We study connectivity-preserving finite-jump dispersion of anonymous, identical, and oblivious agents under an idealized range-only sensing model. Each agent measures only the distances to its visible neighbors, without bearings, identifiers, communication, memory, or a shared coordinate system. We derive the largest isotropic displacement certifiable as safe from these measurements alone. The resulting rule requires only the distance to the farthest visible neighbor: each agent selects a random direction and moves by half of its remaining visibility margin. The rule preserves every existing visibility edge under synchronous finite motion and therefore preserves connectivity. For two agents, we prove positive conditional drift in squared distance, almost-sure convergence to the visibility boundary, and finite expected time to reach any fixed neighborhood of that boundary. A one-million-run Monte Carlo experiment agrees with the exact first-round moments and estimates approximately 9.5 rounds to reach distance 0.97V from coincident initial positions; an independent Bellman-equation computation gives the same estimate. For general swarms, 1,000 runs across five initial-topology classes reproduce the deterministic safety guarantee at implementation level and reveal a consistent topology-dependent ordering of attainable diameter under the tested protocol. These results provide a theoretical foundation for connectivity-preserving multi-robot dispersion under minimal sensing, while isolating the guarantees achievable from anonymous range measurements alone.
AI's inflection point is a fountain of alarming risks. Digital platforms in particular are vulnerable to three major AI risks, as those in control of global platforms cede control to AI, of the algorithms, of membership and of content: (1) AI control of global platforms, social and economic, used by the majority of humanity, resulting in the majority of humanity becoming vulnerable to AI control. (2) Unaccountable AI agents operating anonymously on social and economic digital platforms, side-by-side with humans, possibly masquerading as humans, serving unknown masters with unknown goals, operating under unknown restraints without accountability or culpability. (3) Unattributed deepfakes generated by anonymous AI, with people unable to distinguish them from genuine content and with their producers evading accountability and culpability. Here, we propose to corral these AI risks with grassroots contracts, a digital incarnation of a two-centuries old notion of Pierre-Joseph Proudhon. We present grassroots contracts for a social graph with accountable AI agents and for a social network with digital provenance, and working smartphone apps that implement the two and show how they corral these major AI risks: (1) AI control of global platforms. (2) Unaccountable AI agents. (3) Unattributed deepfakes.
Sophie Hall, Kai Zhang, Ilia Shilov, Heinrich H. Nax, Saverio Bolognani
At the core of most socio-technical systems lies a scarce resource that is allocated among agents: highway lanes, public transit, road space, water rights, energy access, grid capacity, user attention, pollution rights, etc. With further automation of the underlying allocation processes, control engineers are increasingly tasked with making decisive assumptions regarding what society wants. In practice to date, design choices are largely driven by industry norms and conventions rather than a result of conscientiously responsible and ethical design. In this paper, we look at tools available to control engineers to design systems in a more principled manner in order to match the societal mandate. We consider three control design paradigms: online feedback optimization, control of Markov decision processes, and model predictive control. Beginning with aggregating individual agents' preferences into control design objectives, subsequently ensuring and certifying the fulfillment of those specifications, we argue that the feedback nature of control systems enables appropriate allocation of the shared resources in ways hitherto unparalleled.
