Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

LLMs for Cyber Security: New Opportunities

Dinil Mon Divakaran, Sai Teja Peddinti

TL;DR

This paper surveys how large language models (LLMs) can address critical cyber security challenges, from vulnerability detection to content moderation and risk mitigation. It highlights five major opportunity themes and reviews concrete methods, such as automatic code repair, phishing detection, and explainable security analytics, enabled by the vast pretraining and prompting capabilities of LLMs. The authors discuss practical data problems, including labeling cost and privacy, and propose data augmentation and domain-specific network traffic modeling with unlabeled data to overcome these barriers. Significant risks, such as prompt injection, deepfakes, and adversarial evasion, are analyzed with guardrails, watermarking, and red-teaming as countermeasures. Overall, the work advocates a balanced, multi-stakeholder approach to safely leverage LLMs for defense, emphasizing human-in-the-loop decision making and governance.

Abstract

Large language models (LLMs) are a class of powerful and versatile models that are beneficial to many industries. With the emergence of LLMs, we take a fresh look at cyber security, specifically exploring and summarizing the potential of LLMs in addressing challenging problems in the security and safety domains.

LLMs for Cyber Security: New Opportunities

TL;DR

This paper surveys how large language models (LLMs) can address critical cyber security challenges, from vulnerability detection to content moderation and risk mitigation. It highlights five major opportunity themes and reviews concrete methods, such as automatic code repair, phishing detection, and explainable security analytics, enabled by the vast pretraining and prompting capabilities of LLMs. The authors discuss practical data problems, including labeling cost and privacy, and propose data augmentation and domain-specific network traffic modeling with unlabeled data to overcome these barriers. Significant risks, such as prompt injection, deepfakes, and adversarial evasion, are analyzed with guardrails, watermarking, and red-teaming as countermeasures. Overall, the work advocates a balanced, multi-stakeholder approach to safely leverage LLMs for defense, emphasizing human-in-the-loop decision making and governance.

Abstract

Large language models (LLMs) are a class of powerful and versatile models that are beneficial to many industries. With the emergence of LLMs, we take a fresh look at cyber security, specifically exploring and summarizing the potential of LLMs in addressing challenging problems in the security and safety domains.
Paper Structure (13 sections, 1 figure)

This paper contains 13 sections, 1 figure.

Table of Contents

  1. Introduction
  2. New opportunity to address cyber security problems
  3. LLMs for Vulnerability Detection and Management
  4. LLMs for Content Classification and Enforcement
  5. Safety Classifiers for Policy Enforcement:
  6. Phishing Detection:
  7. LLMs for Explainability and Prioritization
  8. Explainability:
  9. Prioritization:
  10. LLMs for Tackling Data Challenges
  11. Traffic Modeling for Network Security:
  12. LLMs for Mitigating LLM Risks
  13. Key takeaways

Figures (1)

  • Figure 1: LLMs offer versatile solutions to address a wide range of cyber security challenges.