Artificial intelligence hallucinations in large language models for medical education and healthcare: a scoping review of detection, mitigation, and governance
- 1Department of Medical Education, Medical Education and Development Center, Shahid Sadoughi University of Medical Sciences, Yazd, Iran
- 2Medical Education and Development Center, Shahid Sadoughi University of Medical Sciences, Yazd, Iran
Abstract
Background and objective Generative artificial intelligence (AI) and large language models (LLMs) are increasingly used in medical education and healthcare. Yet AI hallucinations-plausible but inaccurate, unsupported, or fabricated information or citations-may compromise educational integrity, clinical judgment, patient safety, and trust. This scoping review mapped evidence on their nature, detection and validation, mitigation, and educational and governance implications.Methods This scoping review was conducted in accordance with the Joanna Briggs Institute methodology and informed by the Arksey and O’Malley framework. Reporting followed the PRISMA Extension for Scoping Reviews. PubMed/MEDLINE, Scopus, IEEE Xplore, and Web of Science were systematically searched for English-language sources published from January 1, 2018, to July 31, 2026. Study selection and data extraction were independently performed by two reviewers. Findings were synthesized using descriptive thematic analysis.Results Of 2,051 identified records, 21 sources were included in the final synthesis after duplicate removal and screening. These included five empirical or model-evaluation studies and 16 review, conceptual, or opinion-based sources. The analysis identified five main themes and 13 subthemes: (1) conceptualization and manifestations of AI hallucinations in healthcare; (2) detection, assessment, and validation of hallucinations; (3) mitigation strategies for improving the reliability of generative AI outputs; (4) educational implications and critical AI literacy; and (5) governance, accountability, and trustworthy implementation in healthcare. Retrieval-augmented generation, structured prompt design, output-verification methods, and human oversight were among the most frequently identified approaches for reducing hallucination-related risks.Conclusion AI hallucinations are not merely technical errors; they represent a multidimensional challenge to the safe and responsible use of LLMs in medical education and healthcare. Reducing their risks requires a multilayered approach that includes evidence-based validation, human oversight, critical AI literacy, redesign of educational assessment, and transparent governance and accountability frameworks.
Related
- Balancing innovation and ethics: responsible governance of artificial intelligence hallucinations in healthcare.
- Bias Mitigation Across the Healthcare Artificial Intelligence Lifecycle: A Structured Narrative Review
- Regulation of medical artificial intelligence in public health: governance lessons and implications for China
- Impact of Artificial Intelligence on Operational Efficiency and Cost Reduction in U.S. Healthcare Organizations
- Wetter’s Digital Twin: A Patient-Specific Framework for Artificial Intelligence and Healthcare
BibTeX
@article{heydari2026artificial,
title = {{Artificial intelligence hallucinations in large language models for medical education and healthcare: a scoping review of detection, mitigation, and governance}},
author = {Sara heydari and zohreh sadat alavi},
journal = {Medical Education Online},
year = {2026},
month = nov,
doi = {10.1080/10872981.2026.2756806},
url = {https://doi.org/10.1080/10872981.2026.2756806},
note = {CC BY},
}
RIS
TY - JOUR
TI - Artificial intelligence hallucinations in large language models for medical education and healthcare: a scoping review of detection, mitigation, and governance
AU - Sara heydari
AU - zohreh sadat alavi
PY - 2026
DA - 2026/11/07
JO - Medical Education Online
DO - 10.1080/10872981.2026.2756806
UR - https://doi.org/10.1080/10872981.2026.2756806
AB - Background and objective Generative artificial intelligence (AI) and large language models (LLMs) are increasingly used in medical education and healthcare. Yet AI hallucinations-plausible but inaccurate, unsupported, or fabricated information or citations-may compromise educational integrity, clinical judgment, patient safety, and trust. This scoping review mapped evidence on their nature, detection and validation, mitigation, and educational and governance implications.Methods This scoping review was conducted in accordance with the Joanna Briggs Institute methodology and informed by the Arksey and O’Malley framework. Reporting followed the PRISMA Extension for Scoping Reviews. PubMed/MEDLINE, Scopus, IEEE Xplore, and Web of Science were systematically searched for English-language sources published from January 1, 2018, to July 31, 2026. Study selection and data extraction were independently performed by two reviewers. Findings were synthesized using descriptive thematic analysis.Results Of 2,051 identified records, 21 sources were included in the final synthesis after duplicate removal and screening. These included five empirical or model-evaluation studies and 16 review, conceptual, or opinion-based sources. The analysis identified five main themes and 13 subthemes: (1) conceptualization and manifestations of AI hallucinations in healthcare; (2) detection, assessment, and validation of hallucinations; (3) mitigation strategies for improving the reliability of generative AI outputs; (4) educational implications and critical AI literacy; and (5) governance, accountability, and trustworthy implementation in healthcare. Retrieval-augmented generation, structured prompt design, output-verification methods, and human oversight were among the most frequently identified approaches for reducing hallucination-related risks.Conclusion AI hallucinations are not merely technical errors; they represent a multidimensional challenge to the safe and responsible use of LLMs in medical education and healthcare. Reducing their risks requires a multilayered approach that includes evidence-based validation, human oversight, critical AI literacy, redesign of educational assessment, and transparent governance and accountability frameworks.
SN - 1087-2981
ER -