Concealing LLM-Based Multi-Agent Topology via Phantom Structure Injection
MIRAGE hides genuine multi-agent communication topology behind a phantom structure to blunt inference attacks.
The paper proposes MIRAGE, a framework that conceals the communication topology of LLM-based multi-agent systems from black-box inference attacks. It synthesizes a structurally different phantom topology, realizes phantom edges as plausible semantic dependencies, and suppresses cues that would expose genuine edges. Experiments across three topology-optimization frameworks and four benchmarks show substantially weaker topology inference while largely preserving task utility.
- Black-box observers can infer multi-agent topologies from reasoning traces.
- MIRAGE keeps the real topology while presenting a distinct phantom topology.
- Tests cover three optimization frameworks and four benchmark datasets.
- Inference attacks become less effective while task utility is largely preserved.
From the original article
1 min read at arxiv.orgDriven by the rapid advancement of large language models (LLMs), LLM-based multi-agent systems (MAS) have emerged as a powerful paradigm for collaborative reasoning over complex tasks. A key design element of MAS is the communication topology, which governs information flow among agents and often encodes proprietary knowledge about the system architecture.
By Longzhu He · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.