Researchers warn of cybersecurity risks in healthcare LLM use
Researchers published a special report warning of cybersecurity vulnerabilities in large language models used in healthcare and radiology. The report identifies risks including potential data breaches, manipulation of medical images and analysis results, and calls for security assessments before LLM deployment in clinical settings.
What happened
Researchers published a special report in Radiology: Artificial Intelligence addressing cybersecurity challenges of large language models in health care, particularly in radiology. The report, led by Tugba Akinci D'Antonoli from University Hospital Basell, identifies vulnerabilities in LLM deployment before integration into medical systems.
Reported impact
LLMs are susceptible to security threats that could allow malicious actors to extract sensitive patient data, manipulate information or alter outcomes. In radiology specifically, attackers could manipulate image analysis results, access sensitive patient data or install arbitrary software.
Index analysis
AI-inherent vulnerabilities include adding intentionally wrong or malicious information into training data or bypassing internal security protocols to produce harmful or unethical responses. Non-AI-inherent vulnerabilities involve the broader ecosystem in which LLMs are deployed and can lead to data breaches, data manipulation or loss and service disruptions.
Governance controls
Researchers caution that cybersecurity risks must be carefully assessed before LLM deployment in health care. Radiologists should enact protective measures when dealing with LLMs.
Classification
Sources and evidence
Sources not yet published
This record was researched and written by the Index from original reporting. Its sources are still being verified by an editor, and they are listed here once that check is complete. Until then the record carries no reviewer.