ONTARIO AUDIT: AI DOCTOR'S NOTETAKER INVENTS MEDICAL DATA

Healthcare providers in Ontario are using AI systems to transcribe and summarize patient interactions, but an audit has exposed critical flaws in accuracy. The systems frequently hallucinate—generating medical information that was never discussed or documented. Common errors identified include: - False referrals: The AI created therapy recommendations that doctors never made - Incorrect prescriptions: Medication details were altered or fabricated - Inaccurate patient history: Medical records contained information not provided during visits These mistakes occur because large language models can confidently produce plausible-sounding text without verifying factual accuracy. When applied to medical records, this behavior becomes dangerous. The audit did not specify which AI systems or vendors were involved, but the findings affect multiple healthcare facilities across Ontario. Doctors relying on these summaries risk making decisions based on false information, potentially harming patients. Healthcare professionals have expressed concerns about the increasing integration of AI into clinical workflows without adequate validation. Many are unaware of the limitations of these systems or how frequently errors occur. Ontario health authorities have not yet released formal guidance on remediation, though the audit signals a need for mandatory human review of all AI-generated clinical notes. Some providers are reverting to manual documentation until systems improve. Experts recommend that any AI used in healthcare must be specifically trained and validated on medical data, with transparent error rates disclosed to users. Generic language models designed for general-purpose tasks are unsuitable for clinical documentation. The findings align with similar concerns raised in other jurisdictions about AI reliability in high-stakes environments. Patient safety advocates are calling for regulatory oversight of clinical AI systems before wider adoption occurs.