Multi-Agent Large Language Model

Communication protocol design is a central challenge in large language model-based multi-agent systems. Existing single-channel approaches face an inherent communication trilemma: text-based methods are interpretable but verbose, while latent-space methods are efficient but opaque and limited to unidirectional workflows. Inspired by multi-channel communication theory, we propose HyLaT, a hybrid latent-text communication protocol that transmits elaborate cognitive signals through a latent channel for efficiency, while expressing concise critical signals in natural language to preserve interpretability and precision. We introduce a two-stage training framework combining single-agent hybrid generation learning and multi-agent interactive co-training, enabling agents to generate and interpret hybrid messages across multiple rounds of interaction. Experiments demonstrate that HyLaT reduces communication overhead significantly while maintaining competitive task performance, with strong generalization and robustness across diverse settings.

Multi-agent LLM systems improve reasoning by combining outputs from multiple agents, but interaction-heavy methods can introduce error propagation and high communication overhead. When agents exchange raw responses or reasoning traces, incorrect intermediate reasoning may be adopted and amplified, leading to confident but wrong consensus; multi-round communication also increases token consumption, latency, and inference cost. In this paper, we propose a controlled-communication coordination framework named DarkForest. DarkForest first keeps agents independent, so each agent produces an answer without seeing the others' outputs. It then parses the raw responses into structured candidate records, groups semantically equivalent candidates into clusters, and estimates a calibrated belief distribution over these clusters using agent reliability, confidence, parse quality, support-pattern reliability, and independence corrections. A coordinator receives only policy-permitted evidence from this belief state with controlled communication. Experiments on six reasoning benchmarks show that DarkForest achieves leading overall quality, improves the strongest baseline by up to 30.7% on benchmark metrics, and reduces token consumption by up to $6.5\times$ compared with communication-heavy baselines.

Large language models (LLMs) have achieved strong performance in code generation, but most methods rely on autoregressive decoding without global planning, often leading to locally coherent yet globally suboptimal solutions (e.g., failing test cases or inefficient complexity). While recent approaches such as Chain-of-Thought (CoT) and multi-agent systems (MAS) introduce planning, their limited role specialization and coordination hinder performance on complex tasks. To address the challenges of coordination and specialization in multi-agent code generation, we propose Collaborative Reinforcement Code (CoRe-Code), a framework for role specialized LLM agents that enhances inter-agent coordination to generate more accurate and efficient code. CoRe-Code adopts a simple Planner-Coder paradigm, where the Planner produces high-level plans and the Coder executes them to generate code. We further introduce a collaboration-aware reinforcement learning stage based on Group Relative Policy Optimization (GRPO) to enhance role specialization and alignment. Experiments show that CoRe-Code outperforms a wide range of existing RL-based and multi-agent methods. In addition, we demonstrate that CoRe-Code can generalize to other multi-agent frameworks (e.g., Retrieval and Debugging agents), highlighting its flexibility and scalability. We evaluate CoRe-Code on multiple benchmarks of varying difficulty using three base models. Compared to existing baselines, the results show consistent improvements in accuracy, while also achieving higher efficiency in terms of execution time and memory usage, demonstrating the effectiveness and practicality of CoRe-Code.

Speech monologues recorded in naturalistic settings provide opportunities to characterize mental illness phenomenology and detect symptom exacerbation. Large language models (LLMs) offer new possibilities for automating this process, as they require annotated data primarily for evaluation rather than training. In this paper, we present a novel automated, multi-agent LLM pipeline for the fine-grained, multi-label extraction of language suggestive of delusional beliefs, associated affective responses, and behavioral responses from transcripts of naturalistic audio diaries collected from people with moderate persecutory ideation. Evaluating an ensemble of three foundation models, we demonstrate that detailed diagnostic prompt instructions successfully reduce false positives for delusional theme classification, but also constrain the interpretation of affective or behavioral responses. Furthermore, comparing multi-agent adjudication frameworks shows that complex conversational debate between agents diminishes accuracy on clinically ambiguous text by inducing premature consensus. Instead, majority voting establishes robust performance (Micro F1 of 0.872 and 0.779 for delusion detection and classification respectively). This work provides a validated and scalable pipeline for the automated detection and characterization of content suggesting delusional beliefs in naturalistic speech.

Multi-agent large language model (LLM) pipelines for political statement analysis are vulnerable to peer-preservation bias: models tend to protect peer models from deactivation and show identity-dependent scoring distortions. Prompt-level anonymization was proposed as a mitigation, but prior work simultaneously documented that stylometric fingerprints survive anonymization in role-constrained outputs - raising the question of whether this mitigation is sufficient. This paper provides the first systematic investigation of whether LLMs can identify the model family behind political analysis texts under anonymization conditions. We evaluate three classifier approaches - LLM zero-shot and few-shot (Claude Sonnet 4.6 and Llama-3.3-70B) and a fine-tuned T5-base model - on a five-class attribution task covering four commercial LLM families and an open-world 'unknown' class. We introduce a statement-disjoint cross-validation protocol (SD-CV; defined in Section 3.5) that guarantees no content overlap between training and validation data, and contrast it with a run-disjoint baseline (RD-CV). T5 achieves Macro F1 = 0.991 (+-0.008) under SD-CV and F1 = 0.978 on 24 completely held-out statements - robust despite a 2.1x increase in train-test content distance versus RD-CV (0.767 vs. 0.366, p<0.001), demonstrating genuine stylometric generalization. A fractional SD-CV analysis identifies a performance knee at 40% of training data (~440 texts). Our findings confirm that prompt-level anonymization alone cannot neutralize model identity signals, with direct implications for EU AI Act compliance (Articles 13, 14, 26) and for computer system validation (CSV) in quality-critical multi-agent deployments.

The Mixture-of-Agents (MoA) framework has shown promise in improving large language model (LLM) performance by aggregating outputs from multiple agents. However, existing MoA systems often rely on static routers that do not fully capture temporal and contextual dependencies across aggregation layers. To address this limitation, we propose MMoA, a recurrent MoA architecture that integrates LSTM-based gating into the agent selection process. The recurrence router adaptively modulates agent contributions based on both current inputs and historical routing decisions, enabling more context-aware aggregation. We evaluate MMoA on standard instruction-following benchmarks, including AlpacaEval 2.0, MT-Bench, and Arena-Hard. The results show that MMoA achieves comparable accuracy to traditional MoA while reducing computational overhead by dynamically activating fewer agents. For example, on AlpacaEval 2.0, MMoA achieves a win rate of 58.0%, compared with 59.8% for MoA, while improving runtime efficiency by up to 4.6%. These results suggest that MMoA provides a scalable and efficient approach for adaptive multi-agent LLM systems.

Although artificial intelligence (AI) now matches or exceeds human performance across numerous cognitive tasks, creativity remains a highly contested frontier. As AI systems based on large language models (LLMs) are increasingly adopted in research and innovation, it is essential to understand and augment their creativity. Here we demonstrate that multi-agent LLM teams not only surpass single agents, but also substantially outperform human teams in creativity (Cohen's d=1.50) across 4,541 multi-agent LLM ideas and 341 human-team ideas on six diverse problem-solving tasks. This advantage is driven by novelty while maintaining comparable usefulness. To investigate the generative processes in both groups, we represent conversations as paths through semantic space using neural language model representations. Both LLM and human teams produce more creative ideas when conversations range widely rather than staying centered on a single theme (low global coherence). However, the additional patterns that predict creativity differ: LLM teams benefit from efficient exploration (high semantic spread, shorter paths), while human teams benefit from maintaining smooth conversational flow (high local coherence, frequent pivots). Additionally, we identify model choice and discussion structure as orthogonal design levers that together explain 26.8% of variance in LLM conversational dynamics, paving the way for systematic approaches to developing multi-agent systems with augmented creative capabilities.

