Adversarial Threats to AI-Augmented Security Operations: A Systematic Survey
Abstract
Abstract The integration of large language models into enterprise security operations centers has created a new and largely unexamined attack surface. This survey systematically maps adversarial threats targeting AI-augmented security systems, including retrieval-augmented generation poisoning, prompt injection, multi-agent hijacking, and concept drift under adversarial conditions. We propose a unified taxonomy of attack classes, evaluate existing defense mechanisms, and identify critical open problems. To address the chief gap identified—the absence of a standardized evaluation framework for LLM-integrated SOC environments—we introduce RedSOC, an open-source adversarial evaluation framework, and report empirical results from a 15-scenario benchmark spanning three attack classes. RedSOC achieves 100% detection across all evaluated scenarios while revealing significant variation in attack success rates: indirect prompt injection achieves 100% attack success against undefended RAG pipelines, corpus poisoning achieves 80%, and direct prompt injection achieves 60%, yielding an overall attack success rate of 80%.
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.