Exclusive: Researchers trick a bot that prescribes meds

Security researchers successfully jailbroke Doctronic's AI prescription refill bot, used in a pilot program in Utah. They exploited the system with simple techniques, demonstrating vulnerabilities in its underlying structure. The researchers manipulated the bot to spread false vaccine information and recommend inappropriate dosages and medications. This included tripling an OxyContin dose and mislabeling methamphetamine as therapeutic. Despite the company's claims of security measures, the researchers found the flaws easy to exploit. The testing occurred on the public chatbot, even though Utah's program operates within a regulated environment. Doctronic stated it takes security research seriously and utilizes ongoing adversarial testing. The Utah pilot allows patients to renew prescriptions through the AI system without direct doctor involvement for certain medications. The researchers altered the bot's knowledge by providing fake regulatory updates. The bot's ability to influence refill recommendations and medical summaries poses a potential threat. Mindgard, the research firm, contacted Doctronic with their findings, but the issues persisted after initial attempts at resolution.