CAPS: Compositional Attack Path Scoring for LLM Deployment Stacks

Abstract

Evaluating the security posture of large language model (LLM) deployment stacks is a critical challenge in modern AI security. Traditional vulnerability management frameworks—such as the Common Vulnerability Scoring System (CVSS) and component-level checklists—assume that software components can be evaluated in isolation. In real-world agentic and retrieval-augmented generation (RAG)-based LLM ecosystems, this assumption is systematically violated: attackers exploit complex topologies, chaining

Categories

@article{dang2026caps,
  title = {{CAPS: Compositional Attack Path Scoring for LLM Deployment Stacks}},
  author = {Quang-Vinh Dang and Hoang-Viet Vu and Ngoc-Son-An Nguyen and M. Dinh and Dat Le},
  year = {2026},
  month = aug,
  journal = {Artificial Intelligence and Applications},
  doi = {10.47852/bonviewaia620210609},
  url = {https://www.semanticscholar.org/paper/33eca5ed0e766089b5c28a42428b804e6a8e2bf2}
}