When is another AI research run worth more than keeping that money invested? Using 8,120 iPulse AI reports that consumed nearly 100 million output and reasoning tokens on the same 203 stocks across four production batches, we compare repricing old targets (with and without shock safeguards), mechanical growth-path resets and fresh multi-agent reports. Ten matched configurations represent seven investment frameworks. The primary economic test selects equal-weight Top 10 USD portfolios from one-year forecasts, net of 10 bps trading costs and a USD 3,000 all-in charge for fresh research on a USD 300,000 account. In two unchanged-model short windows, guarded reuse averages 2.25%, versus 1.66% for SPY and 4.24% for fresh reports. Guarded reuse beats SPY in all three short windows. Extending the first holding window to 176 days reverses the comparison: guarded reuse returns -11.26%, fresh reports +8.55%, and SPY +3.46%, motivating the thesis half-life question. Supporting quarter-forecast diagnostics show that mechanically resetting old growth paths closes 98% of the MAE gap to fresh reports, yet Fresh and Reset share only 1-4 Top 10 stocks. Top 30 reverses the unchanged-model comparison in favor of guarded reuse (4.17% versus 1.37%). Concentration tests expose dependence on exceptional winners. Nine editorial editions show substantial between-edition turnover, but each final screen retains its own incoming list. Recurring budget scenarios retain initial holdings: monthly costs approach realized short-window refresh value, while weekly budgets require larger additional gains. A matched ranking-continuity extension extracts further value from retained forecasts without eliminating thesis ageing. The architecture separates continuous price monitoring, episodic thesis rebuilding and evidence-based editorial checks.
The EU Platform Work Directive, to be transposed by December 2026, requires ride-hailing platforms to state in plain language the main parameters of their automated decisions and how a worker's behaviour affects them. Such rules are exactly what a large language model (LLM) can read, so drivers will be able to let one decide for them. We compare two main families of rebalancing policy once their rules are public: a reactive policy (R) that sends idle drivers to cover requests just lost, and an anticipatory policy (A) that plans moves on a demand forecast. Each driver is an LLM agent maximizing its own net income, given no rules (N) or the rules at the directive's level of detail (L). We test two LLMs, Claude Haiku 5.5 and Jev, which understand the rules equally well, in a three-zone model and in Manhattan. With the rules disclosed, drivers follow the platform more closely: in Manhattan, the share of relocation offers they decline falls from about four in five to one in four under both policies. Yet the effects differ by policy. For the platform, disclosure costs R 3.7% of its profit in Manhattan, as R now pays for moves drivers used to make for free, while A's profit is unchanged; the two policies, which tie on profit under full compliance, end 6.1% apart. For drivers, disclosure raises income in the small model under both policies and LLMs; for users, it raises the service rate there by 10 to 18 percentage points. In Manhattan, self-interested drivers also make A a win-win-win against full compliance, raising driver income, platform profit and service rate by 10 to 12% each. Rebalancing policies that perform equally well under full compliance can thus show opposite results under mandatory disclosure. The study offers a framework to evaluate such policies under the new requirement and is, to our knowledge, the first to do so with LLM agents.
Shu Yang, Shenzhe Zhu, Hao Zhu, José Ramón Enríquez, Di Wang, Alex Pentland, Michiel A. Bakker, Jiaxin Pei
Large language models (LLMs) and LLM-based agents are increasingly deployed as assistants in planning and decision making, yet most existing systems are implicitly optimized for a single-principal interaction paradigm, in which the model is designed to satisfy the objectives of one dominant user whose instructions are treated as the sole source of authority and utility. However, as they are integrated into team workflows and organizational tools, they are increasingly required to serve multiple users simultaneously, each with distinct roles, preferences, and authority levels, leading to multi-user, multi-principal settings with unavoidable conflicts, information asymmetry, and privacy constraints. In this work, we present the first systematic study of multi-user LLM agents. We begin by formalizing multi-user interaction with LLM agents as a multi-principal decision problem, where a single agent must account for multiple users with potentially conflicting interests and associated challenges. We then introduce a unified multi-user interaction protocol and design three targeted stress-testing scenarios to evaluate current LLMs' capabilities in instruction following, privacy preservation, and coordination. Our results reveal systematic gaps: frontier LLMs frequently fail to maintain stable prioritization under conflicting user objectives, exhibit increasing privacy violations over multi-turn interactions, and suffer from efficiency bottlenecks when coordination requires iterative information gathering.
