounded, like prior authorization drafting, referral summarization, or internal policy search. Avoid anything that touches diagnosis or treatment decisions as your first move.
Then run it properly:
- Days 1-30: Define the workflow, owner, data boundary, and success measure.
- Days 31-60: Build and test with a small group. Review every output.
- Days 61-90: Go live with monitoring, and document what you learned.
One working, governed deployment does more for your organization than a year of strategy decks. It builds trust with compliance. It gives your staff a better option than the free chatbot. And it gives you a template for the next one.
Where to begin
If you are not sure what your staff are already doing, or where your biggest exposure sits, start with a baseline. Our free AI Readiness Assessment walks through governance, data, and workflow readiness in a few minutes and shows you where the gaps are.
If you already know your priorities and need a structured path to production, the ASCEND framework is how we take organizations from inventory to a governed, running deployment.
Either way, start with what is true today. Your staff already did.