Sick and Wrong: Ontario Auditors Find Doctors' AI Note Takers Routinely Blow Basic Facts

Why It Matters

Inaccurate AI notes risk clinical mis‑treatment and legal liability, challenging trust in health‑tech deployments. The audit highlights the need for stricter procurement standards before scaling AI in patient care.

Key Takeaways

• Half of AI scribes fabricated patient details
• Two‑thirds missed mental‑health information
• Accuracy accounted for only 4% of vendor scores
• OntarioMD recommends manual review of AI notes
• No mandatory attestation for AI‑generated records

Pulse Analysis

The Office of the Auditor General of Ontario released a stark warning about AI‑powered note‑taking tools used by physicians. By testing 20 approved vendors with simulated consultations, auditors uncovered that nine systems invented treatment recommendations, twelve mis‑recorded medication information, and the majority failed to capture essential mental‑health cues. Such hallucinations in clinical documentation could lead to diagnostic errors, inappropriate prescriptions, and compromised patient safety, eroding confidence in AI assistance for frontline providers.

A deeper issue surfaced in the procurement methodology. The scoring rubric heavily favored non‑clinical criteria—30% for a vendor’s Ontario presence—while accuracy, bias controls, and security each contributed less than 5% to the overall rating. This misaligned weighting allowed vendors with subpar performance on core medical tasks to win contracts, exposing the health system to both clinical and privacy risks. Experts argue that future evaluations must prioritize factual fidelity and robust risk assessments to prevent the selection of unreliable tools.

The findings arrive as Canadian provinces and U.S. states race to embed generative AI into health workflows. While AI promises efficiency gains, the Ontario audit underscores that unchecked deployment can amplify errors rather than alleviate workloads. Policymakers and health IT leaders should mandate real‑world validation, enforce mandatory attestation of AI‑generated notes, and establish transparent accountability frameworks. Only through rigorous standards can the industry harness AI’s potential without jeopardizing patient outcomes.