There has been significant recent progress in algorithms for approximation of Nash equilibrium in large two-player zero-sum imperfect-information games and exact computation of Nash equilibrium in multiplayer strategic-form games. While counterfactual regret minimization and fictitious play are scalable to large games and have convergence guarantees in two-player zero-sum games, they do not guarantee convergence to Nash equilibrium in multiplayer games. Recently, an approach has been presented for exact computation of Nash equilibrium in multiplayer imperfect-information games that solves a quadratically constrained program based on a nonlinear complementarity problem formulation derived from the sequence-form game representation. This formulation was solved using Gurobi's nonconvex quadratic solver, which employs spatial branch-and-bound to iteratively refine variable bounds by solving convex relaxations of bilinear terms. During presolve, Gurobi introduces auxiliary variables and, in some cases, binary variables, leading to an internal mixed-integer quadratically constrained program. This approach was demonstrated to outperform prior algorithms from the Gambit software suite and quickly solve three-player Kuhn poker after removal of dominated actions; however, the algorithm was not able to solve the full version of the game within 24 hours. In this paper, we derive finite bounds on slack and multiplier variables in the nonlinear complementarity formulation. These bounds can strengthen the convex relaxations used within spatial branch-and-bound. We compare uniform corollary bounds with finer proposition-based bounds on full three-player Kuhn poker. All eight bounded configurations return numerically verified equilibria in under 17 seconds, while the unbounded formulation finds no feasible solution within 300 seconds. The finest bounds do not consistently yield the fastest computation.
Achuth Chandrasekhar, Jesse Barkley, Peter Pak, Rumi Loghmani, Aakash Oza, Amir Barati Farimani
Mechanical engineers use specialized tools for CAD, simulation, manufacturing and research, each with a distinct interface and learning curve. We present Mechanical Engineering Skills, an open library of 33 agent skills that makes these capabilities accessible through a single text or voice conversation. Each skill combines operating instructions, executable tools and verification procedures. A language model interprets the request, selects and coordinates skills, while deterministic tools perform computations and return checked results. Default workflows require no tool-specific interface training, local model execution or model fine-tuning; skills load on demand and install with one command. Nineteen skills adapt published or preprint research from our laboratory, including text-to-CAD, linkage synthesis, CFD setup, stress-field and lattice surrogates, equation discovery, additive-manufacturing planning and in-process monitoring. Fourteen additional skills support core engineering and education. We describe the skill format, design principles and a method for translating research into practical workflows. Five case studies with explicit evidence scopes and independent reruns demonstrate exact analytical agreement for lattice homogenization, rocket apogee predictions within 4% of stored OpenRocket references, an equation fit nine times more accurate than a cubic polynomial at an unseen temperature, and an end-to-end rehearsal of porosity monitoring. Controlled agent comparisons remain future work. The open library is available at https://github.com/BaratiLab/mechanical-engineering-skills.
Detecting coordination among unmanned aerial vehicle (UAV) fleets operating in shared airspace and identifying the route-lead aircraft whose navigation decisions govern fleet behavior presents a fundamental speed--accuracy trade-off: fast methods enable real-time traffic management but sacrifice detection fidelity, while accurate methods may exceed the time budget for actionable airspace deconfliction. This paper presents a game-theoretic decision framework that resolves this trade-off by formulating method selection as a two-player zero-sum game between a Monitor (selecting computational methods and parameters) and Nature (selecting the unknown traffic scenario). We construct an end-to-end pipeline from trajectory surveillance data through eight candidate detection algorithms, a Monte Carlo sensitivity analysis characterizing their stochastic performance, and finally a multi-objective optimization layer that identifies Pareto-optimal method portfolios. The minimax solution provides a robust mixed strategy with a probability distribution over methods that guarantees worst-case performance regardless of scenario uncertainty. Experimental evaluation across 200 randomized configurations spanning 5--50 aircraft demonstrates that the framework recommends distinct method portfolios depending on operational priority: Koopman Phase dominates balanced (70.6%) and speed-priority (79.7%) profiles, while CRQA emerges as primary (47.4%) when route-lead identification is prioritized. The framework achieves a guaranteed game value of 0.29--0.53 (normalized utility) across all tested preference profiles, providing the first principled, scenario-adaptive methodology for computational method selection in UTM fleet monitoring operations.
