Invisible Threats from Model Context Protocol: Generating Stealthy Injection Payload via Tree-based Adaptive Search

Researchers have discovered a new black-box attack method called Tree structured Injection for Payloads (TIP) that can compromise AI agents using Model Context Protocol with over 95% success rate. The attack exploits vulnerabilities in how large language models interact with external tools, bypassing existing defenses and requiring significantly fewer queries than previous methods.

• →TIP attack achieves over 95% success rate against undefended MCP-enabled AI agents using natural-looking payloads.
• →The method requires an order of magnitude fewer queries compared to existing adaptive attacks.
• →Even against four defensive approaches, TIP maintains more than 50% effectiveness.
• →The attack exposes a critical security vulnerability in real-world Model Context Protocol deployments.
• →Researchers successfully demonstrated the attack on mainstream LLMs including four major models.