Johns Hopkins Bayview employee uploaded 3,182 patients' records to unapproved AI tool

Johns Hopkins Bayview Medical Center is notifying 3,182 patients in Baltimore that their health and treatment information may have been exposed after a hospital employee uploaded a spreadsheet of patient records to a public artificial intelligence tool the hospital had not approved — and the hospital has not confirmed whether the data was ever actually deleted from that platform's servers.

The hospital says it discovered the incident on September 21, 2026, the same day it happened, during an internal review. The employee was instructed to remove the prompt, the conversation thread, and the uploaded document from the tool's history, and the hospital reached out to the AI company asking that the data be deleted. But as Straight Arrow News reported, the hospital's public release did not name the company, and no statement confirms the vendor followed through. Deleting a chat removes it from a user's visible history; what happens on the company's servers depends on that company's own data policies.

The spreadsheet covered patients admitted or treated at Bayview between May and September 2026. Among the details potentially exposed, the hospital listed patients' names, admission and treatment dates, medical record numbers, account numbers, departments of service, medications administered, and provider names. No Social Security numbers or financial account information were included in the file, the hospital said.

Johns Hopkins is committed to maintaining the privacy and security of its patients' information and is taking this incident very seriously. We sincerely regret any inconvenience or concern this may have caused.— Johns Hopkins Bayview Medical Center

No outside attacker was involved. This was not a hack or ransomware event — the exposure came entirely from an insider upload. The hospital has not attributed intent, and sources describe no indication the employee acted maliciously.

The rule the employee broke was already written down. Johns Hopkins University's AI guidelines, issued in 2025, spelled out a clear prohibition: members of the Hopkins community were not to enter clinical records into unapproved third-party generative AI tools, with the warning that such tools offer no guarantee of privacy, according to Straight Arrow News. Johns Hopkins Medicine's own website tells patients that AI tools may use their data only for approved healthcare purposes. The policy existed; the incident happened anyway.

Straight Arrow News reported that Hopkins did not answer its question about whether the hospital had any technical controls — such as data loss prevention software — in place to block such uploads before September 21. The hospital said after the fact that it has implemented corrective measures and reinforced its policies and training on protected health information, including restrictions on unauthorized applications and AI tools.

The broader pattern extends beyond one hospital. According to Straight Arrow News, a Wolters Kluwer survey of more than 500 healthcare providers and administrators, released in January, found that one in six — 17 percent — reported having used an unapproved AI tool on the job. When Wolters Kluwer surveyed physicians and nurses separately in March, just 27 percent said they knew of formal AI governance policies at their organizations. That same month of May also brought a separate disclosure from Loma Linda University Health in California, where a file with limited patient information was accidentally uploaded to an outside AI platform in the course of an approved research study, again according to Straight Arrow News.

Under HIPAA, vendors that handle patient data normally require a signed business associate agreement with the covered healthcare entity. A public consumer AI tool operating without such an agreement has no contractual obligation to protect or delete that data, according to ogunsecurity.com. Many consumer AI products hold onto prompts and uploaded files, and some explicitly reserve the right to draw on that material when refining their models.

The hospital began mailing notification letters to affected patients on October 22, 2026 — 31 days after the discovery date, by calendar count. Affected patients are directed in the letters to enroll in a year of complimentary credit monitoring provided through Experian IdentityWorks. Those who wish to pursue a formal complaint may do so with the Department of Health and Human Services' Office for Civil Rights, the federal body responsible for enforcing HIPAA's privacy rules, according to Straight Arrow News. The hospital has not said publicly whether it has reported the incident to HHS.

Key questions remain open: the hospital has not named the AI platform, has not confirmed whether the vendor deleted the data, has not said whether the data could be used to train the tool, and has not reported whether any affected patient has experienced identity theft or misuse as a result of the exposure. Straight Arrow News noted that affected patients have no way, at present, to determine where their information now resides or what use may be made of it.

Why it matters — The incident shows that a single employee using a free AI tool — in violation of existing written policy — can expose thousands of patients' medical records with no hacker required, and that patients may have no way to verify their data was ever deleted.

Reported by techreview.si, focalnews.org, cbsnews.com, san.com, ogunsecurity.com