60% of evaluated AI Scribe systems mixed up prescribed drugs in patient notes, auditors say

An audit by the Office of the Auditor General of Ontario found that the province’s approved AI clinical note‑taking systems often missed critical details and invented facts not present in doctor — patient recordings, raising patient‑safety and procurement concerns. The review covered 20 systems approved for use by Ontario healthcare providers and is part of a wider report on public‑sector AI use in the province.

Auditors tested the tools with simulated doctor — patient recordings and had medical professionals compare the original audio with the AI‑generated notes. Nine of the 20 systems fabricated information or suggested treatment changes that did not appear in the source recordings; auditors cited examples such as false claims that no masses were found or that patients were anxious when neither point was raised in the audio.

The audit recorded high rates of clinically dangerous errors. Twelve of the 20 systems (60%) inserted incorrect drug information into patient notes, and 17 systems missed key details about patients’ mental health. In six cases auditors found systems either partially or fully omitted mental‑health concerns that were discussed in the recordings. OntarioMD, which supported procurement for the program, advised clinicians to manually review notes produced by AI scribe tools. The audit noted, however, that none of the approved systems require a mandatory user attestation confirming review, leaving a gap between recommended practice and enforced workflow safeguards.

The report also faulted how products were evaluated during procurement. A platform’s domestic presence in Ontario accounted for 30% of its evaluation score, while the accuracy of medical notes contributed just 4%. Controls for bias were weighted at 2%, threat/risk/privacy assessments at 2%, and SOC 2 Type 2 compliance at 4%. Auditors warned that these weightings could lead to selecting vendors whose tools produce inaccurate or biased medical records or that lack adequate protections for sensitive health information.

Provincial officials said more than 5,000 physicians are participating in the AI Scribe program and that there have been no known reports of patient harms linked to the technology. The audit report also records that the Ministry of Health did not provide an immediate response to requests for comment about the findings. The auditors set out clear priorities for builders and deployers: improve transcription and drug‑matching accuracy; enhance capture and documentation of mental‑health details; implement mandatory verification or attestation workflows so clinicians confirm AI notes before they become official records; and increase emphasis on bias, privacy and security controls to ensure tools meet scrutiny during procurement and clinical rollout.