Forewarned is Forearmed: A Survey on Large Language Model-based Agents in Autonomous Cyberattacks

TL;DR

This survey addresses the problem of autonomous LLM-based cyberattack agents that can execute complex intrusions with reduced human input, driving Cyber Threat Inflation. It presents a unified architecture decomposing agents into models, perception, memory, reasoning, and actions, and then examines multi-agent collaboration and defense lessons. It contributes a network-centric taxonomy of eight cyberattack capabilities, analyzes threat bottlenecks across static, mobile, and infrastructure-free networks, and outlines future research directions and defensive strategies for legacy systems. The work's significance lies in guiding blue teams and policymakers to anticipate AI-enabled threats, adapt threat models, and develop defense-in-depth, governance, and deception-based countermeasures. Overall, the paper emphasizes the need for proactive, adaptable protection as LLM-driven agents reshape the cybersecurity landscape across diverse network paradigms.

Abstract

With the continuous evolution of Large Language Models (LLMs), LLM-based agents have advanced beyond passive chatbots to become autonomous cyber entities capable of performing complex tasks, including web browsing, malicious code and deceptive content generation, and decision-making. By significantly reducing the time, expertise, and resources, AI-assisted cyberattacks orchestrated by LLM-based agents have led to a phenomenon termed Cyber Threat Inflation, characterized by a significant reduction in attack costs and a tremendous increase in attack scale. To provide actionable defensive insights, in this survey, we focus on the potential cyber threats posed by LLM-based agents across diverse network systems. Firstly, we present the capabilities of LLM-based cyberattack agents, which include executing autonomous attack strategies, comprising scouting, memory, reasoning, and action, and facilitating collaborative operations with other agents or human operators. Building on these capabilities, we examine common cyberattacks initiated by LLM-based agents and compare their effectiveness across different types of networks, including static, mobile, and infrastructure-free paradigms. Moreover, we analyze threat bottlenecks of LLM-based agents across different network infrastructures and review their defense methods. Due to operational imbalances, existing defense methods are inadequate against autonomous cyberattacks. Finally, we outline future research directions and potential defensive strategies for legacy network systems.

Figures (7)

• Figure 1: The outline of this paper.
• Figure 2: LLM-based cyberattack agent construction. This architecture enables the agent to ingest diverse input types, store and retrieve contextual knowledge, adaptively plan multi-stage attacks, and interact with tools to perform cyberattacks.
• Figure 3: The timeline of LLM-based agent development and their increasing capabilities in cyberattacks.
• Figure 4: LLM-based agents’ cyberattack capabilities of phishing and social engineering.
• Figure 5: LLM-based agents' cyberattack capabilities of malware generation.