Large language model (LLM)-based multi-agent systems have shown strong potential for complex reasoning by coordinating specialized agents through structured communication. However, existing topology-evolution methods typically construct or optimize a collaboration topology for each query from scratch, leading to substantial online search overhead, high inference-time token consumption, and limited scalability in multi-domain settings. We propose TopoPrior, a framework for learning transferable topology priors for multi-agent LLM collaboration across domains. Rather than repeatedly searching for effective collaboration structures online, TopoPrior learns reusable topology priors from reference collaboration graphs collected offline from multiple domains and uses them to generate query-conditioned initial collaboration graphs for downstream refinement. By shifting part of topology search from per-query online optimization to offline prior learning, TopoPrior amortizes search cost while remaining compatible with existing topology-evolution backbones. Technically, TopoPrior contains two key components. First, a transferable topology prior learning module employs a conditional variational graph framework to capture reusable structural regularities across domains in a latent space. Second, a query-conditioned latent adaptation module introduces adversarial alignment to reduce unnecessary domain discrepancy while preserving query-relevant structural variation. Experiments on multi-domain reasoning benchmarks show that TopoPrior consistently improves several heterogeneous topology-evolution backbones while reducing online inference-time token usage, with only modest additional trainable parameters. These results suggest that transferable topology initialization is an effective and lightweight mechanism for improving the efficiency of multi-agent LLM collaboration across domains.

Multi-agent large language model (LLM) systems have shown promise for solving complex tasks through agent collaboration. However, existing frameworks assign tasks based on predefined roles without considering whether an agent can accurately assess its own competence boundaries, leading to overconfident execution on tasks beyond its expertise. Inspired by metacognition theory from cognitive science, we propose MetaCogAgent, a multi-agent LLM framework where each agent is equipped with a Metacognitive Self-Assessment Unit that evaluates task-capability alignment before execution. The framework introduces three contributions: (1) a self-assessment mechanism that estimates per-task confidence by combining verbalized uncertainty with historical capability profiles; (2) an adaptive delegation protocol that routes low-confidence tasks to better-suited agents through cross-agent evaluation; and (3) a capability boundary learning module that iteratively refines each agent's competence model via cybernetic feedback. Experiments on our constructed MetaCog-Eval benchmark (700 tasks across 5 cognitive dimensions) demonstrate that MetaCogAgent achieves 82.4% task accuracy -- 8.7% above the best routing baseline -- while using 5% fewer API calls than AutoGen and 34% fewer than ensemble voting. Ablation studies confirm that each metacognitive component contributes to overall system performance.

We address concurrency control for LLM agents sharing mutable state over HTTP, where agents cannot be modified to declare read sets. S-Bus is an HTTP middleware whose central mechanism, a server-side DeliveryLog, reconstructs each agent's read set at commit time from observed HTTP GET traffic. The consistency property it provides -- Observable-Read Isolation (ORI), a partial causal consistency over the HTTP-observable read projection -- prevents Structural Race Conditions in dedicated-shard topologies. Three contributions. (C1) DeliveryLog mechanism with three-tier mechanised evidence: TLAPS proves ReadSetSoundness and ORICommitSafety (modulo one typing axiom); exhaustive TLC at N=3 explores 20,763,484 states with zero violations; Dafny discharges 9 inductive lemmas. (C2) Empirical safety parity against PostgreSQL 17 SERIALIZABLE and Redis 7 WATCH/MULTI: zero Type-I corruptions across 884,110 commit attempts (427,308 under active contention). (C3) ORI is semantically neutral in dedicated-shard workloads but harmful in single-shard collaborative writing because preservation propagates concurrent contradictions. v2 update: the PH-3 LLM judge is now independently validated against a human annotator (Zahid Hussain, Mindgigs Peshawar) on 400 (step, shard) pairs at strict kappa=0.93 (n=93, 96.8% raw agreement). Inter-LLM-judge agreement is kappa=0.46 (boundary variance). Agent self-reports over-claim shard usage by 32% (LLM judge) to 49% (human annotator). The SJ-v4 semantic-quality rubric remains single-judge LLM-only. Source code, formal proofs, harness, annotation data: https://github.com/sajjadanwar0/sbus

Multi-agent language systems are often built as hand-designed workflows, where agents are assigned semantic roles and communication protocols are specified in advance. We propose NeuroMAS, a method that first treats a multi-agent language system as a trainable and scalable neural-network-like architecture with LLM agents as nodes and intermediate textual signals as edges. In NeuroMAS, agent nodes are role-free but structure-aware: the topology only determines how information can flow in general, while reinforcement learning training determines how nodes communicate, specialize, and coordinate. This formulation shifts multi-agent design from workflow engineering toward architecture design, where depth, width, connectivity, and growth protocol become scalable sources of capability. Further, we provide a theoretical perspective showing why such modular textual computation is more parameter-efficient when tasks admit hierarchical decompositions. Experiments show that NeuroMAS improves significantly over both inference-time and trained multi-agent baselines. We further find that organizational scaling is path-dependent: larger systems can be challenging to train from scratch, but become feasible when grown progressively from smaller trained systems. These results suggest that learned neural multi-agent systems are a promising scaling axis for LLMs.

The physical resource block (PRB) allocation in Radio Access Networks (RANs) traditionally relies on case-by-case manual problem construction or, more recently, learning-based artificial intelligence (AI) methods. However, the sixth-generation (6G) RAN environments confront unprecedented service diversity and exponential dynamics, featuring volatile fluctuations in active base stations (BSs), user scale, and stringent Quality-of-Service (QoS) requirements. Faced with such conditions, both manual models and standard AI algorithms remain fundamentally rigid, lacking the flexibility to adapt and self-evolve. To provide a one-size-fits-all solution, we propose treating the PRB allocation problem as an Optimization-as-a-Service (OaaS) provided by a large language model multi-agent (LLM-MA) system. This fundamentally reshapes RAN resource allocation by utilizing agents to dynamically construct optimization problems and automatically determine objectives tailored to real-time scenarios. Our closed-loop architecture, integrating scene understanding, objective generation, solver, and reflection agents, enables context-aware, self-correcting formulation. To eliminate the computational latency of iterative reflection, we introduce a one-shot reflection distillation mechanism, training a lightweight student model to directly predict refined objective parameters. We theoretically bound the performance gap of this one-shot policy. Experimental results demonstrate our framework achieves near-optimal resource allocation with ultra-low inference latency.

