Attack and defense techniques in large language models: A survey and new perspectives

📅 2025-05-02
📈 Citations: 0
Influential: 0
📄 PDF
🤖 AI Summary
This survey addresses critical security vulnerabilities in large language models (LLMs), including adversarial prompting, model extraction, and application-layer attacks—alongside their corresponding defenses. It identifies key limitations in existing defensive approaches: insufficient adaptability to evolving threats, suboptimal robustness–utility trade-offs, and poor resource efficiency. To address these gaps, the work proposes a novel, end-to-end LLM security taxonomy—the first of its kind—spanning the full attack-defense lifecycle. Three foundational research directions are introduced: (1) scalable adaptive defense mechanisms, (2) interpretable security frameworks, and (3) standardized, reproducible evaluation protocols. Methodologically, the study integrates systematic literature review, rigorous threat modeling, and cross-disciplinary insights from NLP, cybersecurity, and AI governance. The resulting synthesis delivers both theoretical rigor and actionable guidance, offering industry practitioners a practical defense roadmap and advancing secure, trustworthy, and responsible LLM deployment paradigms.

Technology Category

Application Category

📝 Abstract
Large Language Models (LLMs) have become central to numerous natural language processing tasks, but their vulnerabilities present significant security and ethical challenges. This systematic survey explores the evolving landscape of attack and defense techniques in LLMs. We classify attacks into adversarial prompt attack, optimized attacks, model theft, as well as attacks on application of LLMs, detailing their mechanisms and implications. Consequently, we analyze defense strategies, including prevention-based and detection-based defense methods. Although advances have been made, challenges remain to adapt to the dynamic threat landscape, balance usability with robustness, and address resource constraints in defense implementation. We highlight open problems, including the need for adaptive scalable defenses, explainable security techniques, and standardized evaluation frameworks. This survey provides actionable insights and directions for developing secure and resilient LLMs, emphasizing the importance of interdisciplinary collaboration and ethical considerations to mitigate risks in real-world applications.
Problem

Research questions and friction points this paper is trying to address.

Exploring attack and defense techniques in Large Language Models (LLMs)
Analyzing vulnerabilities and security challenges in LLM applications
Identifying adaptive defenses and ethical solutions for LLM security
Innovation

Methods, ideas, or system contributions that make the work stand out.

Classify attacks into adversarial and optimized types
Analyze prevention and detection defense strategies
Highlight adaptive scalable defenses and explainable security
💼 Related Jobs
No related jobs found.
Z
Zhiyu Liao
School of Computer Engineering, Jimei University, Xiamen, China
K
Kang Chen
College of Science, Mathematics and Technology, Wenzhou-Kean University, Wenzhou, China
Y
Yuanguo Lin
School of Computer Engineering, Jimei University, Xiamen, China
K
Kangkang Li
School of Smart Education, Jiangsu Normal University, Xuzhou, China
Y
Yunxuan Liu
School of Computer Engineering, Jimei University, Xiamen, China
H
Hefeng Chen
School of Computer Engineering, Jimei University, Xiamen, China
X
Xingwang Huang
School of Computer Engineering, Jimei University, Xiamen, China
Y
Yuanhui Yu
School of Computer Engineering, Jimei University, Xiamen, China