Shadow AI led to a data breach at Johns Hopkins. This type of breach will happen again.

Johns Hopkins released notification of a data breach at the Johns Hopkins Bayview Medical Center (JHBMC) in Baltimore. The source of the breach was not an external malicious actor but an internal mistake. An employee used an unapproved, public AI tool and exposed the data of more than 3,000 patients.

Unsanctioned use of AI tools is a known security risk for healthcare, and the industry is still figuring out how to reduce this trend, shadow AI, among employees who face mounting workloads and have access to a rapidly evolving technology. While security and governance teams race to put the right guardrails in place, sensitive data remains at risk of exposure through public, unvetted AI tools.

Uploading patient information to an unapproved AI tool led to the breach

On Sept. 21, an employee at JHBMC uploaded a spreadsheet with the data of 3,182 patients to a third-party AI tool that Johns Hopkins had not approved for use. The spreadsheet included patient names, medical record numbers, treatment information and provider names. The medical system discovered the incident during an internal review on the same day.

Johns Hopkins took internal action, and on Oct. 22, it began notifying impacted patients.

“These steps included having the workforce member delete the initial AI prompt and subsequent thread and document from the AI tool history, reaching out to the company that owns and operates the AI tool to request deletion, and implementing corrective measures to prevent similar incidents,” according to an emailed statement from Johns Hopkins.

Shadow AI is the successor to shadow IT, and it is happening all over healthcare

People were turning to unapproved technology long before AI came onto the scene. The advent of personal devices fueled shadow IT in hospitals and health systems. Familiarity and ease of use drive people to leverage tech outside of the approved, and often clunky, options available in the hospital.

“The phone in your pocket is almost invariably eons beyond any IT that’s being used in the hospital,” Joel Selanikio, MD, a pediatric attending physician at Georgetown University Hospital and frequent speaker as a principal of FutureHealth, told Digital Health Insights.

Now, that shadow, and its accompanying risks, is growing with the prevalence of readily available AI tools. A 2025 report from clinician network Offcall found that 81 percent of physicians were unhappy with the pace of their employer’s AI adoption, and 71 percent reported that they had limited influence over which AI tools were selected for approved use. Those findings point to a scenario in which turning to unvetted AI tools that clinicians feel work best is a tempting option.

Wolters Kluwer Health surveyed 518 full-time healthcare professionals, including both providers and administrative workers, and found that 40 percent had seen use of an unauthorized AI tool in their organizations. Plus, 17 percent of respondents said they had used an unauthorized AI tool.

If employees do not have access to the AI tools that they feel work best and a clear understanding of AI policies, they will experiment with other options, putting patient data and their organizations at risk.

Training and technical controls are critical in preventing more breaches like this

Healthcare is embracing AI for clinical and administrative workflows, but the security risks that come with these tools will continue to manifest in data exposure without the right governance and controls.

JHBMC emphasized in its emailed statement the plan “to reinforce our policies and training concerning the appropriate handling of protected health information, including restrictions on use of unauthorized applications and artificial intelligence tools and reeducation on the tools that are institutionally approved.”

Employees need to know which tools they can use, how to use them and the consequences of using unauthorized tools. In addition to that, technical controls that block the use of unsanctioned AI tools can reduce the risk of employees mistakenly or intentionally turning to AI that a hospital or health system hasn’t vetted or approved.

“A year from now, we will have very similar problems unless the hospitals both do the new things that they need to do in terms of signing agreements and using the tools that are going to comply with current healthcare law around patient information but also the old things that apparently they haven’t done yet… training, particular IT changes like blocking particular websites or blocking uploads,” Selanikio said. “Given the advances that you can expect in the next year from AI, the issue might be much worse.

Carrie Pallardy, a Chicago-based freelance writer and editor, began her career covering healthcare more than a decade ago. Her work has taken her into many different industries, but covering healthcare delivery remains a constant focus. She can be reached at [email protected] or on LinkedIn.