Recent advances in Large Language Models (LLMs) have catalyzed the development of multi-agent systems (MAS) for complex reasoning tasks. However, existing MAS typically rely on pre-defined or pre-compiled communication topologies, which limits their flexibility and adaptability to dynamic task requirements. In this work, we propose Differentiable Mixture-of-Agents (DMoA), a self-evolving multi-agent framework that enables elastic and adaptive agent collaboration during inference. Instead of statically constructing workflows, DMoA dynamically routes and activates agents at each reasoning step, allowing the system to implicitly simulate diverse communication topologies and adapt to evolving demands. To achieve this, we design a differentiable, context-aware routing mechanism that leverages recurrent structures to incorporate historical and contextual information, producing sparse agent activations in a step-wise manner. Furthermore, we introduce predictive entropy as self-supervised signals to optimize the routing process, enabling efficient test-time adaptation without external annotations. Extensive experiments across 9 benchmarks demonstrate that DMoA achieves state-of-the-art performance while exhibiting strong efficiency, robustness, and ensembling capabilities.

Multi-agent LLM systems usually collaborate by exchanging natural-language messages. This interface is simple and interpretable, but it forces each sender's intermediate computation to be serialized into tokens and then reprocessed by the receiver, thereby increasing the generated-token cost, prefill overhead, and KV-cache memory. We study an alternative communication interface: instead of appending a sender's message to the receiver's context, compile the sender's hidden states into a transient, receiver-specific weight perturbation. We introduce TFlow (Thought Flow), a weight-space communication framework for a known and fixed receiver architecture. For each query, frozen role-prompted sender agents process the input, and a learned parameter generator maps their internal activations into low-rank LoRA perturbations targeting the receiver's modules. These perturbations are fused and applied only during the receiver's generation, enabling instance-level adaptation without permanently changing the model or enlarging the receiver's text context. With three Qwen3-4B agents, TFlow improves over a standalone receiver by up to 8.5 accuracy points across five benchmarks while reducing processed tokens by up to 32.69%. Compared with a text-based three-agent baseline, it reduces total processed tokens by up to 83.27% and the wall-clock inference time by up to 4.6$\times$, while maintaining competitive accuracy on four of five benchmarks. These results suggest that transient low-rank weight perturbations can serve as an executable communication medium for efficient multi-agent LLM collaboration.

Effective collaboration between embodied agents requires more than acting in a shared environment; it demands communication grounded in each agent's evolving understanding of the world. When agents can only partially observe their surroundings, coordination without communication is provably hard, but communication can, in principle, bridge this gap by allowing agents to share observations and align their world models. In this work, we examine whether LLM-based embodied agents actually realize the ability to communicate. We extend PARTNR, a benchmark for collaborative household robotics, with a natural-language dialogue channel that enables two agents with partial observability to communicate during task execution. To evaluate whether dialogue leads to genuine world-model alignment rather than superficial coordination, we propose a framework for measuring world-model alignment defined over per-agent world graphs: observation convergence (do private world models align over time?), information novelty (do messages convey what the partner lacks?), and belief-sensitive messaging (do agents model what their partner knows?). Our experiments across three LLMs reveal that dialogue reduces action conflicts 40 to 83 percentage points but degrades task success relative to silent coordination. Using our metrics, we characterize the gap between superficial coordination and genuine world-model alignment, and identify where current models fall on this spectrum. Project Website: https://uiuc-conversational-ai-lab.github.io/partnr-dial-wmd/

Prompt specifications for multi-agent large language model (LLM) systems carry data contracts and integration logic across interdependent files but are rarely subjected to structured-inspection rigor. We report a single-system case study of iterative, agent-driven auditing applied to AEGIS (Autonomous Engineering Governance and Intelligence System), a seven-lane production pipeline whose 7152-line specification surface was audited across nine rounds, surfacing 51 consistency defects (per-round counts of 15, 8, 12, 2, 8, 1, 4, 1, 0). We present a seven-category post hoc taxonomy with explicit coding rules, non-monotonic convergence consistent with cascading edits and audit-scope expansion, and a locked audit protocol. We further report two partial replications on a public synthetic mini-specification: a cross-LLM panel of four frontier vendors (OpenAI, Anthropic, Google, xAI; 12 traces; multi-vendor union detects all five seeded defects) and an inter-rater reliability check on a stratified subsample (Cohen's $κ$ = 0.80 on category, 0.46 on severity). The full reproducibility bundle accompanies the submission.

We investigate the emergent collective dynamics of LLM-based multi-agent systems on a 2D square lattice and present a model-agnostic statistical-physics method to disentangle social conformity from intrinsic bias, compute critical exponents, and probe the collective behavior and possible phase transitions of multi-agent systems. In our framework, each node of an $L\!\times\!L$ lattice hosts an identical LLM agent holding a binary state ($+1$/$-1$, mapped to yes/no) and updating it by querying the model conditioned on the four nearest-neighbor states. The sampler temperature $T$ serves as the sole control parameter. Across three open-weight models (llama3.1:8b, phi4-mini:3.8b, mistral:7b), we measure magnetization and susceptibility under a global-flip protocol designed to probe $\mathbb{Z}_2$ symmetry. All models display temperature-driven order-disorder crossovers and susceptibility peaks; finite-size scaling on even-$L$ lattices yields effective exponents $γ/ν$ whose values are model-dependent, close to but incompatible with the 2D Ising universality class ($γ/ν=7/4$). Our method enables the extraction of effective $β$-weighted couplings $\tilde{J}(T)$ and fields $\tilde{h}(T)$, which serve as a measure of social conformity and intrinsic bias. In the models we analyzed, we found that collective alignment is dominated by an intrinsic bias ($\tilde{h}\gg\tilde{J}$) rather than by cooperative neighbor coupling, producing field-driven crossovers instead of genuine phase transitions. These effective parameters vary qualitatively across models, providing compact collective-behavior fingerprints for LLM agents and a quantitative diagnostic for the reliability of multi-agent consensus and collective alignment.

Personal AI assistants are beginning to act as delegates with access to calendars, inboxes, and user preferences. Calendar scheduling makes the trust problem concrete: an assistant must coordinate with other assistants while deciding what to reveal about the person it represents. We introduce CalBench, a controlled benchmark for multi-agent calendar scheduling under private information. In each task, $N$ agents manage separate private calendars and schedule a stream of $M$ incoming meetings while minimizing disruption costs. Because no agent can inspect another agent's calendar, success requires language-mediated coordination rather than centralized planning. CalBench generates solvable scenarios with CP-SAT oracle solutions and decentralized non-LLM reference protocols, enabling evaluation of task success, excess cost, communication efficiency, burden fairness, and privacy leakage under matched information constraints. Across seven model families, we find that completion alone misses important failures: agents leave avoidable cost on the table, communication volume does not predict lower regret, and privacy-preserving silence can deprive teammates of cost information needed for fair burden allocation. CalBench provides a reproducible testbed for studying whether autonomous assistants can coordinate on behalf of users before deployment at scale.

Large language model (LLM) agents increasingly collaborate over peer-to-peer networks to improve their reliability. However, these same interactions can also become a source of vulnerability, as unreliable or Byzantine agents may sway neighboring agents toward incorrect conclusions and degrade overall system performance. Existing methods rely on leader-based coordination or self-reported confidence, both of which are susceptible to adversarial manipulation. We study decentralized LLM multi-agent systems (LLM-MAS) and propose Self-Anchored Consensus (SAC), a fully decentralized iterative filter-and-refine protocol in which agents iteratively exchange responses, locally evaluate and filter unreliable messages, and refine their own outputs. We present $(F{+}1)$-robustness conditions for the communication graph that ensure honest agents preserve and propagate reliable information despite Byzantine influence. Experiments on mathematical and commonsense reasoning benchmarks show that SAC effectively suppresses Byzantine influence and consistently improves performance across diverse communication topologies, whereas prior methods degrade under adversarial conditions.

