An Exploration of Attacks on AI Agent Systems
The paper explores how prompt injection attacks can compromise AI agent systems, focusing on both direct user inputs and inter-agent messages that can leak sensitive data or make malicious behavior spread across agents. Its main value is framing prompt injection as a multi-agent security problem with infection-like propagation rather than just a single-chatbot issue, though the work sounds mostly exploratory and weighted toward single-agent experiments rather than a comprehensive evaluation of real deployed agent frameworks.
Abstract
As autonomous AI agents become more integrated into sensitive domains, it is crucial that single and multi-agent AI systems are secured. This paper investigates how prompt injection attacks can exploit both direct user inputs and inter?agent communication to leak sensitive information and trigger cross-agent infections. While our tests mostly highlight key vulnerabilities in single-agent systems, we also extend these attacks to multi-agent frameworks, investigating how weaknesses can propagate across agents. Our findings emphasize the need for more robust defenses as AI agents become more commonly embedded in real-world workflows.