When medical AI fails outside English
Year: 2026
Type: article
Abstract: Large language models (LLMs) are increasingly positioned as general-purpose medical systems, with demonstrated potential in diagnosis and management reasoning for patient care tasks.1 2 Their rapid progress has fuelled optimism that medical AI will soon support both clinicians and patients across specialties and settings, particularly in regions with limited access to care. Yet this narrative of generalisability obscures a critical limitation: the development, evaluation and empirical validation of current medical LLMs have been achieved in English-language contexts. By contrast, research and validation in non-English languages remain sparse, despite the fact that these languages represent the majority of the world’s patients. As a result, systems described as broadly capable are often least reliable for patients and clinicians who do not operate in English, exposing a foundational gap in the development and evaluation of medical AI.
Source: BMJ Digital Health & AI
Institutions: The University of Tokyo, Stanford University
Language: English
Subfield: Health Informatics
Field: Medicine
Domain: Health Sciences
Open Access status: diamond
OpenAlex ID: W7160840117