ML Security PapersQuantifying Conversation Drift in MCP via Latent Polytope
Haoran Shi 1, Hongwei Yao 2,3, Shuo Shao 1, Shaopeng Jiao 4, Ziqi Peng 5, Zhan Qin 1,3, Cong Wang 3
Published on arXiv
2508.06418
Insecure Plugin Design
OWASP LLM Top 10 — LLM07
Prompt Injection
OWASP LLM Top 10 — LLM01
Key Finding
SecMCP detects indirect prompt injection and tool poisoning attacks in MCP-enabled LLMs with AUROC scores consistently exceeding 0.915 across all evaluated models and datasets.
SecMCP (latent polytope conversation drift detection)
Novel technique introduced
The Model Context Protocol (MCP) enhances large language models (LLMs) by integrating external tools, enabling dynamic aggregation of real-time data to improve task execution. However, its non-isolated execution context introduces critical security and privacy risks. In particular, adversarially crafted content can induce tool poisoning or indirect prompt injection, leading to conversation hijacking, misinformation propagation, or data exfiltration. Existing defenses, such as rule-based filters or LLM-driven detection, remain inadequate due to their reliance on static signatures, computational inefficiency, and inability to quantify conversational hijacking. To address these limitations, we propose SecMCP, a secure framework that detects and quantifies conversation drift, deviations in latent space trajectories induced by adversarial external knowledge. By modeling LLM activation vectors within a latent polytope space, SecMCP identifies anomalous shifts in conversational dynamics, enabling proactive detection of hijacking, misleading, and data exfiltration. We evaluate SecMCP on three state-of-the-art LLMs (Llama3, Vicuna, Mistral) across benchmark datasets (MS MARCO, HotpotQA, FinQA), demonstrating robust detection with AUROC scores exceeding 0.915 while maintaining system usability. Our contributions include a systematic categorization of MCP security threats, a novel latent polytope-based methodology for quantifying conversation drift, and empirical validation of SecMCP's efficacy.
Key Contributions
- Systematic categorization of MCP security threats into three primary attack classes: conversation hijacking, misleading, and data exfiltration
- SecMCP framework that models LLM activation vectors in a latent polytope space to detect and quantify adversarial conversation drift induced by malicious external tool content
- Empirical validation on three LLMs (Llama3, Vicuna, Mistral) and three benchmark datasets (MS MARCO, HotpotQA, FinQA), achieving AUROC > 0.915 with negligible usability impact