Vnitr Lek 2026, 72(5):345-350 | DOI: 10.36290/vnl.2026.067

Large language models in healthcare: benefits, risks and outlook

Ondřej Volný^{1, 2, 3}, Adam Gřunděl^{3, 4, 5}

¹ Komplexní cerebrovaskulární centrum, Neurologická klinika, Fakultní nemocnice Ostrava

² Lékařská fakulta Ostravské univerzity, Katedra AI, digitalizace a inovativních technologií

³ Česká společnost pro umělou inteligenci a inovativní digitální technologie v medicíně ČLS JEP (ČSAIM)

⁴ Ministerstvo zdravotnictví

⁵ Nemocnice AGEL Nový Jičín

Large language models (LLMs) such as ChatGPT, Gemini, or Claude have become widely accessible tools increasingly used by patients and clinicians for health-related purposes, regardless of whether these tools are certified as medical devices. A large-scale analysis of over 500,000 health-related conversations with Microsoft Copilot demonstrated that nearly one in five queries involves personal symptom assessment, and one in seven personal health queries concerns someone other than the user. Health-related queries markedly increase during nighttime hours, when traditional healthcare access is most limited. This trend creates both opportunities (information availability, patient education, administrative support) and significant risks (uncritical trust in non-certified systems, data security concerns, amplification of health anxiety). In January 2026, OpenAI launched ChatGPT Health - a dedicated space for health queries integrating medical records - which is not certified as a medical device and is not available in the EEA. This article summarises current evidence, the Czech Medical Association position statement, and offers a practical framework for clinicians, together with an outlook on regulatory development in the Czech Republic.

Keywords: large language models, LLM, ChatGPT, AI in healthcare, medical device, certification, patient safety.

Accepted: August 20, 2026; Published: August 27, 2026  Show citation