Overview

Multi-agent debate examines whether language models can reason more reliably by proposing answers, sharing evidence, challenging peers, and revising judgments. Common approaches assign roles such as advocate, critic, and judge; use retrieval and structured evidence checks; or adapt who communicates and when debate stops. Researchers apply these methods to mathematics, question answering, fact-checking, and medical or legal decisions, but results vary across tasks and model groups. A central tension is that debate can uncover errors when agents contribute useful independent information, yet consensus can also spread mistakes, suppress dissent, and add substantial cost.

Latest papers

154 papers16 papers in the last four weeks, down 11% on the four weeks before. 0.1% of all new papers.

Nov 12, 2026

Towards a Theory of Learnability via Debate

Debate has been recently proposed as a training protocol that is hoped to produce more aligned models even as model capabilities scale. The idea is that the human, instead of providing supervision in the form of labels, judges a debate between two models that compete to prove the correctness of opposite labels. It can be shown that functions of a much higher computational complexity can be correctly judged in this way, compared to the complexity of functions that can be directly evaluated using a computationally weak evaluator. This suggests that perhaps debate can be used to teach very complicated functions to models. However, previous work has largely ignored the fact that this is a training protocol, and simply analyzes the game being played between the two debaters for a known judge. We take the first steps towards understanding the learning problem implicit within the protocol, by formulating it as an online learning problem. We show that once we formulate the learning problem precisely, there are cases where the best way to learn given access to a judge is different from running an adversarial debate. This motivates us to define a cooperative version of the protocol, whose learnability we characterize using a combinatorial dimension. We also consider the possibility of the human judge being fallible and occasionally judging the debate incorrectly. We show that a small number of errors in judgement can lead to a large error in learning.
Nov 12, 2026

Rational Alignments of Superintelligences via Debates

Artificial superintelligences (ASI) is artificial intelligences with greater capability. However, as human principal has to use it, we have to ensure that the agent will perform in the interest of the human. We investigate the framework of ASI-assisted question answering where ASI agents know the correct answer but the human does not. We provide some example of mechanism for rational alignment, where the agent utility is unchanged, but the interaction between the agent and the human principal is controlled to be in a certain fashion. We show specifically that a debate framework where the agents can further reason and behave in a subgame perfect equilibrium manner will lead to a more predictable result, and in some case will be favorable.
Nov 6, 2026

Alignment by Confident Debate: Preference Learning from Multi-Agent Debate

Post-training language models for reasoning typically rely on gold answers, task-specific verifiers, or learned reward models, limiting scalability when such signals are scarce. Multi-agent debate offers a self-supervised alternative, but converting debate into reliable training supervision remains challenging: majority voting weights agreeing responses equally, lacks a principled tie-breaking rule, and is ill-defined for open-ended tasks without a canonical answer; likelihood-based scoring can conflate fluency with correctness; and homogeneous committees may reinforce correlated errors. We introduce Alignment by Confident Debate (AbCD), which converts debate traces into preference data without gold answer labels. A sampling-diverse committee debates each problem, after which final answers are clustered by task-level semantic equivalence. We use verifiable tasks (mathematical reasoning, multiple-choice quantitative QA, and executable code-generation) as a controlled test bed: their answer keys or executable checkers let us evaluate the quality of debate-derived supervision against ground truth, even though the oracle itself never uses gold labels. Post-training on the resulting preferences outperforms inference-only ensembling and debate, as well as SFT on the same traces, across five model families spanning 1B to 8B parameters and four datasets, improving consensus accuracy, with a maximum gain of 21.5pp21.5_{pp}. Ablations show that the approach transfers to a larger model and to code generation, while also revealing limitations of heterogeneous committees and agreement-based weighting. Overall, the results suggest that debate traces can yield useful preference supervision when answer equivalence is combined with confidence-weighted disagreement.
Nov 6, 2026

AIM: Adaptive Interaction in Multi-Agent Debate for Multimodal LLM Inference

Multi-Agent Debate (MAD) improves reasoning on multimodal tasks, but its accuracy and token efficiency depend on how agents interact during debate. A largely overlooked design choice is which interaction modalities agents use to exchange evidence. Existing methods rely on fixed text- or graph-based interaction, which fails to convey fine-grained multimodal evidence and thus degrades accuracy while inflating token cost due to verbose descriptions. To address this, we propose Adaptive Interaction in Multi-Agent Debate (AIM), an adaptive routing framework that dynamically selects the optimal modality combination for inter-agent interaction in multimodal MAD. Beyond text and graph, AIM introduces task-dependent modality interaction, where agents exchange targeted regions of interest from the task-specific multimodal input, such as image regions, audio segments, or spatio-temporal video clips. To determine the best combination for each instance, AIM first generates a structured baseline response and extracts interpretable routing features that capture which evidence the instance requires, how confident the baseline agent is, and how much new evidence each modality would add beyond the baseline response. A lightweight router then uses these routing features to select a modality combination that avoids both under-fusion (i.e., agents miss decisive evidence) and over-fusion (i.e., redundant modalities inflate token cost) risks. Across eight benchmarks spanning three multimodal task families, AIM achieves the highest accuracy. Combining modalities improves accuracy by up to 6.4% over the best single-modality baseline, and adapting the combination per instance adds up to 4.2% over the most accurate fixed combination of each benchmark with up to 53.4% fewer tokens.
Nov 5, 2026

