T-MAP: Red-Teaming LLM Agents with Trajectory-aware Evolutionary Search

TL;DR

T-MAP introduces a trajectory-aware evolutionary search technique to identify vulnerabilities in LLM agents by generating adversarial prompts that exploit multi-step tool interactions, revealing new security risks.

Contribution

The paper presents T-MAP, a novel method leveraging execution trajectories for more effective red-teaming of autonomous LLM agents, especially in complex ecosystems like MCP.

Findings

T-MAP significantly improves attack success rates over baselines.

Effective against state-of-the-art models including GPT-5.2 and Gemini-3-Pro.

Reveals previously unknown vulnerabilities in autonomous LLM systems.

Abstract

While prior red-teaming efforts have focused on eliciting harmful text outputs from large language models (LLMs), such approaches fail to capture agent-specific vulnerabilities that emerge through multi-step tool execution, particularly in rapidly growing ecosystems such as the Model Context Protocol (MCP). To address this gap, we propose a trajectory-aware evolutionary search method, T-MAP, which leverages execution trajectories to guide the discovery of adversarial prompts. Our approach enables the automatic generation of attacks that not only bypass safety guardrails but also reliably realize harmful objectives through actual tool interactions. Empirical evaluations across diverse MCP environments demonstrate that T-MAP substantially outperforms baselines in attack realization rate (ARR) and remains effective against frontier models, including GPT-5.2, Gemini-3-Pro, Qwen3.5, and GLM-5, thereby revealing previously underexplored vulnerabilities in autonomous LLM agents.