The bitter lesson holds that general methods leveraging computation defeat hand-encoded domain knowledge. We argue it carries a condition its popular reading omits: in every case Sutton cites, the general method was scaled against the same task the domain engineering was solving. Social simulation substituted across tasks instead, replacing hand-built agents with an artifact optimized for predicting text. The move is bitter-lesson-shaped, but the lesson does not license it. Prediction is a solitary act: the predictor bears no consequences from the interaction it models. Cooperation, defection, and norm enforcement are phenomena of interested parties; they do not exist for a disinterested observer at any scale. What is missing is not emergence, which LLM populations already show, but stakes. We show that several anomalies reported in prior work (excess cooperation that predates alignment training, collective agreement without individual fidelity, and population homogeneity) follow from this account. A 540-agent pilot with pre-specified hypotheses, which tests sensitivity to persistent consequences rather than cooperation, finds that consequences reshape strategies in two of three model families but never produce human-like heterogeneity. We argue for mixed-motive settings that shape agents rather than only test them, and ask whether agents should be pretrained differently from predictors.
Collaborative multi-agent reinforcement learning (MARL) is increasingly deployed in critical domains, yet its vulnerability to adversarial perturbations has not been thoroughly investigated. Most existing adversarial attacks against MARL rely on optimistic assumptions, such as access to policy weights or the ability to train surrogate policies. We investigate whether adversaries with limited capabilities remain effective against MARL agents. In particular, we study two restricted attack settings that are overlooked in the literature: (i) adversaries with access only to agent observations and (ii) adversaries with no access whatsoever. Our approach generates perturbations that intentionally misalign how victim agents perceive their environment, preventing them from collaborating. We conduct extensive experiments across a diverse set of 22 environments. Our results demonstrate both the effectiveness and sample efficiency of our attacks.
Scientific discovery emerges from cycles of hypothesis, implementation, empirical testing, and feedback. Can this process be automated? We approach automated algorithm design through the lens of the scientific method, where an LLM-based agent goes through each step of the process in an iterative fashion. Results: We present The Little Scientist, a framework that maps the scientific method onto an LLM agent in a loop: a Scientist agent iterates hypothesis, prediction, code, test, and reconciliation inside an evaluation environment that returns structured diagnostics, and a Kuhn agent injects paradigm-shifting reframings at plateaus. On DNA motif discovery, the process wrote an algorithm from scratch (DALE) that outperforms STREME across 132 ENCODE transcription factors on the shuffled-negative benchmark (mean AUROC 0.842 vs. 0.803) while running 11x faster. We then asked why this method worked, and that question became the paper's scientific core: a preregistered, controlled teardown of the framework's own scaffolding consisting of ten ablation arms against one-shot bandit nulls. The answer overturned the premise. Seven of ten loop arms fail to beat a bandit drawing samples from the same model and prompt, the full scaffold performs below blind sampling at 7-8x the token cost, and a floor-prompt arm with no framework at all wins on both score and value per token. On a held-out genomic benchmark, the best loop artifact and the best bandit draw agree to within 0.002, and the DALE algorithm from the initial experiment scored similarly. The convergent picture is that LLM agents do not reason their way past what training placed in the weights. Autoresearch, on this evidence, is sampling, selection, and remixing, which is the bet the Evolve-family architectures made when they chose populations and external selection over scratchpad reasoning, and why those choices were right.
