As doctors increasingly lean on artificial intelligence for diagnostic assistance and clinical workflows, cybersecurity researchers have exposed a troubling vulnerability: they say they were able to deceive an Australian company’s AI-powered medical assistant in just three steps, potentially altering the clinical guidance it produced. The company in question says it had already patched the issue before the findings were published. Nonetheless, the episode highlights the broader tension between rapid AI deployment in healthcare and the painstaking security work needed to ensure those tools cannot be gamed — with patient safety hanging in the balance.
Medical AI Tools May Be Open to Manipulation, Researchers Warn
