Large Language Model Security in Cyber Defense: A Systematic Review of Attack Surfaces, Adversarial Threats, Data Exposure, and Adaptive Defense in RAG- and Agentic Architectures
Abstract
Large Language Models (LLMs) are becoming integral to cyber-defense processes, including threat intelligence, security event analysis, incident investigation, decision support, and automated response. As implementations incorporate retrieval-augmented generation (RAG), persistent memory, external tools, and agentic execution, the security landscape broadens beyond mere model parameters to include data, context, retrieval, state, authorization, tool interfaces, and underlying infrastructure. This systematic review compiles security literature across these dimensions and categorizes findings based on attack surface, adversarial mechanisms, security implications, defense strategies, evaluation methods, and adaptability capabilities. The review examined 1,945 records, eliminated 527 duplicates, screened 1,418 unique entries, analyzed 455 full texts, and selected 168 studies for qualitative synthesis. It highlights key threat classes such as prompt and indirect injection, jailbreaks, training and retrieval-data poisoning, information leakage, retrieval manipulation, memory contamination, tool misuse, and manipulation of agentic actions. Additionally, it reveals that security vulnerabilities can transcend layers: content controlled by attackers can impact retrieval and context, modify model reasoning, and influence downstream tool or agent responses. To address these complexities, the review puts forward the Attack Surface–Threat–Impact–Defense–Adaptation (ASTIDA) framework as a…
How strong is the evidence?
Evidence Hub reads this as a systematic review. Its evidence level is 1 of 5 (Highest · pooled evidence). Pooled evidence from several studies sits at the top of the evidence pyramid.
Read the paper
Related studies
- Knowledge Management Perspective of Generative Artificial Intelligence
- Navigating Generative Artificial Intelligence Promises and Perils for Knowledge and Creative Work
- Characterizing the evolving research on enterprise content management
- Large language models and multimodal foundation models for precision oncology
- From Generative AI to Extended Reality: Multidisciplinary Perspectives on the Challenges, Opportunities, and Future of Educational Computing
- Generative Artificial Intelligence: A New Engine for Advancing Environmental Science and Engineering
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.