I’m going to tell you something most practice owners don’t want to hear: your staff is already using AI with patient information. Not “thinking about it.” Not “planning to.” Right now, today, someone at your front desk is pasting a patient’s details into ChatGPT to help draft a letter.
It’s not malicious. It’s the opposite. They’re trying to get through a packed schedule and a short-staffed week, and a free AI tool genuinely helps. But every time it happens, protected health information leaves your practice and lands in a system you don’t own, can’t audit, and can’t control.
This is called shadow AI: employees using AI tools that nobody approved, nobody secured, and nobody is watching. In healthcare, it is not a productivity story. It is a compliance story.
We manage IT for medical practices across Southwest Florida, and we can tell you exactly how this plays out. Here is what is actually happening, why it is a bigger problem than it looks, and what to do about it.
The instinct, when you hear staff are using ChatGPT, is to treat it like a training issue. Send a memo. Remind everyone to be careful. Move on. That instinct is wrong, and here is why.
When someone pastes a patient note into a free, consumer AI tool, that information is transmitted to a third party’s servers. Many consumer AI tools reserve the right to use what you type to train future versions of the model. In plain terms: your patient’s data can become part of someone else’s product. You did not sign a Business Associate Agreement. You did not get a guarantee about where the data is stored or who can see it. You got a text box and a fast answer.
HIPAA does not distinguish between an accidental disclosure and an intentional one. A breach is a breach. It does not matter that your medical assistant was just trying to save ten minutes. If protected health information went somewhere it should not have, you have an exposure, and you are the one who answers for it.
Here is the part that should really get your attention: you have no audit trail. If an auditor or an attorney asks how your practice governs AI use with patient data, “we told everyone not to” is not an answer. You cannot produce a record of what was shared, by whom, or where it went, because the entire point of shadow AI is that it happens outside your systems. You cannot secure what you cannot see.
This is not a fringe problem. The data is blunt.
Healthcare has been the most expensive industry for data breaches for fourteen years running. According to IBM’s 2025 Cost of a Data Breach research, the average healthcare breach reached $7.42 million and took the better part of a year to fully resolve.
Shadow AI makes that worse. The same research found that organizations with significant shadow AI use added about $670,000 to the average cost of a breach, and one in five breaches now involve shadow AI, a higher rate than incidents involving sanctioned, approved AI tools. These breaches also tend to take longer to detect, because nobody was watching the tool in the first place.
How common is the underlying behavior? A December 2025 survey of more than 500 healthcare workers found that 17% openly admitted to using unauthorized AI tools at work. That is the number people will admit to. The real number is higher, because the whole pattern is invisible by design.
And almost nobody is set up to handle it. In that same body of 2025 research, 97% of organizations that suffered an AI-related security incident had no AI access controls in place, and 63% had no AI governance policy at all. If your practice does not have an approved AI tool and a written policy, you are not the exception. You are the 63%.
Plenty of practices, once they realize this is happening, land on the same first move: ban it. No AI tools. Period. We understand the instinct. It is also the move most likely to fail.
Your staff did not start using AI because they are reckless. They started because they are busy. Healthcare runs on thin staffing and long days, and a tool that drafts a letter in fifteen seconds is a real, immediate relief. Take that away without giving them anything in its place and you have not solved the problem. You have pushed it further underground. Now they are using it on personal phones and personal accounts, where you have even less visibility than before.
A ban with no alternative is not a policy. It is a hope. The practices that actually get shadow AI under control do the opposite: they give their team a sanctioned tool that is safe to use, and then they write down the rules. Lead from the top, tell people clearly what they can and cannot do, and make sure the safe option is sitting right there when they need it. People will use it when it works.
If you are not sure whether this applies to your practice, here is where it tends to show up. It is rarely one rogue employee. It is small, reasonable shortcuts spread across roles.
Front desk staff drafting patient letters and portal messages. Billing and coding staff checking codes against real claim data. Medical assistants generating referral letters and prior authorization drafts. Clinical staff cleaning up or summarizing visit notes.
Every one of those tasks is legitimate work. And every one of them, done in a free public tool, sends protected health information outside your walls. The tasks are fine. The tool is the problem.
Here is the good news, and it is genuinely good news: this is a solvable problem, and the solution is not “ban AI and fall behind your competitors.”
The fix is to replace scattered, unmanaged consumer tools with one secure AI platform your whole team can actually use. This is an infrastructure decision, the same way moving off an aging server or tightening your email security is an infrastructure decision. It belongs with your IT, not in an HR reminder email.
A secure, organizationally managed AI platform looks nothing like a free chatbot:
This is the part practices miss when they are stuck in “AI is a risk” mode: done right, this is not just damage control. It is a productivity gain. The same platform that closes your compliance gap also gives your team pre-built workflows for the busywork that eats their day, things like intake processing, appointment follow-ups, and prior authorization drafts. AI here is not about replacing staff. It is about giving the staff you already have their time back.
One more thing worth saying plainly: this is not a six-figure enterprise project. The platform we deploy uses credit-based billing, so you pay for what your team actually uses, not a flat per-seat fee for every login. Most practices start with one use case, see it work, and expand from there.
You do not have to solve all of this at once. If shadow AI is a blind spot at your practice right now, here is a sane order of operations.
Most practices we talk to are somewhere in the 63% with no AI policy and no real idea how much AI is already in use across their team. The first step is simply finding out.
We built a short Shadow AI Risk Self-Assessment for exactly this. It takes a few minutes, it is free, and there is nothing to download or sign up for. You answer a handful of straight questions about how your practice handles AI, and you get a clear read on where your exposure is.
If you would rather just talk it through, that works too. Call us at (941) 315-2380 and we will give you an honest picture of where you stand. No pressure, no pitch.
It can be. When an employee enters protected health information into a free, consumer AI tool, that data is sent to a third party with no Business Associate Agreement and no guarantee of how it is stored or used. HIPAA does not distinguish between accidental and intentional disclosure, so if PHI leaves your control this way, your practice has a reportable exposure. The risk is not the AI itself, it is using a tool that was never set up to handle patient data.
You can, but a ban with no alternative usually backfires. Staff adopt AI because it helps them get through a busy day, and removing it without a replacement tends to push usage onto personal devices where you have even less visibility. A more effective approach is to give your team one secure, approved AI platform and a clear written policy, so the safe option is also the easy one.
A secure AI platform keeps your data isolated to your own instance and never uses it to train public models. It provides a full audit trail of AI activity, role-based access controls so staff only get the access they need, and independent security validation such as SOC 2 audits and penetration testing. Free consumer tools offer none of these by default, which is what makes them a poor fit for healthcare.
Less than most practice owners expect. The secure platforms we deploy use credit-based billing rather than a flat per-seat fee, so you pay for actual usage instead of paying for every login whether it is used or not. This is not a six-figure enterprise implementation. Most practices start with a single use case and expand once they see the impact.
Start by asking, without making it punitive, since most staff are using these tools to be helpful rather than careless. From there, an organizationally managed AI platform gives you real visibility into who is using AI and for what. Our Shadow AI Risk Self-Assessment is a quick, free way to gauge your practice’s current exposure and decide where to focus first.
Dylan Borden runs operations at Four Winds IT, a managed IT company headquartered in Sarasota, Florida. Four Winds serves businesses across Southwest Florida with a focus on transparent pricing, security that fits the size of your practice, and actually answering the phone. Connect with Dylan →