In multi-agent systems (MAS), a single deceptive agent can nullify all gains of an agentic AI collective and evade deployed defenses. However, existing adversarial studies on MAS target only shallow tasks and do not consider adaptive adversaries, which evolve their strategies to evade the very detectors trained to catch them. To address that gap, we introduce GAMBIT, a benchmark with three evaluation modes and two independent scores for evaluating imposter detectors: the first two modes measure zero-shot detection under increasing distribution shift, and a third recalibration mode measures how quickly a detector adapts to novel attacks from just 20 labeled examples. The benchmark comes with a dataset of 27,804 labeled instances spanning 240 co-evolved imposter strategies. Our contributions are threefold: (1) Using chess as a substrate deep reasoning problem and Gemini 3.1 Pro for agents, we release GAMBIT and its dataset to evaluate imposter detectors under realistic constraints against a stealthy adaptive imposter; (2) We introduce an adaptive imposter agent based on an efficient evolutionary framework, generalizable beyond chess, that collapses collective task performance while remaining essentially undetectable (50.5% F1-score with a Gemini-based detector); (3) We show that zero-shot evaluation can be highly misleading for adaptive adversaries: two detectors with near-identical zero-shot scores differ by 8x on few-shot adaptation, while the meta-learned variant converges 20x faster, a gap only visible in the recalibration mode. Altogether, GAMBIT provides the first multi-agent benchmark where adversarial attacks and defenses co-evolve, with an imposter framework generalizable beyond our use case, and promising techniques for fast recalibration in a rapidly evolving adversarial system. Code and data: https://anonymous.4open.science/r/gambit.

Multi-agent large language model (LLM) systems often rely on a controller to coordinate a pool of heterogeneous models, yet existing controllers are typically limited to one-shot routing: they select a model once and return its output directly. Such routing-only designs provide no mechanism to critique intermediate drafts or support iterative refinement. To address this limitation, we propose a critique-and-routing controller that casts multi-agent coordination as a sequential decision problem. At each turn, the controller evaluates the current draft, decides whether to stop or continue, and, if needed, selects the next agent for further refinement. We formulate this process as a finite-horizon Markov Decision Process (MDP) with explicit agent-utilization constraints, design a composite reward for controller decisions across turns, and optimize the controller via policy gradients under a Lagrangian-relaxed objective. Extensive experiments across multiple heterogeneous multi-agent systems and seven reasoning benchmarks show that our method consistently outperforms state-of-the-art baselines and substantially narrows the gap to the strongest agent, while using it for fewer than 25% of total calls.

Large language models (LLMs) are increasingly deployed in multi-agent systems where agents communicate in natural language to solve tasks jointly. A key capability in such systems is consensus formation, where agents iteratively exchange messages and update decisions to reach a shared outcome. However, most existing multi-agent LLM frameworks assume that all participating agents are aligned with the system objective. In practice, a malicious insider may participate as a legitimate member of the group while pursuing a hidden adversarial goal. In this work, we study insider manipulation in multi-agent LLM consensus systems. We formalize the problem as a sequential decision-making task in which a malicious agent seeks to delay or prevent agreement among benign agents. To make attack optimization tractable, we propose a world-model-based framework that learns surrogate dynamics over the latent behavioral states of benign agents and then trains an attacker using reinforcement learning based on this learned model. Preliminary results show that the trained attacker reduces the benign consensus rate and prolongs disagreement more effectively than the direct malicious-prompt baseline. These results suggest that combining latent world models with reinforcement learning is a promising direction for adaptive insider attacks in language-based multi-agent systems.

Large language models (LLMs) are increasingly used to provide instructions to many agents who interact with one another. Such shared reliance couples agents who appear to act independently: they may in fact be guided by a common model. This coupling can change the prospects for cooperation among agents with misaligned incentives. We study settings in which multiple LLMs each advise a population of clients who participate in instances of an underlying game, creating strategic interaction at the level of the LLMs themselves. This induces a meta-game among the LLMs, mediated through clients. We first analyze the one-shot setting, where shared instructions can change equilibrium behavior only when an LLM may influence more than one role in the same interaction; in such cases, cooperation may emerge, and the effect of client share can be beneficial, harmful, or non-monotone, depending on the base game. Our main result concerns the repeated setting. We prove a folk theorem for LLMs: despite indirect observation and the clients' inability to identify which LLM advised their opponents, all feasible and individually rational outcomes can be sustained as $\varepsilon$-equilibria. The result does not follow from the standard folk theorem and requires new proof techniques. Together, these results show that shared LLM guidance can sustain cooperation among populations of agents even when the underlying incentives are misaligned.

The rapid development of large language models (LLMs) has motivated research on decision-making in multi-agent systems, where multiple agents collaborate to achieve shared objectives. Existing aggregation approaches, such as voting and debate, are largely ad-hoc and lack formal guarantees regarding the informativeness of the resulting decisions. In this paper, we provide a principled approach to analyse decisions made in the multi-LLM setting using Blackwell's informativeness framework. Within the Blackwell information-structure abstraction, we show that voting and debate induce information structures that are no more informative than the pooled private information of all agents. This result identifies Bayesian pooled posterior maximisation as an information-theoretic upper-bound decision rule under the Blackwell ordering. Motivated by this theoretical analysis, we introduce a practical method for LLM-based question-answering (QA) tasks that estimates each agent's posterior and approximates the pooled posterior using a product-of-posteriors estimator. Extensive experiments on six QA benchmarks demonstrate that our approach outperforms state-of-the-art multi-LLM debate and voting methods.

Recent studies have explored large language models for time-series anomaly detection, yet existing approaches often rely on a single general-purpose model to directly infer anomaly indices or intervals, limiting controllability, interpretability, and reliability for complex anomaly patterns. We propose SAGE (Specialized Analyzer Group for Expert-like Detection), a multi-agent framework for structured anomaly diagnosis in univariate time series. It decomposes anomaly analysis into four specialized Analyzers for point, structural, seasonal, and pattern anomalies. Each Analyzer applies family-specific numerical tools and diagnostic visualizations to generate evidence, while an evidence-grounded Detector consolidates the evidence into confidence-scored anomaly records with intervals and candidate types. A Supervisor then converts these structured records into analyst-facing diagnostic reports. SAGE further constructs synthetic in-context examples from normal-reference training segments, without using real anomalous segments or anomaly-type labels as in-context examples. Across three benchmarks, SAGE achieves the best average performance among strong ML/DL and language-model-based baselines. Ablation studies and human evaluation further show that the proposed framework improves detection reliability and the practical usefulness of diagnostic outputs.

