Large language models provide unsafe answers to patient-posed medical questions - Summary - MDSpire

Large language models provide unsafe answers to patient-posed medical questions

  • By

  • Rachel L. Draelos

  • Samina Afreen

  • Barbara Blasko

  • Tiffany L. Brazile

  • Natasha Chase

  • Dimple Patel Desai

  • Jessica Evert

  • Heather L. Gardner

  • Lauren Herrmann

  • Aswathy Vaikom House

  • Stephanie Kass

  • Marianne Kavan

  • Kirshma Khemani

  • Amanda Koire

  • Lauren M. McDonald

  • Zahraa Rabeeah

  • Amy Shah

  • February 13, 2026

  • 0 min

Share

Objective:

To assess the safety of four publicly available chatbots in providing medical advice to patients, focusing on their reliability and potential risks.

Key Findings:
  • Statistically significant differences in safety among chatbots, with implications for patient care.
  • Problematic response rates ranged from 21.6% (Claude) to 43.2% (Llama), indicating varying levels of reliability.
  • Unsafe responses varied from 5% (Claude) to 13% (GPT-4o, Llama), raising concerns about patient safety.
  • Qualitative analysis revealed responses that could potentially lead to serious patient harm, underscoring the need for caution.
Interpretation:

The findings indicate that millions of patients may be receiving unsafe medical advice from chatbots, highlighting an urgent need for improvements in clinical safety protocols.

Limitations:
  • The study only evaluated four specific chatbots, which may not represent the broader landscape of available models.
  • Responses were limited to primary care topics, which may not encompass all medical inquiries, potentially skewing the results.
  • The selection of chatbots may introduce bias, as the chosen models may not reflect the full range of capabilities and safety profiles.
Conclusion:

Further work is necessary to enhance the clinical safety of large language models used in medical advice.

Original Source(s)

Large language models provide unsafe answers to patient-posed medical questions

  • by Rachel L. Draelos, Samina Afreen, Barbara Blasko, Tiffany L. Brazile, Natasha Chase, Dimple Patel Desai, Jessica Evert, Heather L. Gardner, Lauren Herrmann, Aswathy Vaikom House, Stephanie Kass, Marianne Kavan, Kirshma Khemani, Amanda Koire, Lauren M. McDonald, Zahraa Rabeeah, Amy Shah

  • February 13, 2026

Related Content

Effective Treatment of Hemophagocytic Lymphohistiocytosis Linked to Scrub Typhus Using Doxycycline and Immunomodulatory Therapy: A Case Study

Survival Duration and Influencing Factors in Cervical Cancer Patients in Tigray, Ethiopia: A Retrospective Analysis

Reevaluating the Alcohol–Cancer Link: Long-Term Cancer Mortality Outcomes in the REGARDS Study