The Attack and Defense Landscape of Agentic AI: A Comprehensive Survey

ABSTRACT

This comprehensive survey analyzes the emerging security landscape surrounding agentic AI systems capable of autonomous reasoning, tool usage, and multi-step decision-making. The authors categorize threat vectors such as prompt injection, tool-chain manipulation, environment poisoning, and agent coordination failures. The paper synthesizes current research on defensive architectures designed to mitigate these risks. It also outlines future research directions for building secure, resilient agent ecosystems.