MADBench is a benchmark that evaluates the security of Multi-Agent Debate (MAD) systems, which allow large language models to exchange and critique answers. The study categorizes attacks into a layered taxonomy aligned with the MAD workflow and tests six attack families across 356 source tasks and 3,958 test cases. Results indicate that while MAD can reduce answer accuracy attacks compared to single-agent baselines, it may amplify unauthorized reads or writes, and even with collusion among agents, the final answer changes from correct to wrong only 28.30% of the time.
By Yuwan Liu, Jiaming Zhang, Yue Huang, Sisi Duan
arXiv:2606. 13385v1 Announce Type: cross Abstract: Web agents driven by large language models (LLMs) are increasingly deployed in real-world environments, where they operate over untrusted web content and execute actions with direct consequences.
By Zihao Wang, Yiming Li, Yutong Wu, Zheyu Liu, Kangjie Chen, Fok Kar Wai, Pin-Yu Chen, Vrizlynn L. L. Thing, Bo Li, Dacheng Tao, Tianwei Zhang
arXiv:2510. 20963v2 Announce Type: replace Abstract: Multi-agent debate (MAD) was proposed as a promising approach for ensembling the wisdom of multiple large language models (LLMs) to improve reasoning and provide effective supervision to superhuman LLMs.
By Yongqiang Chen, Gang Niu, James Cheng, Bo Han, Masashi Sugiyama
POIROT (Phased Oversight via Introspection, Review and Opinion Tallying) is a modular protocol that lets agents in large language model multi‑agent systems attribute failures to themselves by casting private votes in a hazard space and aggregating those votes. The protocol progresses through phases—from independent self‑assessment to shared assessments and peer interrogation—to evaluate the contribution of each phase. Experiments on the Who&When benchmark and the newly released BLAME benchmark show that agent‑based attribution, especially the independent self‑assessment phase, outperforms a single‑LLM evaluator across multiple models and benchmarks.
By I\~naki Dellibarda Varela, R. Sendra-Arranz, Pablo Romero-Sorozabal, J. M. Valverde-Garc\'ia, Annemarie F. Laudanski, \'Alvaro Guti\'errez, Eduardo Rocon, Manuel Cebrian
arXiv:2501. 00745v3 Announce Type: replace-cross Abstract: The increasing integration of Large Language Model (LLM) based search engines has transformed the landscape of information retrieval.
By Xiyang Hu
The paper introduces response manipulation attacks (ReMA) that iteratively optimize malicious agent responses to compromise multi‑agent system outputs. It proposes SecMAS, a secure aggregation rule that selects responses most consistent with a large subset of others, providing theoretical security guarantees against both untargeted and targeted attacks. Experiments show existing aggregation rules are vulnerable to ReMA, while SecMAS defends effectively without sacrificing utility.
By Xilong Wang, Xiaoxing Chen, Neil Gong
The paper "Leaderboard Hacking: Preference-Based Model Evaluations are Vulnerable to Manipulation" examines how arena-style rankings of language models, which rely on pairwise comparisons and the Bradley‑Terry model, can be manipulated. It identifies three attack vectors—strategic voting, strategic prompting, and strategic nomination—that can each boost a model’s rank by 15‑20 places, and combined attacks can improve a rank by up to 45 places. The authors analyze the conditions that enable these vulnerabilities and propose safeguards for any preference‑based evaluation that permits open prompting, voting, or model submission.
By Eve Fleisig, Joachim Baumann, Dylan Hadfield-Menell, Dirk Hovy
Multi-agent systems solve complex tasks by decomposing them into subtasks, delegating the subtasks to agents, and aggregating the agents' responses using an aggregation rule. Malicious agents can mani...
arXiv:2605. 01133v3 Announce Type: replace-cross Abstract: Large language model (LLM)-powered multi-agent systems (MAS) enable agents to communicate and share information, achieving strong performance on complex tasks.
By Lingxi Zhang, Guangtao Zheng, Hanjie Chen
The paper introduces a test-time aggregation framework for multi-agent reasoning with large language models. It mitigates non-semantic biases by normalizing reasoning traces and injects bias to create decision instability, which signals reliability. The method improves accuracy and token efficiency on several mathematical reasoning benchmarks without requiring validation data or prior model strength knowledge.
By Sungjae Lee, Soheil Abbasloo, Jabin Koo, Jungseul Ok
arXiv:2508. 16481v3 Announce Type: replace Abstract: Ensuring the safe use of agentic systems requires a thorough understanding of the range of malicious behaviors these systems may exhibit.
By Jonathan N\"other, Adish Singla, Goran Radanovic
The paper introduces Bayesian Dialectical Argumentation (BDA), a method for aggregating answers from multiple large language models (LLMs) in a council setting. BDA treats each LLM’s typed moves—proposals, challenges, and concessions—as evidence in a classical annotator model, estimating per-agent reliability even when some agents are persistently unreliable. By weighting evidence according to these inferred reliabilities, BDA produces calibrated posterior probabilities for candidate answers and can invert unreliable agents instead of merely outvoting them, achieving superior calibration and robustness on both binary and multi-class benchmarks without extra LLM calls.
By Ionel Eduard Stan, Paolo Napoletano