Cybersecurity and AI: A Comprehensive Implementation of Advanced Large Language Models for Threat Detection, Digital Forensics, SOC Automation, and Security Vulnerability Mitigation

• Harsh Dankhara

  Author

This report encompasses the cutting-edge implementation of advanced Large Language Models (LLMs) for cybersecurity and digital forensics applications at the intersection of cybersecurity and AI. The project brings together innovative research on AI-based security solutions in four major areas: threat detection and intelligence analysis, digital forensics and incident response, Security Operations Centre (SOC) automation and mitigation of security challenges in cybersecurity frameworks. Our implementation is capable of detecting threats with greater than 94% accuracy. It decreases the workload of a SOC by 70%. Most importantly, it mitigates various serious threats, like the OWASP LLM01 prompt injection attack. The system combines and integrates the use of various models, such as ForensicLLM (4-bit quantized LLaMA-3.1-8B) and custom security agents, serving as a one-stop solution for cyber experts in industry and law enforcement. The ForensicLLM is used only for correlating digital-forensics evidence and generating investigative narrative; threat detection (general) is done through a fine-tuned BERT classifier; SOC incident-severity classification is performed through a fine-tuned RoBERTa model. Contribution of this work contributes to the rapidly expanding field of cybersecurity and AI through practical implementations, performance benchmarks and frameworks for their responsible deployment in security-critical environments. Apart from that, we illustrate that large language models can be effectively used as force multipliers for under-resourced security teams while incorporating strong operational guardrails.

• PDF