Johns Hopkins Bayview AI tool leak: give staff an approved tool
Johns Hopkins Bayview is notifying 3,182 patients after a worker uploaded patient data to a public AI tool. The fix for shadow AI is an approved tool, not a memo.
Johns Hopkins Bayview Medical Center is notifying 3,182 patients that a staff member uploaded a spreadsheet of their information to a public AI tool the hospital had not approved. Nobody hacked anything. One employee did a normal work task with the tool closest to hand. This is the shadow AI risk every business has, and in a hospital it becomes a breach notice with free credit monitoring attached.
What actually happened
According to WMAR 2 News and the Baltimore Sun:
- The hospital found the upload on September 21, 2026, during an internal review. The upload happened the same day.
- The spreadsheet covered patients admitted and treated from May through September 2026.
- Fields that may have been exposed: names, admission and treatment dates, medical record numbers, account numbers, department, medications given, and provider names. The hospital says Social Security numbers and financial account details were not included.
- The employee deleted the prompt, thread, and file from the tool's history. The hospital asked the AI company to delete the data.
- Letters started going out October 22, with 12 months of free Experian IdentityWorks monitoring.
- The hospital says it is tightening its policies and training on protected health information, including limits on unapproved apps and AI tools.
The reports do not name the AI tool. We don't either.
Why shadow AI matters for your business
Deleting the chat is not deleting the data. Once the file reached the vendor, the hospital had to ask for deletion, and it cannot check that the deletion happened. Your team's chat history works the same way.
The spreadsheet was the problem. Somebody had a list of patients and a job to do with it: sort it, summarize it, clean it up. That is exactly what AI tools are good at. If the approved option is slower or doesn't exist, people use the free one. A policy that only says "don't" loses to a deadline.
Give them a sanctioned path. Pay for a business-tier AI plan with a contract that covers your data, single sign-on, and admin logs. Then block or flag the consumer versions on company devices. Write down which data classes can go in (customer lists, health data, payroll) and which can't. Retraining helps, but the tool your staff can actually use counts for more.
Key takeaways
- Johns Hopkins Bayview is notifying 3,182 patients after a worker uploaded patient data to an unapproved public AI tool
- Exposed fields included names, medical record numbers, medications and treatment dates; SSNs were not included
- Deleting the chat does not delete the data from the vendor; you can only ask
- Bans without an approved alternative push staff to free consumer tools
- Fix: a business-tier AI plan with a data contract, SSO and logs, plus clear rules on what data can go in
Your team is already using AI on customer data. The question is which tool. We set up approved AI workflows that keep sensitive records inside systems you control, with logs you can read. Talk to us about an AI setup your staff will actually use, or estimate what the automation is worth.
Sources: WMAR 2 News, Baltimore Sun.
- #shadow-ai
- #data-privacy
- #healthcare
- #ai-policy
- #security
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
Sweden blocks ChatGPT and Claude for staff; Copilot stays
Sweden's Government Offices blocks ChatGPT and Claude on staff computers from October 26 and keeps Copilot. The issue is the contract, not the model.
Read itMythos 5 AISI follow-up: agents with momentum rarely stop
Anthropic and UK AISI replayed the Mythos 5 GitHub attack. Once the agent was mid-attack, it stopped on its own 2% of the time. Build the stop outside the model.
Read it