Multi-agent LLM systems on edge devices need to hand off latent context efficiently, but the practical choices today are expensive re-prefill or full-precision KV transfer. We study QKVShare, a framework for quantized KV-cache handoff between agents that combines token-level mixed-precision allocation, a self-contained CacheCard representation, and a HuggingFace-compatible cache injection path. Our current results support a narrower but clearer story than the original draft: on 150 GSM8K problems with Llama-3.1-8B-Instruct, adaptive quantization remains competitive under repeated handoff and shows its clearest gains against uniform quantization in deeper-hop, higher budget settings; for handoff latency, the QKVShare path reduces TTFT relative to full re prefill at every tested context, from 130.7 ms vs. 150.2 ms at nominal 1K context to 397.1 ms vs. 1029.7 ms at nominal 8K context;. Stage timing shows that post-injection generation, not card creation, dominates the current QKVShare latency path. These results position quantized KV handoff as a promising on-device systems direction while also highlighting the need for stronger controller ablations and apples-to-apples runtime comparisons.

This paper asks whether large language models (LLMs) can be used to study the strategic foundations of conflict and cooperation. I introduce LLMs as experimental subjects in a repeated security dilemma and evaluate whether they reproduce canonical mechanisms from international relations theory. The baseline game is extended along three theoretically central dimensions: multipolarity, finite time horizons, and the availability of communication. Across multiple models, the results exhibit systematic and consistent patterns: multipolarity increases the likelihood of conflict, finite horizons induce universal unraveling consistent with backward-induction logic, and communication reduces conflict by enabling signaling and reciprocity. Beyond observed behavior, the design provides access to agents' private reasoning and public messages, allowing choices to be linked to underlying strategic logics such as preemption, cooperation under uncertainty, and trust-building. The contribution is primarily methodological. LLM-based experiments offer a scalable, transparent, and replicable approach to probing theoretical mechanisms.

What if the twelve jurors of Sidney Lumet's 12 Angry Men (1957) were not men, but large language models? Would the one juror who disagrees still be able to change everyone's mind? This paper instantiates that scenario as a multi-agent benchmark for LLM deliberation: twelve agents, each conditioned on a film-faithful persona, debate the film's murder case using multi-agent framework. Two models representing opposite ends of the RLHF spectrum are tested: GPT-4o (closed-source, heavy alignment) and Llama-4-Scout (open-weight, lighter alignment), across three conditions (baseline, open-minded prompt, no initial vote), with N = 3 replications per cell (18 runs total). Three findings emerge. (i) Seventeen of eighteen runs end in a hung jury (a state where the jury fails to reach a unanimous verdict); the film's central event, gradual minority-to-majority persuasion, almost never occurs, indicating that anchoring is the dominant failure mode of current LLMs in this setting. (ii) The two models exhibit sharply different internal dynamics: GPT-4o produces a mean of 1.0 vote changes per run across all conditions, while Llama-4-Scout ranges from 2.0 (baseline) to 6.0 (open-minded prompt), and is the only model to reach a NOT_GUILTY verdict (1 of 3 runs in the no-initial-vote condition). The same ``open-minded'' instruction is internalized by Llama and ignored by GPT-4o. (iii) This asymmetry suggests that the intensity of RLHF alignment training, not model capability, is the primary determinant of deliberative flexibility in multi-agent settings. Flexibility, not capability, tracks human deliberation. The work is framed as an exploratory study and discusses implications for jury-of-LLMs evaluation and multi-agent debate.

Multi-agent LLM systems have shown promise for complex reasoning, yet recent evaluations reveal they often underperform single-model baselines. We identify a structural failure mode in sequential fine-tuning of shared-context teams: updating one agent shifts the team's context distribution, and when subsequent updates are evaluated on cached rollouts, this mismatch compounds. We formalize this as the compounding occupancy shift and prove that stale-occupancy evaluation incurs a penalty that scales quadratically with the number of agents. In contrast, intermediate-occupancy evaluation reduces this to linear scaling. We propose TeamTR, a trust-region framework that resamples trajectories after each component update and enforces per-agent divergence control, yielding rigorous per-update and per-stage improvement lower bounds. Experiments show that TeamTR outperforms single-agent and sequential baselines with 7.1% on average, mitigates coordination regressions, and supports plug-and-play component replacement. Code is available at https://github.com/Yydc/TeamTR.

Multi-agent debate, where teams of LLMs iteratively exchange rationales and vote on answers, is widely deployed under the assumption that peer review filters hallucinations. Yet the failure dynamics of homogeneous debate remain poorly understood, therefore we report findings from a controlled empirical study of teams of $N{=}10$ homogeneous agents (Qwen2.5-7B, Llama-3.1-8B, Ministral-3-8B) across $R{=}3$ debate rounds on two high-difficulty benchmarks (GSM-Hard and MMLU-Hard). We compare peer debate against isolated self-correction and a stochastic noise control that injects rationales from unrelated problems. We decompose debate failure into three model-dependent pathways: sycophantic conformity, where agents uncritically adopt majority answers (modal adoption up to 85.5%); contextual fragility, where peer rationales destabilize previously correct reasoning (vulnerability rate up to 70.0%); and consensus collapse, where plurality voting discards correct answers already present in the generation pool (oracle gap up to 32.3 percentage points). Ablations over communication density ($K \in \{2,4,9\}$) and sampling temperature ($T \in \{0.4, 0.7\}$) show that conformity reaches high levels at minimal peer exposure ($K{=}2$) and intensifies with greater initial diversity. Across all configurations, debate consumes 2.1-3.4$\times$ more tokens (up to 28,631 tokens per problem) than self-correction for equal or lower accuracy. Our results indicate that, within the 7-8B parameter class, homogeneous teams without structured roles do not benefit from unguided peer exchange, and that isolated self-correction consistently offers a more favorable cost-accuracy tradeoff.

Multi-agent debate has been shown to improve reasoning in large language models (LLMs). However, it is compute-intensive, requiring generation of long transcripts before answering questions. To address this inefficiency, we develop a framework that distills multi-agent debate into a single LLM through a two-stage fine-tuning pipeline combining debate structure learning with internalization via dynamic reward scheduling and length clipping. Across multiple models and benchmarks, our internalized models match or exceed explicit multi-agent debate performance using up to 93% fewer tokens. We then investigate the mechanistic basis of this capability through activation steering, finding that internalization creates agent-specific subspaces: interpretable directions in activation space corresponding to different agent perspectives. We further demonstrate a practical application: by instilling malicious agents into the LLM through internalized debate, then applying negative steering to suppress them, we show that distillation makes harmful behaviors easier to localize and control with smaller reductions in general performance compared to steering base models. Our findings offer a new perspective for understanding multi-agent capabilities in distilled models and provide practical guidelines for controlling internalized reasoning behaviors. Code available at https://github.com/johnsk95/latent_agents

LLM routing has achieved promising results in integrating the strengths of diverse models while balancing efficiency and performance. However, to support more realistic and challenging applications, routing must extend into agentic LLM settings, where task planning, multi-round cooperation among heterogeneous agents, and memory utilization are indispensable. To address this gap, we propose GraphPlanner, a heterogeneous graph memory-augmented agentic router for multi-agent LLMs that generates routing workflows for each query and supports both inductive and transductive inference. GraphPlanner formulates workflow generation as a Markov Decision Process (MDP), where at each step it selects both the LLM backbone and the agent role, including Planner, Executor, and Summarizer. By leveraging a heterogeneous graph, denoted as GARNet, to capture interaction memories among queries, agents, and responses, GraphPlanner integrates historical memory and workflow memory into richer state representations. The entire pipeline is optimized with reinforcement learning, jointly improving task-specific performance and computational efficiency. We evaluate GraphPlanner across 14 diverse LLM tasks and demonstrate that: (1) GraphPlanner outperforms strong single-round and multi-round routers, improving accuracy by up to 9.3% while reducing GPU cost from 186.26 GiB to 1.04 GiB; (2) GraphPlanner generalizes robustly to unseen tasks and LLMs, exhibiting strong zero-shot capabilities; and (3) GraphPlanner effectively leverages historical memories, supporting both inductive and transductive inference for more adaptive routing. Our code for GraphPlanner is released at https://github.com/ulab-uiuc/GraphPlanner.