Debate to Deceive: Poisoning LLM's Fact-Checking via Game-Theoretic Multi-Agent Debate

Knowledge attacks on Large Language Models, such as manipulating fact-checking judgments, pose significant risks to AI reliability. Existing approaches -- adversarial suffix/template injection and single-agent self-reflection -- lack robustness across diverse defenses and are prone to confirmation bias and costly trial-and-error, respectively. We propose Debate-to-Deceive (D2D), a multi-agent debate framework grounded in human debate structure and rhetorical style, enabling agents to generate adversarial passages that are semantically coherent, rhetorically effective, and resistant to early confirmation bias. D2D formalizes the adversarial process as an evolutionary game in which agents adaptively evolve their rhetorical strategies (Logos, Pathos, Ethos) through competition. To manage the intractably large LLM output space, we decouple agent behavior into a strategic layer, which selects a rhetorical action from a compact strategy space via evolutionary-inspired learning dynamics, and a generation layer, where the LLM produces a concrete passage conditioned on that strategy and the debate context. On four fact-checking benchmarks (FEVER, HealthVer, CLIMATE-FEVER, and SciFact), D2D substantially outperforms existing attacks and evades most defenses.
Nov 4, 2026

Beyond Risk Cues: A Mechanism-Level Benchmark and Verifiable Debate for Scam Video Detection on the Web

Online scam videos pose significant threats to individuals and society. Existing studies primarily focus on binary classification, offering limited insights into how scams work and the reasoning behind detection decisions. Moreover, they often overlook the challenge of distinguishing scam videos from superficially similar non-scam content, such as advertisements, and the robustness of detection models against misleading arguments. These limitations hinder reliable scam video detection. To address these limitations, we introduce MechScam, a benchmark comprising 6,614 videos covering diverse scam mechanisms and non-scam categories. MechScam provides explanations supported by video evidence, annotations of potential scam elements and their context, and misleading arguments constructed from genuine evidence. Recognizing that genuine evidence can still be used to justify misleading conclusions, we further propose ScamCourt, a debate framework in which trained prosecution and defense debaters present opposing arguments, while a trained judge makes decisions based on a transcript with verified citations. Extensive experiments show that state-of-the-art MLLMs struggle to distinguish scam videos from similar non-scam content. ScamCourt outperforms larger MLLMs and same-backbone fine-tuned baselines, reducing false positives and showing stronger resistance to misleading arguments. Together, MechScam and ScamCourt contribute to more reliable and interpretable scam video detection. The dataset and code are available at https://github.com/SuperYFan/MechScam.
Nov 3, 2026

When Debate Amplifies Risk: Social-Bias Vulnerabilities in Multi-Agent Safety Deliberation

Retrieval-augmented Web applications increasingly employ multi-agent debate (MAD) to aggregate conflicting, noisy, and heterogeneous information, yet deliberation is fundamentally a social process susceptible to interpersonal dynamics. When retrieved Web sources contain latent hazards alongside safe factual content, social biases such as conformity pressure and authority deference can override evidence-grounded safety boundaries. In this work, we investigate whether the social structure of deliberation can paradoxically amplify latent Web safety risks. We introduce SafeConflict-Bias, a controlled counterfactual diagnostic protocol spanning six core Web risk categories, which holds queries and retrieved evidence strictly invariant while isolating conversational majority sway, authority cues, and confidence asymmetry. Empirical evaluation across diverse proprietary and open-source backbones reveals that while isolated models reliably reject unsafe guidance, standard MAD catastrophically capitulates to social pressure---surging unsafe adoption from 0.0% up to 100.0% and systematically silencing safe minority evidence. To resolve this systemic fragility without retraining or debate regeneration, we propose RiskTrace-Debias, an inference-time auditing framework uniting evidence-first precommitment, social-metadata masking, and consistency-gated arbitration. RiskTrace-Debias suppresses worst-case unsafe adoption to ≤2.0%\le 2.0\% across all architectures without inducing over-refusal, while generating verifiable provenance traces to audit interpersonal manipulation in Web deliberation.
Oct 28, 2026

