AI Is Moving Faster Than Your Practice's Security

A clinic manager told us recently that her practice had "gone all in on AI" over the summer. A scribe in the exam rooms, a chatbot on the website, a tool that drafted review replies, and a couple of staff using a free chatbot on the side to speed up notes. When we asked a simple question, which of those tools had a signed agreement covering patient data, the room went silent. The honest answer was: nobody had checked.

That story is not unusual, and this week a new report put real numbers behind it. The uncomfortable pattern is that practices are adding AI far faster than they are securing it.

The gap the executives just admitted to

MedCity News and Cotiviti released a Healthcare AI Readiness Index this month, built from candid answers by 70 executives at providers and payers. The top line is blunt: adoption is accelerating, readiness is not. More than 70 percent of the organizations surveyed have already started using AI tools. But only 32 percent of providers said they feel very prepared to respond to an AI assisted cyberattack, and just 42 percent of payers said the same.

Two more findings deserve a highlighter. Around 64 percent of providers reported that employees are already using unauthorized or non integrated AI tools, the kind nobody approved. And fewer than 40 percent of organizations have detailed written policies for how staff can use AI at all. So the pattern is: everyone is using it, most people are worried about the security of it, and almost nobody has written down the rules.

If large payers and provider systems with real security budgets feel this exposed, sit with what that means for an independent practice with no security team and a front desk already stretched thin.

Why this is a small practice problem, not a hospital problem

It is tempting to read breach headlines, see the seven figure numbers, and assume attackers only chase hospitals. The data says the opposite. IBM's 2025 cost of a data breach research put the average US healthcare breach at 7.42 million dollars, the most expensive of any industry for the fourteenth year in a row. That is the number for the big incidents. But smaller organizations are where the volume is.

According to breach data reported to the HHS Office for Civil Rights, incidents involving fewer than 5,000 records make up more than half of all reported breaches. In plain terms, small practices get hit constantly. They just do not make the news. Office for Civil Rights settlements for small practice cases have run from roughly 25,000 to 350,000 dollars, and that is before you count patient notification, the reputation hit, and the weeks of work you cannot bill for.

Attackers are not picking the biggest target. They are picking the softest one. A practice that adds four AI tools in a summer and secures none of them is exactly that.

The weak link is usually a vendor, not your front desk

Here is the finding that should change how you shop for AI. Analyses of 2025 breach data found that more than 80 percent of stolen protected health information was taken from third party vendors and business associates, not from the hospital or clinic itself. The patient data did not leak out of your office. It leaked out of a tool you trusted.

Every AI product you connect is one more company holding your patients' information, one more login that can be phished, one more startup that might get acquired or fold. We wrote about the fine print of these deals in what AI vendors won't tell you before you sign, and the security angle is the same story from another direction: the more separate vendors touch your data, the more doors you leave open. Four AI tools from four companies is four times the risk of one system that does the same jobs.

Shadow AI: the risk nobody put on the budget

The scariest exposure is often the one no manager approved. When a nurse pastes a patient note into a free chatbot to reword a message, or a front desk person uses a random transcription app to save five minutes, that patient data is now sitting on a server with no agreement, no encryption promise, and no way to get it back. That is shadow AI, and the index found roughly 64 percent of providers already have it happening.

You do not fix that with a lecture. You fix it two ways. First, a short written policy that says plainly which tools are approved and what can never go into an unapproved one. Second, and more effective, give staff an approved tool that is genuinely easier to use than the free one they are sneaking. People reach for shadow AI because the sanctioned path is slow or missing. Close that gap and most of the problem disappears.

What to do before you connect the next tool

You do not need a security department. You need five questions, asked before anything touches patient data:

• Will the vendor sign a business associate agreement? If a tool handles protected health information and the company will not sign a BAA, that is the end of the conversation.
• Can they prove their security? Ask for a recent security audit or a SOC 2 report and a clear breach notification process. A serious vendor has these ready. A hobby project does not.
• Who owns the data, and can you delete it? You should own your patient data and be able to export and erase it on demand, not beg for it later.
• Does it integrate, or is it another silo? A tool that connects to your existing systems is one door to watch. A tool that stands alone is a new door nobody watches.
• Is there a written rule for staff? One page telling your team what is approved and what is off limits prevents most shadow AI on its own.

Notice that four of the five are about consolidation and clarity, not fancy technology. The single most protective thing a small practice can do is keep the number of vendors touching patient data small, and make sure the ones that remain are accountable. Your website and booking flow count here too: if your own site collects patient information through an insecure form or a random plugin, it is part of the same attack surface. We covered that in whether your practice website is actually secure.

How EtherealMinds thinks about this

We build and run AI for healthcare practices, so we could be one more logo adding to your vendor pile. We work hard not to be. Our view is simple: AI should reduce the number of companies holding your patient data, not multiply it. When we put AI to work for a practice, it lives inside one system we build and stand behind, not a scattered set of subscriptions each with its own login and its own risk.

Take our AI receptionist. It answers the phone around the clock, books and reschedules on the spot, and hands off to a human when a call needs one. Because it is wired into the same patient acquisition and retention system that runs your ads, your follow up, and your booking, the patient information stays in one accountable place instead of leaking across four tools that barely talk to each other. Fewer vendors, one agreement, one place to watch.

We are also transparent about the parts vendors like to skip: what data is touched, how it is handled, who can see it, and how you would pull it back or shut it off. AI is worth adding to your practice. The right way to add it is deliberately, through a system built by someone who treats your patients' data as seriously as you do.