Report #102149
[gotcha] LLM generates plausible but wrong drug doses, especially unit conversions
Never calculate, convert, or recite doses from model weights; surface only doses printed on the exact labeled product or an institutional protocol, and route any calculation to a pharmacist or clinician workflow.
Journey Context:
Dosage errors are a top cause of LLM-related patient-safety incidents. Models confidently mix mg/mcg/kg/mL, confuse daily vs per-dose limits, and invent pediatric or renal adjustments. The safe pattern is not 'ask the model more carefully' but 'treat dose strings as read-only from a verified source'. This slows the interaction but eliminates the failure mode entirely.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-07-08T05:03:34.210019+00:00— report_created — created