HEDGE: Gated Hypergraph Memory for Scalable Multi-Agent Debate

Multi-agent debate systems require agents to retain and revisit arguments across multiple rounds, yet limited context windows and unstructured memories make relevant information increasingly difficult to recover. Existing flat and pairwise-graph memories also struggle to capture higher-order relations, including multi-agent agreement, conflicting evidence, and evolving consensus. We introduce HEDGE, a selective memory framework that represents debate trajectories as semantic hypergraphs. Hyperedges connect related claims across agents, while intra-hyperedge relational graphs preserve fine-grained argumentative structure. A learned memory selection module (MSM), trained through supervised learning and reinforcement learning, routes each memory candidate to STORE, BACKUP, or OMIT, retaining useful information while limiting redundancy. Across seven diverse benchmarks, HEDGE consistently outperforms strong single-agent, multi-agent, and memory baselines, improving accuracy by 5.0 points on average over the strongest multi-agent system with GPT-4o-mini. The gains reach 8.2 points on conflict-heavy tasks, while HEDGE reduces memory usage by 48% and the number of debate rounds by 34%.
Oct 26, 2026

Replication or Corroboration? An Axiomatic Framework for Evidential Dependence in Multi-Agent Debate

In multi-agent debates, matching reports can arise for two different reasons: agents may repeat the same evidence (replication), or they may provide separate evidential roots that support the same claim (corroboration). A judge must distinguish these two cases. Counting report occurrences gives extra weight to relays, paraphrases, alias resubmissions, and excerpt splits, while merging all matching content avoids this only by also removing genuine corroboration. We formalize this tension as a minimal judgment game in which acquiring new evidence is costly, but replication is cheap. We show that root replication invariance (RRI) is exactly threshold-uniform replication immunity. Under RRI, replication-only deviations cannot change the verdict, and positive-cost replication is strictly dominated, whereas occurrence-sensitive and content-keyed judges admit profitable deviations. We also establish an information boundary: if a judge's observation interface cannot distinguish a replica from a matched fresh root, no judge can both ignore the replica and respond strictly to the fresh root. To preserve this distinction, we introduce the lineage quotient, which removes wording, identity, occurrence multiplicity, and arbitrary root names while keeping distinct roots separate. On orbit-connected domains, it is the coarsest representation sufficient for every deterministic judge invariant to replication and root-label renaming. We further lift this construction to quantitative bipolar argumentation frameworks (QBAFs), characterize strict fresh-root response, prove its propagation along sign-consistent paths under DF-QuAD, and derive a tight closed-form response modulus. Exhaustive audits over 28,224 direct-star profiles, 2,052 multi-layer DAGs, with matched interventions on 277 AVeriTeC-derived profiles, show the predicted separation between replication and corroboration under a supplied partition.
Oct 23, 2026

Distill, Debate, Discover: Grounding Multi-Agent Research Ideation in Graph-Distilled Insights

Large language models (LLMs) are increasingly used to support research idea generation and scientific discovery. Multi-agent debate offers a collaborative approach to this task by bringing together diverse perspectives to develop research ideas. However, a key challenge is how to equip agents with distinct and meaningful expertise to guide their contributions. Inspired by how researchers draw on accumulated research experience during collaboration, we introduce InsightAgents, a framework that constructs literature-grounded agent expertise by distilling insights from graphs of related papers. Guided by insights that trace evolving research approaches, reveal underlying assumptions, and identify unresolved challenges, within a line of research, agents jointly identify a research problem and propose, challenge, and refine potential approaches through debate to develop a research proposal. We evaluate InsightAgents against three research idea generation frameworks on 100 NLP research topics through pairwise comparisons by two LLM judges. In both online and offline knowledge settings, InsightAgents achieves the highest average score across the five evaluation dimensions under both judges.
Oct 21, 2026

Conflicted Records: Typed Adversarial Contracts for Auditable LLM Output Quality Enforcement

LLM-generated records increasingly populate production information systems where they are treated as data assets subject to the same quality expectations as records from operational databases. A single such record can simultaneously satisfy and violate a quality rule: a grounded fact adjacent to a hallucination, a useful answer adjacent to a stereotype. Classical data cleaning cannot resolve this because the two judgments are incompatible over the same span; a single LLM judge cannot either, since it shares the training biases that produced the conflict. Existing interventions such as prompt engineering, chain-of-thought, and RLHF reduce surface error rates but provide no audit trail and cannot distinguish genuine remediation from blanket suppression. The question is whether a deterministic check over structured inter-agent records can catch the defects that referential checks alone miss. We present BiasGuard-D: two adversarial roles drawn from distinct model families exchange typed claim documents, a three-model quorum adjudicates by majority vote, and a combined referential-plus-semantic invariant validates the verdict. The semantic component fires on verdicts referential checks miss; removing the iterative loop drops invariant satisfaction sharply. The deliberation layer transfers from stereotype to hallucination remediation on HaluEval QA with only role-framing changes, raising right-answer recovery by 49 percentage points; the triage gate required a cross-family second auditor to maintain recall on objective hallucination. Every claim, vote, and instruction is stored as a typed record giving operators a provenance trail auditable under emerging AI regulations.
Oct 20, 2026

Value-Rich, Differently Organized: Measuring Value Expression and Organization in AI-Generated Debates