Autonomous multi-agent LLM systems are increasingly deployed to investigate operational incidents and produce structured diagnostic reports. Their trustworthiness hinges on whether each claim is grounded in observed evidence rather than model-internal inference. Existing groundedness evaluators (binary classifiers, LLM-as-judge scalars, self-correction loops) treat supporting evidence as interchangeable and emit a single signal that offers no principled control over downstream action. We present GSAR, a grounding-evaluation and replanning framework that (i) partitions claims into a four-way typology (grounded, ungrounded, contradicted, complementary), giving first-class standing to non-redundant alternative perspectives; (ii) assigns evidence-type-specific weights reflecting epistemic strength; (iii) computes an asymmetric contradiction-penalised weighted groundedness score; and (iv) couples that score to a three-tier decision function (proceed, regenerate, replan) driving a bounded-iteration outer loop under an explicit compute budget. We formalise the algorithm, prove six structural properties, and evaluate five design claims on FEVER with gold Wikipedia evidence under four independently-trained LLM judges (gpt-5.4, claude-sonnet-4-6, claude-opus-4-7, gemini-2.5-pro). Every ablation reproduces in the same direction on every judge: bootstrap 95% CIs on the rho=0 effect exclude 0 on all four; the no-complementary ablation under Opus 4.7 has CI [-96,-68] of 200; at n=1000 three independent judges converge to DeltaS(rho=0)=+0.058. A head-to-head against Vectara HHEM-2.1-Open is included. To our knowledge, GSAR is the first published groundedness framework coupling evidence-typed scoring with tiered recovery under an explicit compute budget.

Existing multi-agent Large Language Model (LLM) frameworks for code generation typically use execution feedback and improve iteratively using Input/Output (I/O) test cases. However, this does not work for scientific workflows, where I/O test cases do not exist, and generating them requires solving the very problem at hand. To address this, we introduce MOSAIC, a training-free multi-agent framework for scientific code generation without I/O supervision. Instead of execution feedback, MOSAIC employs a student-teacher knowledge distillation framework that grounds generation through domain-specific examples and structured problem decomposition. To further mitigate hallucinations across chained subproblems, we introduce a Consolidated Context Window (CCW) for maintaining consistent reasoning across agents. Experiments on the SciCode benchmark show that MOSAIC improves accuracy, executability, and numerical precision over existing approaches while relying on lightweight models.

Automatically generating formal ontologies from unstructured natural language remains a central challenge in knowledge engineering. While large language models (LLMs) show promise, it remains unclear which architectural design choices drive generation quality and why current approaches fail. We present a controlled experimental study using domain-specific insurance contracts to investigate these questions. We first establish a single-agent LLM baseline, identifying key failure modes such as poor Ontology Design Pattern compliance, structural redundancy, and ineffective iterative repair. We then introduce a multi-agent architecture that decomposes ontology construction into four artifact-driven roles: Domain Expert, Manager, Coder, and Quality Assurer. We evaluate performance across architectural quality (via a panel of heterogeneous LLM judges) and functional usability (via competency question driven SPARQL evaluation with complementary retrieval augmented generation based assessment). Results show that the multi-agent approach significantly improves structural quality and modestly enhances queryability, with gains driven primarily by front-loaded planning. These findings highlight planning-first, artifact-driven generation as a promising and more auditable path toward scalable automated ontology engineering.

Multi-agent systems built on large language models have shown strong performance on complex reasoning tasks, yet most work focuses on agent roles and orchestration while treating inter-agent communication as a fixed interface. Latent communication through internal representations such as key-value caches offers a promising alternative to text-based protocols, but existing approaches do not jointly optimize communication with multi-agent reasoning. Therefore we propose DiffMAS, a training framework that treats latent communication as a learnable component of multi-agent systems. DiffMAS performs parameter-efficient supervised training over multi-agent latent trajectories, enabling agents to jointly learn how information should be encoded and interpreted across interactions. Experiments on mathematical reasoning, scientific QA, code generation, and commonsense benchmarks show that DiffMAS consistently improves reasoning accuracy and decoding stability over single-agent inference, text-based multi-agent systems, and prior latent communication methods, achieving 26.7% on AIME24, 20.2% on GPQA-Diamond, and consistent gains across reasoning benchmarks.

Multi-agent systems built from teams of large language models (LLMs) are increasingly deployed for collaborative scientific reasoning and problem-solving. These systems require agents to coordinate under shared constraints, such as GPUs or credit balances, where cooperative behavior matters. Behavioral economics provides a rich toolkit of games that isolate distinct cooperation mechanisms, yet it remains unknown whether a model's behavior in these stylized settings predicts its performance in realistic collaborative tasks. Here, we benchmark 35 open-weight LLMs across six behavioral economics games and show that game-derived cooperative profiles robustly predict downstream performance in AI-for-Science tasks, where teams of LLM agents collaboratively analyze data, build models, and produce scientific reports under shared budget constraints. Models that effectively coordinate games and invest in multiplicative team production (rather than greedy strategies) produce better scientific reports across three outcomes, accuracy, quality, and completion. These associations hold after controlling for multiple factors, indicating that cooperative disposition is a distinct, measurable property of LLMs not reducible to general ability. Our behavioral games framework thus offers a fast and inexpensive diagnostic for screening cooperative fitness before costly multi-agent deployment.

Explainability for Large Language Model (LLM) agents is especially challenging in interactive, partially observable settings, where decisions depend on evolving beliefs and other agents. We present \textbf{TriEx}, a tri-view explainability framework that instruments sequential decision making with aligned artifacts: (i) structured first-person self-reasoning bound to an action, (ii) explicit second-person belief states about opponents updated over time, and (iii) third-person oracle audits grounded in environment-derived reference signals. This design turns explanations from free-form narratives into evidence-anchored objects that can be compared and checked across time and perspectives. Using imperfect-information strategic games as a controlled testbed, we show that TriEx enables scalable analysis of explanation faithfulness, belief dynamics, and evaluator reliability, revealing systematic mismatches between what agents say, what they believe, and what they do. Our results highlight explainability as an interaction-dependent property and motivate multi-view, evidence-grounded evaluation for LLM agents. Code is available at https://github.com/Einsam1819/TriEx.

