Topology Matters: Measuring Memory Leakage in Multi-Agent LLMs

TL;DR

This paper introduces MAMA, a framework to quantify how different network topologies in multi-agent LLM systems influence memory leakage, revealing that denser and more connected structures increase vulnerability.

Contribution

The study systematically evaluates the impact of various graph topologies on memory leakage in multi-agent LLMs using a novel measurement framework, providing practical design guidance.

Findings

Denser connectivity increases leakage.

Leakage is higher with shorter attacker-target distances.

Most leakage occurs early and then plateaus.

Abstract

Graph topology is a fundamental determinant of memory leakage in multi-agent LLM systems, yet its effects remain poorly quantified. We introduce MAMA (Multi-Agent Memory Attack), a framework that measures how network structure shapes leakage. MAMA operates on synthetic documents containing labeled Personally Identifiable Information (PII) entities, from which we generate sanitized task instructions. We execute a two-phase protocol: Engram (seeding private information into a target agent's memory) and Resonance (multi-round interaction where an attacker attempts extraction). Over 10 rounds, we measure leakage as exact-match recovery of ground-truth PII from attacker outputs. We evaluate six canonical topologies (complete, ring, chain, tree, star, star-ring) across $n\in\{4,5,6\}$, attacker-target placements, and base models. Results are consistent: denser connectivity, shorter attacker-target distance, and higher target centrality increase leakage; most leakage occurs in early rounds and then plateaus; model choice shifts absolute rates but preserves topology ordering; spatiotemporal/location attributes leak more readily than identity credentials or regulated identifiers. We distill practical guidance for system design: favor sparse or hierarchical connectivity, maximize attacker-target separation, and restrict hub/shortcut pathways via topology-aware access control.