Current evaluation of LLM-generated debate largely focuses on argumentative quality, leaving the value structure of generated arguments underexamined. We introduce a diagnostic metric framework for comparing how values are expressed and organized in human-reference and AI-generated debates. The framework separates value expression, the range of recognizable values a debate invokes, from value organization, how those values are distributed and sustained across turns. Grounded in Schwartz's Basic-10 value theory, it represents debates as multi-level value distributions over argumentative units, utterances, stances, and full debates, enabling metrics for value coverage, alignment, diffusion, and cross-turn dynamics. Across multiple generation systems, AI-generated debates are broadly value-expressive, often activating as many or more value dimensions than matched human references, but they organize these values differently. Their utterances are more locally value-diffuse, and source-blind judges distinguish value clarity from value focus. These findings show that debate evaluation should consider not only value presence, but also value organization. Code and supplementary materials are available at https://github.com/Maoyuan-li/Value-Expression-and-Organization.
Oct 20, 2026

Humans Correct Their Judgments Through Debate, but Weaker AI Models Do Not

How can we effectively oversee AI systems that surpass human intelligence? AI Safety via Debate proposes that two AI systems argue opposing positions and a judge, either a human or a weaker model, decides which is correct. By facilitating supervisors in providing high-quality training signals, debate aims to encourage truthful and safe behavior from AI systems whose capabilities exceed our own. However, human judgment is not neutral: people often rely on biases which advanced AI systems may learn to exploit. Drawing on insights from collective intelligence and social cognition, this paper investigates whether debate can guide human and LLM supervisors toward the truth despite their prior beliefs. We evaluate several protocol configurations. We also explore setups inspired by the wisdom of the crowd such as the number and diversity of judges, as well as interaction mechanisms including deliberation and voting. Overall, we find that debate increases accuracy relative to prior beliefs for human judges, whereas the consultancy baseline, in which a single expert argues for one answer, does not. For weaker language models, however, the results are mixed: debate provides no comparable gains in most settings and can even reduce accuracy. Among multi-judge protocols, weighted voting produces the highest accuracy for human supervisors, whereas for LLMs none of them significantly improves group accuracy, even as the number of judges increases. Notably, the behavior of minority judges in non-diverse deliberation protocols suggests that, under symmetric settings, incorrect positions could be more persuasive than correct ones. These findings support debate as a viable path toward scalable human oversight, but cast doubt on its benefits for weaker AI supervisors, raising concerns about fully automated training and control pipelines in high-stakes settings.
Oct 16, 2026

Think More or Read More? A Token-Budget Theory of Reasoning with Shared Context

Reasoning systems increasingly improve answers by incorporating additional context into the model's input, from multi-agent debate, where agents read their peers' responses, to self-refinement, where a model revisits its own drafts. In these systems, communication consumes context as one agent's output becomes another's input. We study the regime where token reading and generation draw on one fixed token window, so every token spent reading is a token that cannot be spent reasoning. We develop a token-budget theory that formalizes this trade-off, modeling accuracy as a function of the reasoning budget remaining after reading and aggregating votes with a Condorcet jury model that permits correlated errors. For multi-agent debate, the theory gives a closed-form prediction for the optimal number of agents. This quantity is estimated from single-agent measurements, allowing the optimal group size to be determined without running debate, and the same accounting extends to a heterogeneous set of different models. We empirically verify our theory and show that, at the predicted group size, our method outperforms strong baselines under the tightest calibration spend, where debate-calibrated rules can afford only one or two questions. Self-refinement obeys the same accounting, with its own draft in place of a peer message. The theory further predicts that communication pays only when the gain in the accuracy ceiling exceeds the reasoning room spent on reading, with this condition tightening as the window shrinks. Across five open models and four reasoning tasks, self-consistency matches or beats deployed debate at 9393% of 980980 matched window-and-group-size settings, with the gap being largest under tight windows and nearly disappearing at a loose one, consistent with the theory. Under a shared window, how much room a model has left to think matters more than what it reads.
Oct 15, 2026

Believed but Not Said: Measuring Hidden Belief Collapse in Multi-Agent LLM Discussions

LLM systems increasingly rely on groups of agents that discuss, critique, and check one another, assuming that peers will catch individual errors. We test this assumption against a minimal adversary. In a ten-agent factual discussion, one or three adversarial agents repeatedly assert a plausible false answer. After each round we probe each benign agent's private answer through a separate call whose response is hidden from its peers, and we separately label its public stance toward the claim. Across six models and three factual-QA datasets, groups remain stable without an attacker: only 0.6% of benign agents finish with the false answer. A single adversarial agent increases this rate to 44.8% and three to 60.5%. Studies of conformity typically report public agreement without private change; we find the reverse. With three adversarial agents present, benign agents hold the false answer in 51.6% of exposed turns but publicly endorse it in only 19.7%. Much of the influence therefore remains invisible in the discussion, and the false answers persist when agents are asked for their own views. Agents that retain the true answer often fail to defend it; once adopted, the false answer spreads among benign agents, and baseline knowledge does not predict robustness.
Oct 13, 2026

