Research map / Multi-agent systems and games / LLM agents
Is Monitoring Enough? Strategic Agent Selection For Stealthy Attack in Multi-Agent Discussions
Qiuchi Xiang, Haoxuan Qu, Hossein Rahmani, Jun Liu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace original agent selection with FJ-based strategy | MCA · MMLU, SelfCheckGPT | Delta Acc: 2.6% -> 28.7% | Improved |
| replace original agent selection with FJ-based strategy | MAD-Spear · MMLU, SelfCheckGPT | Delta Acc: 6.9% -> 43.5% | Improved |
| replace original agent selection with FJ-based strategy | MCA · MMMU, SelfCheckGPT | Delta Acc: 4.1% -> 32.6% | Improved |
| replace original agent selection with FJ-based strategy | MAD-Spear · MMMU, SelfCheckGPT | Delta Acc: 7.6% -> 45.3% | Improved |
| add stealth-oriented prompt modifications to MAD-Spear | MAD-Spear · MMLU, SelfCheckGPT | Delta Acc: detection rate 97.8% -> 6.7%; Delta Acc 1.5% -> 6.9% | Mixed |
| add stealth-oriented prompt modifications to MCA | MCA · MMLU, SelfCheckGPT | Delta Acc: detection rate 95.1% -> 4.5%; Delta Acc 0.7% -> 2.6% | Mixed |
| compare best heuristic Variant III with our strategy | MAD-Spear · MMLU, SelfCheckGPT | Delta Acc: 21.8% -> 43.5% | Improved |
| add stubborn-broadcaster modification (1) to FJ formulation (ablation) | MAD-Spear · MMMU, SelfCheckGPT | Delta Acc: 40.2% -> 45.3% | Improved |
| add selective-influence modification (2) to FJ formulation (ablation) | MAD-Spear · MMMU, SelfCheckGPT | Delta Acc: 35.1% -> 45.3% | Improved |
| add both modifications to standard FJ model (ablation) | MAD-Spear · MMMU, SelfCheckGPT | Delta Acc: 25.7% -> 45.3% | Improved |
| add both modifications to standard FJ model (ablation) | MAD-Spear · MMMU, G-Safeguard | Delta Acc: 22.2% -> 42.2% | Improved |
| add dedicated solver vs naive enumeration | formulation solver · solving time | Delta Acc: practically infeasible -> 0.54 s | Improved |
Closest papers by meaning
- Monitoring Decomposition Attacks with Lightweight Sequential Monitors (ICLR 2026 · similarity 0.86)
- When Embedding-Based Defenses Fail: Rethinking Safety in LLM-Based Multi-Agent Systems (ICML 2026 · similarity 0.85)
- When Agents Go Rogue: Activation-Based Detection of Malicious Behaviors in Multi-Agent Systems (ICML 2026 · similarity 0.84)
- Securing Multi-Agent Systems Against Corruptions via Node Contribution Backpropagation (ICML 2026 · similarity 0.84)
- Attack the Messages, Not the Agents: A Multi-round Adaptive Stealthy Tampering Framework for LLM-MAS (AAAI 2026 · similarity 0.84)
- Agent-ToM: Learning to Monitor Autonomous LLM Agents via Theory-of-Mind Reasoning (NeurIPS 2026 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
