Johns Hopkins Bayview Medical Center is mailing letters to about 3,182 patients because a staff member put a spreadsheet into a public AI tool. Nobody broke in. The data left through a browser tab that was already open.
The facts come from the hospital's own October 23 release, as reported by Becker's Hospital Review. On September 21, a workforce member uploaded a spreadsheet to an AI tool the hospital had not approved. It covered admissions and treatment from May through September: names, admission and treatment dates, medical record and account numbers, departments, medications given and provider names. No Social Security numbers or financial account information. The hospital learned of it the same day, had the employee delete the prompt, the thread and the file, asked the tool's operator to delete the data, and is offering a year of credit monitoring. It says it is also restricting unapproved apps and AI tools and retraining staff on approved ones. The tool was not named.
Hopkins found this on day one and said so in public. Many hospitals might not have found it at all.
Picture a person with a deadline and a spreadsheet. The question is usually ordinary: sort these admissions by department, flag the duplicates, draft the summary for Thursday's meeting. In a public chatbot that takes thirty seconds. In many hospitals the approved path takes a ticket, a VPN and a second login, if one exists at all.
Clinicians make this trade every week with notes, patient letters and prior authorization appeals. I have watched smart, careful people do it, because the policy lives in a PDF and the deadline lives on the screen in front of them.
A policy memo does not compete with convenience. A faster sanctioned tool does.
The reflex is to block the consumer AI domains on the hospital network. Do that and the same spreadsheet gets photographed with a personal phone and typed into an app on cellular data, where no filter sees it and no log records it. You have moved the risk somewhere you can no longer measure. Restricting unapproved tools, which Hopkins is doing, works best when the approved alternative is already sitting there, quicker than the workaround.
Want the next decision guide in your inbox the morning it publishes? The Sarah Matt Briefing carries one every Tuesday. Subscribe here.
Find out what your staff already use, without punishing the answer. The list you get back is your real AI inventory, and it is almost certainly longer than the one IT keeps. Then take the two tasks that show up most and put a sanctioned option in front of them that is faster than the open tab.
Training matters, but retraining people to avoid a tool they find useful has a short half-life. Speed of the right path is the control that holds.
What is your hospital's approved alternative to the tab your team already has open?
Source: Becker's Hospital Review, "Johns Hopkins alerts 3,182 patients after worker uploads data to unapproved AI tool," October 26, 2026, citing the hospital's October 23 release.
Dr. Sarah Matt is a surgery-trained physician-executive currently practicing internal medicine (charity care). She advises health systems and digital health companies on clinical AI implementation, vendor contracting and operational strategy.
Weekly clinical reality on AI, governance, and what actually works inside health systems.