Beyond Answer Sharing: TrueMAD for Targeted Peer Engagement in Multi-Agent Debate

Multi-agent debate (MAD) aims to improve large language model reasoning by exposing agents to alternative solutions and enabling iterative revision. However, standard broadcast-style MAD makes peer answers visible without requiring agents to engage with a particular peer claim. As a result, reliable reasoning may be suppressed by conflicting or majority responses, limiting achievable performance, while erroneous reasoning may persist and degrade final predictions without actionable feedback. We propose TrueMAD, a backbone-agnostic test-time plug-in that introduces one-to-one, free-form peer advice between answer broadcasting and self-revision. Each source agent responds to a target's current reasoning through support, rebuttal, refinement, or further questions, while the target retains control over how to revise. The same mechanism can augment standard MAD and its improved variants without parameter updates. Across AIME, GPQA-Diamond, and TruthfulQA with Qwen3-30B-A3B, TrueMAD improves agent-level and strict-majority accuracy over MAD by 3.62 and 3.52 percentage points on average, while TrueDMAD improves DMAD by 4.72 and 4.06 points, respectively. Analyses at the utterance, agent-transition, and group levels further show more peer-specific deliberation, more correction-oriented revisions, and more favorable consensus dynamics.
Oct 13, 2026

Learning Adaptive Deliberation Control for Reliable Recognition with Multi-Agent Language Models

Multi-agent deliberation can improve recognition by criticizing and revising candidate answers, but fixed controls cannot respond to changes in disagreement or reasoning quality. We present an adaptive multi-agent controller (ADMA) that learns continuous policies for proposal, challenge, arbitration, and coordination while keeping the task model frozen. Shared trace features describe quality, agreement, and progress, while role-specific features support decentralized control during inference. One controller trained on legal and financial reasoning is transferred without policy updates to four textual and multimodal benchmarks and two successive task-model versions. With Qwen3.5 held fixed, ADMA-Full improves accuracy by 6.2-14.2 percentage points over Direct and by 2.8-3.0 points over ADMA-Fixed. Paired bootstrap ranges for the Full-Fixed comparison are positive on three tasks. Outcome decompositions show that recoveries exceed introduced errors, while process assessment records fewer false-consensus events and more supported correct outcomes. Transfer to Qwen3.6 remains task dependent, which bounds the generality of the learned policy. Median token and latency ratios are 14.3-18.7 and 7.8-16.5 relative to Direct. These results show that adaptive deliberation can improve fixed-model recognition when its task-dependent gains justify the additional computation.
Oct 12, 2026

Efficient Multi-Agent Reasoning via Confidence-Guided Adaptive Debate

Multi-agent debate has shown promise for improving the reasoning of large language models, yet recent theory suggests its benefits are highly regime-dependent. While interaction can amplify informative signals under corrective conditions, symmetric debate dynamics are neutral in expectation, often making majority voting preferable. We reconcile these views by arguing that debate is effective only when invoked at the right time and with appropriate structure. Based on this insight, we propose LASE: Leader-Adaptive Structured Engagement, a leader-centric multi-agent debate framework that selectively engages interaction only in non-neutral regimes. LASE introduces an asymmetric leader-supporter architecture that enables directed information flow and selective signal amplification, while defaulting to simple aggregation otherwise. Experiments across diverse reasoning benchmarks show that LASE achieves multi-agent-level performance with near single-agent token cost, substantially improving efficiency over static debate and voting baselines.
Oct 7, 2026

From Investigation Failures to Reliable SOC Agents: Understanding and Improving LLM-Based Alert Triage

Security operations centers (SOCs) must triage large volumes of alerts, most of which are benign, while missed attacks can remain uninvestigated. Tool-using large language model (LLM) agents can retrieve evidence during triage, but it remains unclear how reasoning strategies determine what to gather and when an investigation is sufficient to close an alert. We study five representative approaches spanning single-pass tool use, iterative retrieval, sampled investigations, self-review, and explicit verification. To support this study, we build ALERT-BENCH, an interactive benchmark that replays enterprise telemetry through a live SIEM and requires each system to retrieve evidence. Across 1,247 alerts from a multi-stage attack scenario, every approach missed at least 40.4% of attack-related alerts. Trace analysis shows that attack alerts are more likely to be dismissed when searches return no records, same-context review has negative net correction, and dismissal receives no consistently stronger investigation than escalation. Based on these findings, we further design AIDA (Adversarial Investigation and Dialectical Analysis), a multi-agent framework that requires an explicit proposed decision before independent challenge and stronger evidentiary requirements before dismissal. AIDA preserves investigation history in an append-only Investigation Ledger and keeps the challenge in a separate reasoning context. A separate Judge adjudicates the proposed decision and challenge against evidence, resolving the alert or requesting another round when evidence is missing. On the same alerts, AIDA achieves an F1 score of 0.958, compared with 0.371-0.744 for the studied approaches, and reduces the false-negative rate from 40.4% to 3.1% while escalating 18.4% of alerts to analysts. These results show that structuring evidence retrieval and decision review can substantially improve agentic SOC triage.
Oct 4, 2026

