Work

A Four-Axis Systematization of AI and Cybersecurity in the Era of Large Language Models

Year: 2026
Type: conference-paper
Study design: Systematic Review
Abstract: The convergence of artificial intelligence (AI) and cybersecurity has created a complex ecosystem of capabilities and threats. We systematize this space along four axes: AI-for-Cyber, Cyber-for-AI, LLM-for-Cyber, and Cyber-for-LLMs. Using a PRISMA-guided review of 422 recent works, we develop a unified taxonomy of AI–cyber interactions. AI-for-Cyber enables anomaly detection, intrusion detection, vulnerability analysis, and automated response. Cyber-for-AI highlights threats such as data poisoning, model extraction, and privacy leakage. Cyber-for-LLMs exposes vulnerabilities including prompt injection and jailbreaking, while LLM-for-Cyber introduces reasoning-driven capabilities for threat intelligence and offensive operations. We further identify compound cross-axis threats, where attackers combine techniques across axes (e.g., LLM-generated attacks exploiting AI defenses). This Systematization of Knowledge (SoK) consolidates fragmented literature, highlights systemic risks, and outlines open challenges for securing AI-enabled cyber systems.
Language: English
Open Access status: hybrid
OpenAlex ID: W7196952257
FWCI:
Cites: 165
Cited by:
Related to: 10