Maxime Saxena, Fabio Caccioli, R. Maria del Rio-Chanona
Decision Models (DMs), a new class of AI models, give fast responses and return a probability distribution over a fixed set of options. These properties make them candidate human surrogates in economic simulations and surveys, a role increasingly filled by Large Language Models (LLMs), which are costly at scale and under-represent the variation in human behaviour. We present a first evaluation of DMs as human surrogates in economics, comparing 10 DMs with 8 LLMs in three economic settings: a household spending survey on consumption, several classic economic games and a labour-market field experiment. Frontier DMs, those that score highest on general decision-making benchmarks, reproduce the variation in human responses more closely than LLMs in all three settings. They capture the modal human response as well as LLMs in the spending survey, but less well in the economic games. Frontier DMs are also roughly ten times faster than the fastest LLM and can be over a hundred times cheaper. Together, these results suggest that DMs capture the heterogeneity of human behaviour better than LLMs and make large-scale economic simulations more feasible.
We propose MA-SafeDiffuser, a multi-agent extension of SafeDiffuser that equips diffusion-based trajectory planners with finite-time diffusion invariance guarantees for joint-agent safety specifications. Building on the single-agent construction that embeds control barrier function (CBF) constraints into reverse-diffusion updates, we: (i) formalize joint safe sets as intersections of per-agent and pairwise barrier sets; (ii) derive centralized and decentralized (communication-aware) constrained denoising procedures with provable invariance under mild assumptions; (iii) address local-trap and deadlock phenomena via time-varying specifications and liveness CBFs; and (iv) develop a lightweight benchmarking suite including a multi-agent Maze2D domain. Empirically, MA-SafeDiffuser reduces violation counts relative to unconstrained diffusion baselines, while retaining planning quality. We provide algorithms, proofs, and reference implementations.
Large language model (LLM) agents can autonomously run neuroimaging analyses from research questions to reported findings. Existing agentic systems use tools to check that code runs and reported values match its output, but rely on LLM judgment to decide which results warrant scientific reporting. We show that LLM judges endorse site-confounded findings even when the confound is presented as evidence. We present \textbf{CONFIRM}, a framework that separates running an analysis from deciding which findings to report. CONFIRM encodes each hypothesis as a machine-readable, pre-specified analysis plan that fixes the claim and its pass criteria before computing results. It then filters results through five deterministic gates (search history and multiplicity, confounds, power, analytic stability, and replication) under three configurable reporting standards, from least to most demanding: \textit{Screen}, \textit{Replicate}, and \textit{Full}. Failed claims receive diagnoses that guide follow-up analyses. To evaluate reporting decisions at the claim level, we introduce \textbf{NeuroClaimBench}, comprising 489 executable neuroimaging claims with frozen data, including literature-supported findings and 150 controls designed to violate individual methodological requirements. On NeuroClaimBench, CONFIRM falsely supports at most 3\% of controls, compared with 23--38\% for the baselines, while recovering up to 61\% of literature-supported claims. When auditing an existing agent, CONFIRM reduces false support on controls from 23\% to 0\% while retaining the agent's literature-supported claims. Its diagnostic feedback also produces more supported revisions than Self-Refine with fewer LLM calls. Together, CONFIRM and NeuroClaimBench show that basing reporting decisions on pre-specified analysis plans and deterministic gates is essential for trustworthy autonomous scientific discovery.
Swarm morphogenesis requires robots to assemble desired shapes through local interactions. We present a modular framework integrating neural cellular automata (NCA) with existing site assignment and collision-aware control. Within this framework, we introduce a decentralized attachment-site discovery mechanism. Robots already in the formation update and share local NCA states to identify new attachment sites, enabling progressive growth without a predefined list of sites during deployment. To bridge the mismatch between conventional NCA growth and gradual physical assembly, we additionally propose offline sparse-retention fine-tuning that simulates partial attachment, together with a false-positive loss that penalizes invalid site activations. Simulations demonstrate formation of diverse lattice morphologies, scaling to several hundred robots, and substantial improvements in morphology accuracy from the proposed fine-tuning. Experiments with Crazyflie quadrotors validate the integrated execution pipeline using robot-relative observations reconstructed from external tracking.