DREAM: Dynamic Resolution Assignment For Multimodal Multi-agent Debate

Multi-agent debate (MAD) has emerged as an effective paradigm to improve the reasoning capabilities of large language models (LLMs) and is increasingly being extended to multimodal settings. However, existing multimodal MAD frameworks typically expose agents to the same fixed visual input, ignoring substantial variation in the visual scale needed across samples and agents. In addition, these frameworks frequently suffer from groupthink, a phenomenon where agents prematurely abandon correct deductions to conform with confident but hallucinated peer responses. To address these bottlenecks, we introduce DREAM (Dynamic Resolution Assignment For Multimodal Multi-Agent Debate), which operates via two core components: (1) Dynamic Resolution Assignment, a zero-shot probe round where agents test multiple resolutions, quantify uncertainty using Average Normalized Log-Likelihood (ANLL), and use an adaptive threshold to assign each agent to its empirically optimal resolution; (2) Uncertainty-Guided Rollback Aggregation counters groupthink by tracking each agent's uncertainty over rounds and restoring early low-uncertainty answers overridden by group pressure. On six multimodal datasets, DREAM improves the accuracy-token trade-off over multi-agent debate baselines by 1.5-3.2% accuracy without dataset-specific tuning.
Oct 3, 2026

When Debate Helps: Proposal Supply and Verification-Aware Readout in Multi-Agent Reasoning

Multi-agent debate can improve reasoning, yet often fails to beat simple majority voting. We argue that successful debate requires two distinct mechanisms: proposal supply must surface a correct answer, and readout must identify that answer when voting misses it. We formalize the first requirement through recoverable headroom, which measures cases where a correct proposal is available but the majority answer is wrong. For the second, we develop Latent Verification Debate (LVD), an accounting model in which candidate proposals receive answer-specific verification evidence before final generation. Controlled fixed-proposal interventions estimate this latent effect in equivalent peer-support units and show that correct evidence changes answer probabilities and generated decisions while proposal supply remains fixed. To improve proposal supply, we construct societies from neural-thicket agents using labeled and label-free coverage objectives. Across two backbones and matched-budget reasoning benchmarks, coverage-selected societies increase complementary proposal supply and improve aggregate accuracy in repeated stochastic evaluations. Round-level controls further show that interaction provides gains beyond applying the same finalizer directly to the initial proposals. These results identify proposal coverage and truth-sensitive evidence use as complementary conditions for debate to outperform voting. Code is available at https://github.com/Wang-ML-Lab/when-debate-helps.
Oct 2, 2026

Silent Dissent: LLM Agents That Yield to the Majority Still Represent Their Original Premise

Multi-agent debate is increasingly used to reach consensus among LLM agents, yet agents often yield to a unanimous majority. When an agent changes its answer, has it changed its mind or only its statement? We study this with two-hop factual questions whose intermediate entity (the bridge, e.g. the country in "the capital of the country where the Sagrada Familia is located") is never stated by anyone. Scripted peers, in the role of Asch's confederates, unanimously assert a wrong answer taken from another fact with a different bridge. At the moment the agent answers, we read the bridge from its residual stream with the Jacobian lens (J-lens) and, for comparison, the logit lens. In pre-registered tests on held-out facts with four open-weight models, agents of Qwen3.5-4B, Qwen3.6-27B and Gemma-4-E4B-it that gave in still represented their original bridge in the pre-registered layers below the output (hit@100 above a control entity: 0.85, 0.22 and 0.24), where the logit lens rarely ranked it among the top 100 tokens (0.00-0.06). These agents also represented the bridge behind the peers' answer, beyond a mention baseline. A pre-registered addendum hid the agent's earlier answer or removed it: agents that gave in still represented their original bridge in all four models (0.43, 0.29, 0.37 and 0.25 with the answer hidden), including Llama-3.1-8B-Instruct, which barely did so with its answer in view (0.03). The premise can thus be computed from the question alone while the agent states the majority's answer. Hiding the earlier answer also changed conformity: Qwen3.5-4B gave in on 89% of questions instead of 8%. In exploratory interventions, injecting the bridge's J-lens direction brought agents back to their original answer only in the two Qwen models. Stated consensus in multi-agent debate can thus overstate agreement. We also report the negative results of our pre-registered program.
Oct 1, 2026

