Blog|Articles|November 19, 2026

12 ways to find the AI tools your staff signed up for without asking

Fact checked by: Chris Mazzolini

Staff are pasting patient data into shadow AI tools nobody approved. Here's how to find them, shut down the risky ones and offer something better.

Right now, someone at your front desk may be pasting a denial letter into a free chatbot to draft an appeal. One of your physicians may have a browser extension summarizing notes on the EHR screen. Neither of them asked first. In a Wolters Kluwer Health survey of 518 health care professionals, 40 percent said they had come across unauthorized AI tools at work. Nearly 20 percent said they had used one, and some of that use involved direct patient care.

Not knowing is getting more expensive. IBM's 2026 Cost of a Data Breach Report found shadow AI in 43 percent of the incidents it studied, up from 20 percent a year earlier, and those breaches cost more. Health care had the highest average breach cost of any industry for the 13th year running, at $6.64 million. HIPAA adds its own problem: if staff paste patient information into a vendor's tool and there's no business associate agreement, that can be an impermissible disclosure, whatever the intent.

Banning AI outright rarely works. Staff use these tools because they save time. The better move is to find out what's already in use, close the riskiest gaps and give people an approved option worth using. Here are 12 places to start.

The 12 ways to find shadow AI at a glance

1. Ask staff, and promise amnesty first

Send staff a short anonymous survey that promises no discipline for honest answers. Ask which AI tools they use, for what tasks and with what information, so you can see where the demand is.

2. Inventory browser extensions on every workstation

AI writing assistants and summarizers that run as browser extensions can read what's on screen, including the EHR. Have IT inventory extensions across practice computers and set an allow list in Chrome or Edge.

3. Check your network logs for AI traffic

Ask your IT vendor for a firewall or DNS report on traffic to common AI services, broken out by device. Heavy use from a single workstation points to a workflow you need to understand.

4. Review apps connected to your email and office accounts

Microsoft 365 and Google Workspace list the third-party apps users have authorized. Revoke unfamiliar AI assistants with mailbox access and require administrator approval for new app connections.

5. Find the note-taking bots in your meetings

AI note-takers can join calls on their own and transcribe patient discussions. Check attendee lists for bots, block outside bots in Zoom, Teams or Meet, and standardize on one transcription tool covered by a business associate agreement.

6. Look at the phones staff use for work

Consumer scribe apps, AI keyboards and recording apps on personal phones carry real risk. Spell out allowed apps in your bring-your-own-device policy, consider mobile device management and ask clinicians directly about recording apps.

7. Search expense reports and card statements

Many AI tools cost $20 a month or less and slip through on cards and expense reports. Search 12 months of charges for recurring AI subscriptions, since each one is a lead and a price point for an approved replacement.

8. Ask your current vendors what AI they've turned on

EHR, practice management, phone and messaging vendors are adding AI features, sometimes on by default. Ask each vendor in writing which features are active, what data they use and whether it trains models, and confirm your business associate agreement covers them.

9. Sort what you find by the data it touches

Tier the tools you find: anything receiving patient information without a business associate agreement stops now, tools touching internal business data get reviewed, and tools that only see public information can usually stay with a no-patient-data rule.

10. Get a business associate agreement or get off the tool

Each first-tier tool either gets a business associate agreement or stops handling patient information. HHS cloud computing guidance treats a service storing or processing electronic patient data as a business associate even if the data is encrypted.

11. Give staff an approved tool that's actually good

In the Wolters Kluwer survey, 45 percent of unapproved-tool users cited faster workflows and 24 percent said approved tools didn't work as well. License a tool that handles staff's top two or three tasks under a business associate agreement and train people on it.

12. Write a one-page AI policy and recheck every quarter

A one-page policy should list approved tools, banned information, how to request a tool and who decides, with sign-off at onboarding. Repeat the extension, network, account and expense checks every quarter.


Related to this article