A Review of Large Language Models in Healthcare: Taxonomy, Threats, Vulnerabilities, and Framework

Abstract

Due to the widespread acceptance of ChatGPT, implementing large language models (LLMs) in real-world applications has become an important research area. Such productisation of technologies allows the public to use AI without technical knowledge. LLMs can revolutionise and automate various healthcare processes, but security is critical. If implemented in critical sectors such as healthcare, adversaries can manipulate the vulnerabilities present in such systems to perform malicious activities such as data exfiltration and manipulation, and the results can be devastating. While LLM implementation in healthcare has been discussed in numerous studies, threats and vulnerabilities identification in LLMs and their safe implementation in healthcare remain largely unexplored. Based on a comprehensive review, this study provides new findings which do not exist in the current literature. This research has proposed a taxonomy to explore LLM applications in healthcare, a threat model considering the vulnerabilities of LLMs which may affect their implementation in healthcare, and a security framework for the implementation of LLMs in healthcare and has identified future avenues of research in LLMs, cybersecurity, and healthcare.

Keywords

• large language models
• ChatGPT
• taxonomy
• threat model
• healthcare
• vulnerabilities