AI Safety 🇺🇸 27.07.2026 13:06

Ontario Auditors: AI-Assisted Medical Scribes Routinely Distort Basic Facts

MicrosoftMicrosoft Nuance CommunicationsNuance Communications Suki.AISuki.AI
An investigation by Ontario auditors revealed that AI systems for transcribing doctor-patient conversations, such as DAX Copilot by Nuance, Suki AI, and Robin AI, make serious errors. In 18 out of 55 cases reviewed, factual inaccuracies were found, including invented symptoms and incorrect recommendations, potentially compromising patient safety.
According to an investigation by the Auditor General of Ontario, AI tools for medical record-keeping used by physicians in the province systematically make significant factual errors. In 18 of 55 reviewed cases, auditors found inaccuracies in generated records, including fabricated symptoms, incorrect dates, medications, and even recommendations that the physician never made. Among the systems reviewed were DAX Copilot from Microsoft/Nuance, Suki AI from Suki.AI, and Robin AI from Robin Healthcare. Auditors noted that physicians often sign off on generated notes without checking them, relying on the AI. In one case, DAX Copilot attributed symptoms to a patient that the patient had not reported; in another, Suki AI incorrectly recorded a medication dosage. Although vendors claim their AI has undergone clinical trials and HIPAA certification, auditors called for mandatory real-time accuracy audits of such systems.
Source: The Register AI/ML — original
Our earlier posts on this topic ↓
Fresh news