AI Medication Dosing Errors Pose Risks to Physicians

New research shows general-purpose AI models frequently fail to calculate correct drug dosages for patients.

Updated on Oct. 29, 2026 in Artificial Intelligence

Isometric editorial illustration of an industrial scale weighing a glass vial, representing the need for verification in medical AI. — Doximity research found that major AI models frequently output incorrect medication dosages, leading the firm to launch DoseCheck to verify clinical accuracy. AI Illustration. Upload story photo >

Live Poll

Would you trust an AI tool to provide accurate medication dosing for you or your family?

Doximity research reveals that major AI models produce incorrect medication dosing in 5.6% of responses, with 19.5 per 1,000 answers containing serious or potentially lethal errors. In response, Doximity has introduced DoseCheck, a specialized verification tool designed to improve accuracy for clinical settings.

Why it matters

As physician AI adoption reaches 81%, these findings highlight a significant safety gap in general-purpose large language models, which often fail to account for patient-specific factors like age or renal function. This research emphasizes the move toward specialized verification layers to mitigate high-stakes errors in medical decision-making.

Researchers tested 8,500 dosing queries across 2,200 FDA-approved drugs, finding that recently approved medications face five times the error risk compared to established ones. Doximity claims its DoseCheck verification layer operates with a 99.9% accuracy rate.

The players

Doximity

A digital platform for U.S. physicians that provides specialized clinical AI tools and secure communication software.

The details

The errors occurred because models frequently failed to adjust for critical patient variables, including age, kidney function, and known drug interactions. DoseCheck acts as an integrated verification layer within the Doximity Ask medical search engine to flag these discrepancies. These models, which underpin modern clinical tools, struggled most significantly with newer FDA-approved drugs, highlighting the limitations of current training datasets in high-precision clinical environments.

2023: Physician AI adoption rates were half of 2026 levels.
Summer 2026: Research showed general-purpose AI outperforming specialized tools.
October 2026: Doximity released research on AI medication dosing errors.

The Tech Race

This development marks a departure from the assumption that general-purpose AI models are sufficiently robust for clinical practice. It directly contrasts with recent benchmarks from Summer 2026 that had suggested general-purpose AI was closing the performance gap with specialized medical platforms.

Physicians should be aware that general-purpose AI currently carries a 5.6% risk of medication dosing errors when used for clinical advice. Practices are now beginning to shift toward specialized tools like DoseCheck to provide a necessary safety buffer in daily workflows.

The takeaway

With 43% of AI-generated dosing errors classified as significant, the industry is shifting focus from raw model performance to safety-critical verification layers. Watch for further benchmarks on how these verification tools handle complex drug-to-drug interactions in real-world clinical use.