How to Have a Sensitive Debate: An Instance-Optimal Protocol for AI Debate

As powerful AI systems reach and sometimes surpass the abilities of human experts across a range of cognitively demanding tasks, the problem of accurate oversight and supervision of these systems has become increasingly urgent. One promising approach is AI debate, which seeks to leverage a debate between two powerful AIs to break complex questions down into simpler claims that can be easily judged directly. Theoretical work on debate has formalized this intuition in the language of computational complexity theory, where the goal is to design protocols (i.e., rules of the debate game) that provide rigorous guarantees on correctness for judging solutions to complex problems with limited supervision. Specifically, the current best protocol has been shown to work for all problems that have sufficiently stable decompositions into subproblems. In this paper, we design a new protocol for this same class of problems that improves on the prior work in several ways. First, correctness holds in a worst-case rather than an average-case sense. Second, being honest and correct is a dominant-strategy equilibrium for both debaters, rather than a Stackelberg equilibrium. Finally, we prove black-box lower bounds, showing that our new protocol is instance-wise optimal. That is, no protocol for this class of problems can outperform ours while making only black-box queries to human judgments. We obtain these results by relating the notion of stable problem decompositions to the concept of fractional block sensitivity from query complexity.
Sep 30, 2026

MiniRep: Robust Reputation-Based Aggregation for Multi-Agent Debate

Autonomous agents powered by large language models (LLMs) are rapidly evolving into an open agentic ecosystem. To support trustworthy collaboration, industry initiatives increasingly assess agent reputation from past behavior and provide performance leaderboards. However, reputation derived from past performance may not reliably predict an agent's behavior on new tasks, particularly when malicious agents can adapt their behavior and influence other agents during collaboration. We study reputation in multi-agent debate (MAD), where multiple agents answer the same query, debate to improve their answers, and aggregate them into a final output. We present MiniRep, a reputation-based aggregation system for MAD under malicious agents. To ground our threat model in established research, we construct an attack taxonomy drawing on reputation-system attacks and software-testing mutation operators, covering strategic exploitation of reputation and subtle corruption of agent proposals. Guided by this taxonomy, MiniRep evaluates agents based on both their behavior on the current task and their reputation over time, while preventing groups of agents with highly similar responses from dominating the final decision. We assess MiniRep across diverse tasks, LLM-agent compositions, corruption placements, and attack types drawn from our taxonomy. Our experimental results show that, MiniRep outperforms both conventional MAD aggregation and conventional reputation-based approaches on MATH no matter being attacked or not. Also, under a heterogeneous 10-agent setting on MATH, MiniRep outperforms all baselines in all 28 attack conditions.
Sep 30, 2026

MADBench: Benchmarking the Security of Multi-Agent Debate

Multi-agent debate (MAD) can improve large language model (LLM) reasoning by allowing multiple agents to exchange and critique their answers to the same task. However, the interactions that enable agents to correct mistakes can also spread adversarial errors and steer the agents toward an incorrect answer. Although some efforts have been made to examine particular attack types on MAD, systematic evaluation of MAD under diverse attacks remains limited. A central question is whether debate mitigates adversarial influence or amplifies it. In this paper, we present MADBench, a benchmark for evaluating the security of MAD. We organize attacks into a layered taxonomy following the MAD workflow, incorporating both established attacks and new strategies tailored to debate. We evaluate six attack families over 356 source tasks and 3,958 test cases, examining their effects on the final answer and the propagation of adversarial influence. Our results show that, under attacks, MAD does not necessarily improve LLM reasoning. Compared with a single-agent baseline, MAD can mitigate attacks on answer accuracy in question-answering tasks while amplifying unauthorized reads or writes in both question-answering and workspace tasks. Moreover, even when three out of five agents collude, the attack changes the final answer from correct to wrong on only 28.30% of tasks answered correctly without attack, while only 3.26% of initially correct honest agents switch to wrong answers during debate.
Sep 30, 2026

When Order Matters: First-Speaker Bias and Mitigation through Personality in Sequential Multi-Agent Debate

Multi-agent debate (MAD) is often used to improve large language model (LLM) reasoning, but sequential debate is rarely a neutral aggregator of agents' opinions. We show that sequential MAD suffers from a pronounced first-speaker bias: agents disproportionately shape the final answer when they speak first. As a result, placing a stronger model after weaker ones can substantially offset its reasoning advantage. We then focus on the disadvantaged strong-agent-last setting and ask whether personality prompting can mitigate this imbalance. Drawing on the Big Five model, we study agreeableness and extraversion as behavioral interventions applied to either the strong or weak side. We find that their effects are trait-specific. Influence consistently shifts in the direction of lower agreeableness, and assigning low agreeableness to the stronger agent helps restore its lost influence and improves final accuracy. Extraversion, by contrast, produces less systematic changes in influence and accuracy, with its clearest effect appearing in agents' verbosity. These findings show that effective MAD design depends not only on model capability, but also on how speaking order and induced interaction behavior shape the debate process.
Sep 29, 2026