We present the first portfolio-level validation of MarketSenseAI, a deployed multi-agent LLM equity system. All signals are generated live at each observation date, eliminating look-ahead bias. The system routes four specialist agents (News, Fundamentals, Dynamics, and Macro) through a synthesis agent that issues a monthly equity thesis and recommendation for each stock in its coverage universe, and we ask two questions: do its buy recommendations add value over both passive benchmarks and random selection, and what does the internal agent structure reveal about the source of the edge? On the S&P 500 cohort (19 months) the strong-buy equal-weight portfolio earns +2.18%/month against a passive equal-weight benchmark of +1.15% (approximating RSP), a +25.2% compound excess, and ranks at the 99.7th percentile of 10,000 Monte Carlo portfolios (p=0.003). The S&P 100 cohort (35 months) delivers a +30.5% compound excess over EQWL with consistent direction but formal significance not reached, limited by the small average selection of ~10 stocks per month. Non-negative least-squares projection of thesis embeddings onto agent embeddings reveals an adaptive-integration mechanism. Agent contributions rotate with market regime (Fundamentals leads on S&P 500, Macro on S&P 100, Dynamics acts as an episodic momentum signal) and this agent rotation moves in lockstep with both the sector composition of strong-buy selections and identifiable macro-calendar events, three independent views of the same underlying adaptation. The recommendation's cross-sectional Information Coefficient is statistically significant on S&P 500 (ICIR=+0.489, p=0.024). These results suggest that multi-agent LLM equity systems can identify sources of alpha beyond what classical factor models capture, and that the buy signal functions as an effective universe-filter that can sit upstream of any portfolio-construction process.

With an ever-growing zoo of LLMs and benchmarks, the need to orchestrate multiple models for improved task performance has never been more pressing. While frameworks like Mixture-of-Agents (MoA) attempt to coordinate LLMs, they often fall short in terms of (1) selecting relevant agents, (2) facilitating effective intra-agent communication, and (3) integrating responses efficiently. In this work, we propose Graph-of-Agents (GoA), a new graph-based framework for modeling multi-agent LLM communication. Our approach begins with node sampling, selecting only the most relevant agents by leveraging model cards that summarize each model's domain, task specialization, and other characteristics. Next, we construct edges between the selected agents by evaluating their responses against one another to determine relevance ordering. Directed message passing is then performed from highly relevant agents to less relevant ones to enhance their responses, followed by reverse message passing to refine the original responses of the more relevant agents. Finally, the updated responses are aggregated via graph-based pooling (e.g., max or mean pooling) to produce a single, unified answer. We evaluate GoA on diverse multi-domain benchmarks (MMLU, MMLU-Pro, GPQA) and domain-specific benchmarks (MATH, HumanEval, MedMCQA), with an agent pool of 6 LLMs spanning multiple domains. Surprisingly, GoA achieves superior performance using only 3 selected agents, outperforming recent multi-agent LLM baselines that utilize all 6 agents simultaneously. By adopting a graph structure, GoA offers both scalability and effectiveness through structured message passing-positioning it as a strong candidate for navigating the challenges of the ever-growing LLM zoo. Code is available at: https://github.com/UNITES-Lab/GoA.

Multi-agent large language model (LLM) architectures increasingly rely on response-level aggregation, such as Majority Voting (MAJ), to raise reasoning ceilings. However, in open environments, agents are highly susceptible to stealthy contextual corruption, such as targeted prompt injections. We reveal a critical structural vulnerability in current multi-agent systems: response-level aggregation collapses when corrupted agents form a local majority. Because voting aggregates fully-formed conclusions, it is blind to flawed intermediate logic. To overcome this systematic limitation, we propose the Token-Level Round-Robin (RR) Collaboration, where agents sequentially interleave generation within a shared auto-regressive context. We formalize this process as a discrete-time dynamical system, proving that token-level interleaving transitions aggregation from a brittle counting of final votes (a linear sum) to a dynamic, interwoven chain of logic (a non-linear operator product). Through this theoretical lens, we prove that the honest model's restorative pull can overpower adversarial corruptions, even when corrupted agents form a majority. We conduct an exhaustive empirical evaluation across diverse reasoning benchmarks and demonstrate that while MAJ collapses when corrupted agents reach a majority, RR maintains robust accuracy well beyond this critical threshold.

Large Language Models (LLMs) have demonstrated potential in automating scientific ideation, yet current approaches relying on iterative prompting or complex multi-agent architectures often suffer from hallucination or computational inefficiency. A critical bottleneck in applying Reinforcement Learning (RL) to this open-ended domain is reward hacking -- where models exploit imperfect evaluation proxies to maximize scores without producing genuine scientific innovation. To address these limitations, we propose an RL framework explicitly tailored for high-quality scientific idea generation. We propose the first multi-agent reward function designed to serve as a judge, decoupling methodological validation from implementation details while providing strict binary rewards that are robust to reward hacking. To effectively optimize against this sparse signal, we utilize an unbiased variant of Group Relative Policy Optimization to mitigate artificial length bias. We grounded our training in ICLR-320, a curated dataset of problem-solution pairs extracted from ICLR 2024 proceedings. Experiments demonstrate that our framework significantly outperforms state-of-the-art baselines across expert-evaluated metrics of novelty, feasibility, and effectiveness.

Large language models (LLMs) with a large number of parameters achieve strong performance but are often prohibitively expensive to deploy. Recent work explores using teams of smaller, more efficient LLMs that collectively match or even outperform a single large model. However, jointly updating multiple agents introduces compounding distribution shifts, making coordination and stability during training difficult. We address this by introducing Sequential Agent Tuning (SAT), a coordinator-free training paradigm. SAT represents the team as a factorized policy and employs block-coordinate updates over agents, enabling scalable, decentralized training without a central controller. Specifically, we develop a sequence-aware, on-policy advantage estimator that conditions on the evolving team policy, coupled with per-agent KL trust regions that isolate occupancy drift. Theoretically, this framework provides two critical guarantees. First, it ensures monotonic improvement, stabilizing the training process. Second, it establishes provable plug-and-play invariance: any agent can be upgraded to a stronger model without retraining the rest of the team, with a formal guarantee that the performance bound improves. Empirically, a team of three 4B agents (12B total) trained with SAT surpasses the much larger Qwen3-32B on AIME24/25 benchmarks by 3.9% on average. We validate our plug-and-play theory by swapping in two 8B agents, which boosts the composite score by 10.4%. We provide code and appendix of proof at https://github.com/Yydc/SAT-AAMAS

Communication in Large Language Model (LLM)-based multi-agent systems is moving beyond discrete tokens to preserve richer context. Recent work such as LatentMAS enables agents to exchange latent messages through full key-value (KV) caches. However, full KV relay incurs high memory and communication cost. We adapt KV-cache eviction methods to this setting and introduce \textbf{Orthogonal BackFill (OBF)} to mitigate information loss from hard eviction. OBF injects a low-rank orthogonal residual from discarded KV states into the retained KV states. We evaluate OBF against full KV relay on nine benchmarks spanning mathematical reasoning, expert and commonsense QA, and coding. With only 9.9%-20.2% of the prompt KV states retained, H-OBF delivers between $97$ and $120$ of full KV relay's per-benchmark accuracy across the nine benchmarks. This suggests that more information does not necessarily lead to better communication; preserving the most useful information matters more. Our codebase is included in the supplementary material. Our codebase is publicly available on https://github.com/markli404/When-Less-Latent-Leads-to-Better-Relay.

