Weill Cornell Medicine reports unauthorized AI use of patient radiology data

Weill Cornell Medicine disclosed on November 5, 2026 that protected health information from its radiology locations was put into AI systems. Machine learning then processed the data without authorization between June 10, 2026, and June 20, 2026. The activity covered ten days in June 2026 and involved radiology research.

Patient information originated at Weill Cornell Imaging at NewYork-Presbyterian and at NewYork-Presbyterian Hospital. Researchers traced the information to both clinical locations. Both sites served as sources for the material used in the AI tools. The datasets put into the AI tools included these administrative details.

The datasets also contained dates and times of radiology services, facility names, and health care provider names. The involved material also covered clinical information tied to radiological examinations, including diagnoses, medication information, and surgical and procedural histories. Machine learning handled these clinical details from the radiological exams.

The institution stressed that common personal identifiers were left out. Patient names and addresses were not involved. Telephone numbers, e-mail addresses, Social Security numbers, and patient-specific financial information were also excluded from the research datasets.

Federal law calls medical record numbers critical patient identifiers. Under federal guidelines, these identifiers qualify as protected health information when they can be linked to individual patients along with other metadata, and abstract data points may become identifiable when paired with additional information.

On September 6, 2026, the institution discovered that patient data had been used without permission in a radiology study. It began an investigation that day. It found the problem right away and opened an investigation to figure out what happened.

About sixty days passed between the September 6 discovery and the public notice, and Weill Cornell Medicine issued the notice on November 5, 2026. Federal health privacy law says people affected by potential breaches must be notified without unreasonable delay, usually within 60 days after discovery. This requirement falls under the Health Insurance Portability and Accountability Act. It followed that schedule by putting out the notice at that point.

Weill Cornell Medicine said the use violated its policies and occurred without patient permission. The institution called the activity unauthorized and said it conflicted with internal privacy rules. Someone inadvertently crossed policy boundaries inside an authorized research setting, which separates this incident from traditional hacker breaches. The event started inside an authorized research setting instead of coming from outside attackers. Using external AI platforms meant data left the traditional safeguards of the hospital’s network.

Weill Cornell Medicine said it saw no sign that an unauthorized person viewed the information, and it has not found proof that an unauthorized individual accessed the patient information. The notice said the data were limited and stressed that no misuse of any patient information has been indicated. The team checked whether outsiders obtained or shared the patient information. They also sought to confirm whether third parties got the information or if AI providers spread it further.

Weill Cornell Medicine suspended the research activity after discovering the unauthorized use. It also suspended the specific project for an indefinite period so reviewers could examine protocols and software tools. It asked that all information processed by the AI tools be removed. Officials contacted AI software providers and demanded confirmation that the data had been taken out of their systems.

The institution added more training to strengthen compliance with rules for proper AI tool use and patient information protection. Staff training aims to prevent similar policy violations. The institution put in place a retraining program for staff and research faculty centered on ethical AI use and patient privacy rules.

It reminded employees that approval is required before external software handles clinical records. The institution started improving internal technical safeguards to watch and limit data transfers to unvetted AI platforms. These upgrades work like digital guardrails that can flag or block unauthorized data exports.

Purging all data from AI training sets can be technically difficult even when providers are asked to remove it, so taking every piece of data out of machine learning training sets in full can create technical problems. Weill Cornell did not disclose the individuals responsible or the specific subject of the related research.

Weill Cornell Medicine sees nearly 3.6 million patient encounters each year. The institution has gathered more than 1,300 National Institutes of Health grants and has 1,800-plus enrolled students in its medical and graduate school. It sits on Manhattan’s Upper East Side and is considered one of the nation’s leading medical and graduate schools.

For people wishing to verify whether their records were included, Weill Cornell Medicine provided a general inquiry line at (212) 746-5454. Under federal rules, patients can lawfully ask for an accounting of how their health information was disclosed. People who consider the response insufficient can file a formal complaint with the HHS Office for Civil Rights. People were told to watch insurance documentation and explanation-of-benefits forms for unusual activity or unfamiliar medical claims to guard against medical identity misuse.

Health systems run into complex challenges when blending advanced technology into clinical workflows while keeping privacy standards stringent. The incident offers an example of challenges in adopting AI within healthcare privacy constraints, and the need for institutional oversight is growing because of this event.