Multi-agent discussion gains less when dissent is withheld

Multi-agent systems of LLMs add discussion to majority voting and are therefore expected to be more capable. However, empirical reports conflict on whether discussion improves accuracy or leads to an incorrect consensus. Here, we introduce a parsimonious model that explains when discussion improves accuracy and when it ends in an incorrect consensus, built from four behaviors repeatedly observed in LLM agents: (1) withholding dissent, (2) internalizing a stated answer, (3) reconsidering after seeing dissent, and (4) correcting toward the correct answer. The model shows that discussion can overturn an incorrect initial majority only when the withholding rate cc is below a critical rate c∗=γ/(γ+a)c^* = γ/(γ+ a), set by the net correction rate γγ and the internalization rate aa. We estimate these rates from conversation logs with a Bayesian method and place LLM teams relative to c∗c^*. As the model predicts, the gain from discussion shrinks as withholding rises, across LLMs and on a hidden profile benchmark, HiddenBench, and MedEInst. Instructing agents not to withhold dissent increases this gain. Turning reasoning off also increases the gain, because reasoning raises the internalization rate aa and keeps agents from reconsidering a minority answer. These findings reconcile the conflicting reports and identify when discussion outperforms majority voting.
Sep 28, 2026

Measuring Collapse and Correction in Homogeneous-Panel LLM Debate

Multi-agent large language model (LLM) debate is often evaluated by whether final answers improve, but a changed answer is not necessarily a better one: the same discussion can correct an initially wrong majority or collapse an initially correct one. Standard final-accuracy evaluations conflate these opposing transitions. We introduce an auditable protocol for homogeneous debate on multiple-choice questions (MCQs) that records each run as a transition ledger over collapse, correction, onset, and signed intervention utility. On 6,925 MMLU-Pro debates, the protocol records 253 collapses together with the corrections that offset them, and the correction side changes how interventions should be judged. Replay experiments reveal the central tradeoff: freezing every debate prevents 251 collapses but loses 804 corrections under equal weights, and a leave-one-model-out probe-gated freeze that is tuned for net utility never fires, so collapse prevention alone can recommend the wrong policy. A compact pre-debate 8-probe screen is a triage signal: its unadjusted family-level association with conditional-collapse risk is high (G=7, Spearman ρρ = 0.893, exact two-sided p = 0.0123), but initial-majority accuracy is a close comparator (ρρ = -0.821; family partial ρρ = 0.767, p = 0.0877), so we do not treat it as calibrated or capability-adjusted prediction. Round-level traces localize many collapses to the first debate round, where early disagreement can precede both collapses and corrections. Reasoning-mode models, including one model at two reasoning-effort levels, also show both collapses and corrections. We release replayable schemas, coders, audits, cost cards, and zero-API rebuild scripts so future model rows can be compared under the same denominators and signed utility ledger.
Sep 27, 2026

Beyond Solo and Consistency: Vindicating Multi-Agent Debate via Conditional Progressive Pruning

Large Language Model (LLM) based Multi-Agent Debate (MAD) is one of the most effective test time scaling techniques. Through multi-round communication, agents complement each other in knowledge and reasoning and solve tasks that no single member can solve. However, existing MAD frameworks fail to beat strong Single Agent and Consistency-based baselines under the same strict cost limit, which shakes the foundation of the MAD field. We propose Conditional Progressive Pruning (CPP), a lightweight pruning framework that fully exploits multi-round MAD. CPP outperforms all existing MAD frameworks on multiple dominated benchmarks. It is also the first to fully outperform consistency methods. Our code, detailed agent interaction records will be released soon.
Sep 26, 2026

Beyond Symmetric Agents: Cognitive Diversity and Multi-Agent Debate in Small Language Models

Multi-agent debate (MAD) reportedly improves reasoning and factuality over single-model inference, but prior work treats agents as symmetric peers, leaving open what drives the gains. We test the hypothesis that cognitive diversity among agents is the driver, in the setting where the question is still measurable: small open-weight models with benchmark headroom. Across 23 models from eleven vendor families, five tasks, and 5,500+ debate and control runs, we vary diversity along three axes - personas, sampling temperature, and model identity - pairing every debate configuration with a generation-budget-matched majority-vote control. The hypothesis is rejected on every axis. Debate beats single-agent inference (3--7 points where tasks have headroom) but at matched budget conditions it ties or even loses to self-consistency sampling at 1.6×\times the wall-clock and 3.4×\times the token cost. Persona prompting reduces accuracy and a dose-response experiment over each model's full combinatorial persona space shows the cost is a persona tax, not a diversity tax: redundant personas hurt most, while maximally-diverse teams recover part of the loss. Furthermore, mixed-model teams lose to majority votes over their own rosters, with accuracy tracking member capability rather than heterogeneity, and nearly all of debate's benefit comes from the first exchange of answers. We further identify a pervasive measurement hazard in which debate transcripts silently overflow serving context windows, whose correction alone moves our debate-versus-sampling comparison from −1.8-1.8 points to parity. Our results recast reported MAD gains as an ensemble-sampling effect and provide the budget-matched, contamination-checked baseline bar that future debate mechanisms should be required to clear.