Three “T’s” of Large Language Models in Cybersecurity : LLMs As a Tool, LLMs As a Target, LLMs As a Threat, Offensive and Defensive Applications Of LLMs In Cybersecurity Systems Bakhtiyorov Boburjon Bakhtiyor Ugli
- 주제(키워드) Large language models application in defensive and offensive cybersecurity
- 발행기관 아주대학교 일반대학원
- 지도교수 Min Jae Park
- 발행년도 2025
- 학위수여년월 2025. 2
- 학위명 석사
- 학과 및 전공 일반대학원 비즈니스애널리틱스학과
- 실제URI http://www.dcollection.net/handler/ajou/000000034730
- 본문언어 영어
- 저작권 아주대학교 논문은 저작권에 의해 보호받습니다.
초록/요약
In the last few years, Large Language Models have been the main topic of discussion and demonstrated extensive impact across various fields including cybersecurity. The evolution of Generative AI(GenAI) and its subset of Large Language Models(LLMs - GPT-4, LLama, Gemini, PaLM) have been transforming all the industries and at the same time enhancing the complexity of the cybersecurity field. This research paper explores the multifaceted role of LLMs in cybersecurity, addressing their limitations, challenges, potential risks, and opportunities. It focuses on three primary dimensions: LLMs as a tool for enhancing cybersecurity defenses, as a target vulnerable to exploitation, and as a threat that malicious actors can leverage for cyber-attacks. The work analyzes vulnerabilities within LLMs, including exploitable scenarios where malicious users may bypass ethical constraints through techniques like jailbreaks, prompt injections, and reverse psychology attacks. As for the defensive side, the paper further explores how LLMs might strengthen cybersecurity measures through threat intelligence, automated attack identification, secure code generation, incident response, and malware detection. Additionally this paper examines survey of highly skilled 196 cybersecurity professionals on the usage and implementation of LLMs into their everyday cybersecurity related tasks which helps get insights various views such as “Skeptics”, “Adopters”, “Potential Adopters” and as to why the integration of LLMs have been challenging along with expected trends for the usage of LLMs or overall GenAI in cybersecurity from their perspective.
more목차
Chapter 1: Introduction 1
1.1. Brief Look Into The Development Of GenAI and LLMs 2
1.2. Research Outline 5
Chapter 2: Literature Review 6
2.1. Types of traditional cyberattacks 6
2.2. LLM enhanced cyber attack 8
2.3. LLMs for cyber defense 12
Chapter 3: Research Methodology And Research Results 17
3.1. Research methodology outline 17
3.2. RQ1: 17
3.3. RQ2: 21
3.4. CompDoc: 26
3.5. ComDoc enhanced with t=bio Memory Injection 30
3.6. Experiment results 31
3.7. RQ3: 38
3.8. Survey Design and Objectives 38
3.9. Target Audience and Sample Size 38
3.10. Data Collection 39
3.11. Survey Findings and Data Analysis 40
Chapter 4: Discussion 49
4.1. Summary of findings 49
4.2. Academic Contribution: 50
4.3. Limitations: 51
4.4. Future Research Directions: 52
Chapter 5: Conclusion 53
Appendix 55
Reference 57