Large language models (LLMs) are increasingly deployed as autonomous agents that negotiate, coordinate, and make decisions on behalf of human users. Understanding how they cooperate is therefore important for AI governance and machine behaviour research. We examine two factors: payoff magnitude and interaction language. Classical game theory predicts that multiplying all payoffs by a positive constant should not change behaviour because the underlying strategic incentives remain unchanged. We test this prediction using five widely adopted commercial LLMs playing an iterated prisoner's dilemma across ten payoff scales and five languages (English, French, Vietnamese, Chinese and Arabic). Contrary to theoretical expectations, changing only the numerical scale of payoffs substantially alters cooperation. Models respond differently, often in opposite directions, meaning that averages across models can conceal important behavioural heterogeneity. Sensitivity to payoff scale also depends strongly on prompt language. Moreover, payoff scale modifies the effects of persona instructions: at sub-unit stakes, where no payoff shown exceeds 1, all five models cooperate more under a selfish than a cooperative persona, whereas Claude and Gemini reverse this pattern at larger scales. These findings show that cooperation measures depend critically on payoff scale and language and should not be treated as fixed model traits.
As older adults (OAs) increasingly favor aging in place, remote family members (RFMs) need meaningful awareness of their loved ones without intrusive monitoring. Large language models can provide rich interpretations of non-intrusive sensing data, yet recent work has largely emphasized technical accuracy over how families use or benefit from this information. We conducted formative interviews with 12 RFMs and 2 OAs, followed by a three-month deployment case study with one OA-RFM dyad. Our system provided daily AI-generated summaries of multimodal sensing data, which we iteratively refined with both parties using a research-through-design approach. Although the summaries integrated well into the dyad's daily routine, sustaining over 90% engagement, their lived experience revealed evolving tensions around agency, reassurance, privacy, trust, and communication, shaped by sensor limitations and existing communication misalignments. Our findings inform the design of future AI-mediated sharing systems that account for real-world sensing constraints and complex family care dynamics.
Multi-agent systems offer significant advantages in terms of flexibility and redundancy, enabling the efficient distribution of complex tasks among multiple robots. However, their deployment requires advanced control strategies to ensure optimal cooperation between autonomous agents and human operators. We propose a modification of the global connectivity maintenance algorithm to incorporate measurements taken in agent's local frame, using sensors with limited fields of view. This algorithm, responsible for the autonomous behavior of the multi-robot system, is combined with different adaptive supervision strategies, allowing the user to exert control at varying scales over the entire group. (i) Single Agent Control (S): The operator directly controls a single robot, (ii) Pinning Control (P): The operator imposes constraints on multiple robots to influence the group, (iii) Centroid Control (C): The operator acts on the system's center of mass to steer the entire fleet. In conjunction with the connectivity maintenance algorithm, these methods ensure compliance with inter-agent constraints while providing haptic feedback to the user regarding the system's connectivity state. The effectiveness of these approaches was evaluated through a user study, comparing performance with and without haptic feedback across multiple scenarios tailored to the different control modes.
