UMVA has learned that a disturbing pattern of errors and fabrications is plaguing artificial intelligence tools currently being used by doctors to document patient visits.
The technology, designed to alleviate the administrative burden on physicians by transcribing conversations into medical notes, is instead creating a dangerous web of misinformation. Instead of streamlining care, these systems are actively hallucinating data that never existed, potentially compromising the integrity of patient health records.
UMVA has uncovered details about how these systems operate, listening to private doctor-patient dialogues to compile clinical summaries. However, the reality of the output is far from reliable, with investigations revealing that a majority of tested programs produced inaccurate or incomplete information.
The most alarming findings involve the generation of outright fabrications. These AI tools have been caught inventing treatment plans, suggesting blood tests that were never ordered, and even documenting medical conditions or mental health statuses that were never discussed during the appointment.
In one unsettling instance, an AI system recorded that a patient had no physical masses when the topic had never even been raised. In other cases, the software incorrectly identified medications, listing drugs that the physician never prescribed, creating a hazardous discrepancy in the patient's medical history.
Beyond the accuracy of the notes, there are profound concerns regarding the sanctity of patient privacy. The use of these digital scribes requires sensitive, intimate health information to be processed by third-party systems, raising significant questions about whether confidential data is being adequately protected.
Patients trust their doctors to safeguard their most private details, yet this reliance on experimental technology threatens that foundation of trust. As these tools continue to be integrated into clinical settings, the risk of misdiagnosis or improper treatment plans based on machine-generated errors remains a growing reality for the healthcare system.