LEGACY & INSIGHTS: HEALTH & TECHNOLOGY
Hopkins Deleted the Chat. Did the AI Company?
Johns Hopkins data breach began with one employee’s upload, Hopkins says. The file held records on 3,182 patients and went to an AI tool the hospital had not approved.
Johns Hopkins data breach started with an employee’s upload, Hopkins says. A worker at Johns Hopkins Bayview Medical Center uploaded a spreadsheet on 3,182 patients to an AI tool the hospital had not approved. Hopkins has not named the tool. So the question now is where that file went.
Johns Hopkins Data Breach: What Hopkins Says Happened
Hopkins found the upload behind the Johns Hopkins data breach on September 21 during an internal review, WMAR reports. The spreadsheet covered patients treated from May through September. The Banner reports the data included provider information, admission dates, treatment information, medications, account numbers and medical record numbers. WMAR adds that names may have been included. However, Hopkins says it held no Social Security numbers or financial account details. No report cited here describes any misuse of the data.
Where the Johns Hopkins Data Breach File Went
The employee deleted the prompt, thread and document, and Hopkins asked the tool’s operator to delete the data, The Banner reports. Still, deleting a chat removes it from a user’s history, SAN reports, but that does not necessarily remove it from the company’s servers. Hopkins has not said whether the company confirmed deletion. Nor has it said whether the data from the Johns Hopkins data breach could train the tool.
Shadow AI Beyond the Johns Hopkins Data Breach
Hopkins is not alone. In May, Loma Linda University Health in California uploaded a file with limited patient information to an outside AI platform by accident, SAN reports. A January Wolters Kluwer survey found 17% of surveyed health care providers and administrators had used an unapproved AI tool at work, SAN says. The risk behind the Johns Hopkins data breach sits in everyday work, not only in attacks.
Hopkins Had Rules
Before the Johns Hopkins data breach, Johns Hopkins University’s 2025 AI guidelines told its community not to enter clinical records into unapproved third-party generative AI tools, SAN reports. However, Hopkins has not said whether technical controls could have blocked this upload before September 21. Separately, I. Glen Cohen, a Harvard law professor, says a health record that leaves the hospital system is “losing a lot of the protections that would otherwise apply under federal law,” Inc. reports.
WATCH / WBAL-TV 11 REPORT
WBAL-TV 11 Baltimore reports that Johns Hopkins Bayview is notifying more than 3,100 patients after an employee uploaded their information to an unapproved AI tool. Source: WBAL-TV 11.
What Johns Hopkins Data Breach Patients Can Do
Hopkins began mailing letters to patients in the Johns Hopkins data breach on October 22, WMAR reports. The hospital offered affected patients 12 months of credit monitoring through Experian IdentityWorks. A dedicated call center answers questions at 1-833-931-7766. In addition, patients can file a complaint with HHS’s Office for Civil Rights, which enforces HIPAA privacy rules, SAN reports. Still, none of that tells patients where the file is now.
What the Johns Hopkins Data Breach Leaves Unanswered
Hopkins has not named the AI tool or its company. It has not confirmed whether the company deleted the file. SAN reports the hospital also did not answer whether it notified HHS. Until those answers are public, the Johns Hopkins data breach leaves a deleted chat as a claim, not a guarantee.
Signal Breakdown
Not every claim about the Johns Hopkins data breach carries the same weight.
Reported and Documented
- Upload. A worker at Johns Hopkins Bayview uploaded a spreadsheet on 3,182 patients to an unapproved AI tool, per WMAR and The Banner.
- Discovery. Hopkins found it on September 21 in an internal review.
- Notice. Letters began mailing October 22, with 12 months of credit monitoring.
- Data. Hopkins says no Social Security or financial account numbers were included.
Claims and Counterclaims
- Hopkins. It says it asked the tool’s operator to delete the data and is retraining staff.
- SAN. Deleting a chat does not necessarily remove it from company servers, yet no one has confirmed what happened here.
- Survey. Still, SAN cites a Wolters Kluwer survey in which 17% of providers and administrators used an unapproved AI tool.
- Cohen. However, the Harvard law professor’s warning about lost protections comes through Inc., not from Hopkins.
Not Established
- which AI tool and company received the file
- whether deletion was confirmed
- whether the data was used to train a model
- whether Hopkins told HHS
- whether technical controls could have blocked the upload
- what happens to the employee
Sourcing: the incident facts come from The Banner (October 25), WMAR (October 23), WBAL-TV 11 (video report) and SAN. The Wolters Kluwer survey, the Loma Linda case and Johns Hopkins University’s 2025 AI guidance come from SAN. The Cohen quote comes from Inc., which cites Harvard Law Today; I did not open that source. I did not read Hopkins’ patient letter, and the Baltimore Sun page could not be read.
HOPKINS / KEY FACTS
RADAR / WHAT TO WATCH
KMOB1003 SELECTS
Music, style, tools, art and objects worth your attention.
