Secure and Enhanced Cyber-Threat Detection in IoMT Using Locally Deployed Large Language Models

Area of Science:

• Cybersecurity
• Health Informatics
• Artificial Intelligence

Background:

• The Internet of Medical Things (IoMT) is rapidly expanding, integrating advanced technologies like image processing and large language models (LLMs).
• Processing patient-identifiable information (PII) in cloud environments presents significant security and compliance challenges.
• Current threat detection tools offer basic protection but fail to identify complex vulnerabilities, and cloud-based LLMs compromise data privacy.

Purpose of the Study:

• To propose a novel, locally deployed, and cost-efficient LLM solution for enhancing IoMT cybersecurity.
• To improve the detection of complex cyberthreats and provide mitigation strategies.
• To address the privacy concerns associated with cloud-based LLM approaches.

Main Methods:

• Implementation of a locally deployed Llama 3.1 LLM (8B parameters) for cybersecurity threat analysis.
• Comparison of the local LLM's detection capabilities against existing tools like SPYDERISK.
• Evaluation of the model's effectiveness in identifying additional cyberthreats and recommending mitigation strategies.

Main Results:

• The locally deployed Llama 3.1 model successfully identified 12 additional cyberthreats beyond those detected by SPYDERISK.
• The enhanced detection coverage significantly improved the overall security posture for IoMT environments.
• The privacy-preserving nature of local deployment was maintained throughout the threat detection process.

Conclusions:

• Local deployment of Llama 3.1 offers a superior and privacy-preserving method for IoMT cybersecurity.
• This approach effectively addresses the limitations of existing tools and cloud-based LLM solutions.
• The study demonstrates the potential of localized LLMs for robust threat identification and mitigation in sensitive healthcare settings.