Beyond Input Guardrails: Reconstructing Cross-Agent Semantic Flows for Execution-Aware Attack Detection

TL;DR

This paper introduces \\SysName, a novel framework for detecting complex cross-agent attacks in multi-agent systems by reconstructing and analyzing semantic flows, moving beyond traditional input filtering methods.

Contribution

It presents a new execution-aware analysis approach that reconstructs semantic flows and uses a supervisor LLM to detect sophisticated attack vectors in multi-agent systems.

Findings

Achieves 85.3% node-level attack detection F1-score.

Detects over ten different compound attack types.

Provides a holistic view of system activity for improved security.

Abstract

Multi-Agent System is emerging as the \textit{de facto} standard for complex task orchestration. However, its reliance on autonomous execution and unstructured inter-agent communication introduces severe risks, such as indirect prompt injection, that easily circumvent conventional input guardrails. To address this, we propose \SysName, a framework that shifts the defensive paradigm from static input filtering to execution-aware analysis. By extracting and reconstructing Cross-Agent Semantic Flows, \SysName synthesizes fragmented operational primitives into contiguous behavioral trajectories, enabling a holistic view of system activity. We leverage a Supervisor LLM to scrutinize these trajectories, identifying anomalies across data flow violations, control flow deviations, and intent inconsistencies. Empirical evaluations demonstrate that \SysName effectively detects over ten distinct compound attack vectors, achieving F1-scores of 85.3\% and 66.7\% for node-level and path-level end-to-end attack detection, respectively. The source code is available at https://anonymous.4open.science/r/MAScope-71DC.