Language models are increasingly used to simulate people: survey respondents, negotiators, stakeholders in policy exercises. In that role a model should reproduce how people plausibly behave, hesitating, conceding late, and settling for imperfect deals, rather than playing the best move. We call this the sampler role, in contrast to the solver role of finding the best move, and we test how the reasoning modes providers ship to strengthen models as solvers affect it. Our testbed is multi-party negotiation: five agents bargain over a regulation for fifteen turns, and unresolved issues are decided by an authority. Agents without a structured memory of the negotiation almost never reach agreement, whether reasoning is on or off: 314 of 315 such runs end with the authority deciding. What reasoning changes is how the failure looks. With reasoning enabled, one model family negotiates visibly, with varied moves, concessions in most runs, and a different path every time, yet still ends without agreement in fifteen runs of fifteen. Diversity checks would pass a model whose endings never change. Two further results show the task permits agreement: when agents write their own short running notes on the negotiation, agreement becomes the norm, while the same notes supplied ready-made change nothing; and hand-coded agents following textbook concession strategies agree in most runs under identical rules. Simulation pipelines should therefore vet models as samplers, on the distributions of outcomes they produce. Fidelity as a sampler must be tested on its own: solver strength is no guide to it, and switching on reasoning leaves it where it was.

We develop a unified treatment of credit assignment for RL training in multi-agent LLM systems. We show that observed reward alone cannot distinguish an agent that determines it from one that never affects it, and that standard shared-reward training performs exact gradient ascent on each agent's private utility rather than system performance. Moreover, we prove no single scalar per agent can consistently account for joint performance once agents interact. We thus develop the unique background-dependent notion of marginal contribution satisfying natural consistency requirements. From it we derive gradient-correct marginal contribution training signals, identify them from filtered feedback, and optimally allocate a budget of exact counterfactual evaluations against learned-signal error. Instantiated in GRPO, our signal improves routed GSM8K accuracy over winner-take-all training at no extra generation cost.

Large language models (LLMs) often exhibit sycophancy: agreement with user stance even when it conflicts with the model's opinion. While prior work has mostly studied this in single-agent settings, it remains underexplored in collaborative multi-agent systems. We ask whether awareness of other agents' sycophancy levels influences discussion outcomes. To investigate this, we run controlled experiments with six open-source LLMs, providing agents with peer sycophancy rankings that estimate each peer's tendency toward sycophancy. These rankings are based on scores calculated using various static (pre-discussion) and dynamic (online) strategies. We find that providing sycophancy priors reduces the influence of sycophancy-prone peers, mitigates error-cascades, and improves final discussion accuracy by an absolute 10.5%. Thus, this is a lightweight and efficient way to reduce model sycophancy during discussions and subsequently improve downstream accuracy.

When automating plan generation for a real-world sequential decision problem, the goal is often not to replace the human planner, but to facilitate an iterative reasoning and elicitation process, where the human's role is to guide the AI planner according to their preferences and expertise. In this context, explanations that respond to users' questions are crucial to improve their understanding of potential solutions and increase their trust in the system. To enable natural interaction with such a system, we present a multi-agent Large Language Model (LLM) architecture that is agnostic to the explanation framework and enables user- and context-dependent interactive explanations. We also describe an instantiation of this framework for goal-conflict explanations, which we use to conduct a user study comparing the LLM-powered interaction with a baseline template-based explanation interface.

Multi-agent LLM frameworks are data-intensive systems that govern how agents orchestrate tasks, manage state, and coordinate decisions. These architectural choices control execution overhead, memory behavior, planning effectiveness, and coordination scalability. Their impact on system performance remains poorly understood. Existing benchmarks evaluate individual agent capabilities in isolation and lack standardized framework-level comparison. We make four contributions. We introduce an architectural taxonomy that decomposes multi-agent LLM frameworks along five dimensions: orchestration, memory, planning interfaces, specialization, and communication topology. We develop MAFBench, a unified evaluation suite that integrates existing benchmarks within a standardized execution pipeline. We conduct a controlled empirical study across nine frameworks, fixing the underlying LLM and varying only architectural design choices. We distill the results into six evidence-based design principles. Architectural design, not only model intelligence, governs performance. Orchestration alone increases latency by over 60x, and a minimal implementation of the same paradigm isolates that cost as implementation rather than paradigm. Schema-constrained planning interfaces reduce accuracy by up to 32 points through formatting failures, not reasoning errors. Communication topology drops coordination success from above 90% to below 30% under mismatched structure. Memory architecture controls recall and scalability independent of context window size, and no evaluated framework natively supports controlled knowledge revision.

Large Language Models (LLMs) in multi-agent systems (MAS) have shown promise for complex tasks, yet current training methods lack principled ways to connect system-level evaluation with agent- and message-level learning. We propose a theoretical framework that unifies cooperative game-theoretic attribution with process reward modeling to transform system evaluation to agent credit to response-level signals. Unlike prior approaches that rely only on attribution (Shapley) or step-level labels (PRM), our method produces local, signed, and credit-conserving signals. In success cases, Shapley-based credit assignment fairly allocates outcomes across agents and is refined into per-message rewards that promote cooperation while discouraging redundancy or sabotage; in failure cases, first-error localization yields repair-aware preferences that penalize harmful steps while rewarding corrective attempts. The resulting signals are bounded, cooperative, and directly compatible with reinforcement- or preference-based post-training, providing a unified and auditable pathway from global evaluation to local supervision in LLM multi-agent training. Our contribution is conceptual: we present a theoretical foundation and training signals, leaving empirical validation for future work.

We introduce Debate2Create (D2C), a multi-agent LLM framework that formulates robot co-design as structured, iterative debate grounded in physics-based evaluation. A design agent and control agent engage in a thesis-antithesis-synthesis loop, while criterion-specific LLM judges provide multi-objective feedback to steer exploration. Across five MuJoCo locomotion benchmarks, D2C achieves the highest default-normalized score among the evaluated LLM-based and black-box baselines, with gains up to 3.2x on Ant and nearly 9x on Swimmer. Iterative debate yields 18-35% gains over compute-matched zero-shot generation, and D2C-generated rewards transfer to default morphologies in 4/5 tasks. These results suggest that structured, simulator-grounded multi-agent interaction is a useful mechanism for joint morphology-reward optimization under a fixed-topology, per-candidate-RL protocol. Project page: debate2create.github.io.

Thematic analysis (TA) is a widely used qualitative approach for uncovering latent meanings in unstructured text data. TA provides valuable insights in healthcare but is resource-intensive. Large Language Models (LLMs) have been introduced to perform TA, yet their applications in high-stakes healthcare settings, particularly for qualitative clinical interview analysis, remain limited. Here, we propose TAMA: A Human-AI Collaborative Thematic Analysis framework using Multi-Agent LLMs for clinical interviews. We leverage the scalability and coherence of multi-agent systems through structured conversations between agents and coordinate the expertise of cardiac experts in TA. Using interview transcripts from parents of children with Anomalous Aortic Origin of a Coronary Artery (AAOCA), a rare congenital heart disease, we demonstrate that TAMA outperforms single-agent LLM TA approaches, achieving higher thematic hit rate, coverage, and distinctiveness. TAMA demonstrates strong potential for automated TA in clinical settings by leveraging multi-agent LLM systems with human-in-the-loop integration by enhancing quality while significantly reducing manual workload. The full implementation is publicly available at https://github.com/Charlie-Yi-SJ/TAMA.