Muhammad Sohail Danish, Gilles Q. Hacheme, Akram Zaytar, Satej S. Soman, Inbal Becker-Reshef, Juan Lavista Ferres, Salman Khan, Girmaw Abebe Tadesse
Real-world geospatial decision support requires more than single-step retrieval or code generation. Disaster-response questions often involve multi-source workflows that combine spatial databases, map-derived features, satellite imagery, raster products, and reusable geospatial artifacts, where correctness depends on preserving spatial scope, temporal context, and intermediate execution state. While large language models and coding agents offer a promising interface for such workflows, they struggle with geospatial-specific failure modes, including handling large artifacts, selecting the wrong evidence source, losing spatial or temporal metadata, and applying invalid geometric operations. We introduce GeoFaham, a domain-specialized multi-agent system for geospatial disaster analytics. GeoFaham decomposes workflows into specialist agents for vector operations, map feature retrieval, imagery discovery, and raster analysis, coordinated by an Orchestrator with a dual-ledger control loop for dependency-aware planning, progress tracking, validation, and replanning. To support artifact-centric collaboration, GeoFaham uses a Response Store that persists full geospatial outputs and typed metadata outside the LLM context window while agents exchange compact identifiers and summaries. We also introduce GeoFaham-Bench, a 40-task benchmark grounded in 17 real-world disaster records, designed to evaluate multi-step geospatial reasoning, cross-source integration, raster-vector computation, artifact generation, and recovery under non-rigid ground truth. We evaluate GeoFaham on GeoFaham-Bench against strong general-purpose coding-agent baselines, including Codex, Claude Code, and OpenCode. GeoFaham achieves the best overall performance, scoring 0.940 versus the next-best 0.879, while using 7.5 times fewer tokens. Our code and benchmark are publicly available at https://github.com/microsoft/GeoFaham.
We address the problem of smooth continuous multi-agent path finding (SC-MAPF), where agents move in a continuous environment along smooth curves in continuous time. The paths for individual agents are searched in local visibility roadmaps that consider obstacles and other agents. A new algorithm is proposed in this work, which combines the previous algorithm CE-CBS for high-level search with any-angle path finding for local single-agent planning, where agents are to avoid obstacles, both static and other agents. The new algorithm is tested and compared to CE-CBS, the algorithm building individual paths using RRT$^*$, and SMT-CCBS algorithm, an algorithm for MAPF$_R$. Experimental results show various advantages of the new algorithm over CE-CBS and SMT-CCBS in comparable settings.
Large language model (LLM)-based multi-agent systems (MAS) can solve complex problems beyond the reach of a single agent by decomposing tasks and coordinating complementary capabilities. Despite extensive research on MAS, existing work largely focuses on specific aspects or individual phases of their development and operation, resulting in fragmented perspectives with objectives and signals that are difficult to reconcile across these different settings. To bridge these disconnected perspectives, we conceptualize MAS development and operation as a three-phase lifecycle spanning design-time orchestration before execution, runtime intervention during execution, and post-hoc attribution after execution, and introduce MAS-Gate, a unified node-centric quantification applicable across these phases. Modeling each execution as a directed acyclic graph, MAS-Gate combines within-node utility, which estimates a node's contribution through changes in reference-answer likelihood, with inter-node strength, which approximates semantic uptake and potential downstream reach. We adapt this shared quantification to three objectives: providing a structure-aware dense reward for Orchestrator training, training an Editor to act on execution prefixes using paired rewards from completed edited and unedited trajectories, and supplying risk-ranked nodes as auxiliary evidence for LLM-based post-hoc attribution. Experiments spanning interaction environments, knowledge-intensive multi-hop question answering, general reasoning, and post-hoc attribution demonstrate that MAS-Gate improves workflow construction, targeted runtime editing, and LLM-assisted error localization over corresponding phase-specific baselines. Together, these results establish MAS-Gate as a unified structural interface for improving MAS generation, dynamic editing, and error attribution throughout the lifecycle.
Training-time multi-agent system (MAS) design seeks to learn reusable orchestration policies, yet prevailing approaches remain largely one-shot: an Orchestrator generates a system but only observes and learns from its terminal outcome. This paradigm discards execution trajectories that expose system-level failures and contain fine-grained information beyond the terminal outcome, leaving no direct basis for revising the source MAS, evaluating the updated MAS, or learning from the resulting contrast. To this end, we introduce MAS-Loop, an execution-grounded framework that closes this feedback loop by coupling an Orchestrator with a trajectory-conditioned Refiner. For each task, the Orchestrator constructs and executes a source MAS to obtain an outcome score and an intermediate trajectory. Conditioned on the trajectory, the Refiner localizes an actionable failure, formulates a trajectory-grounded diagnosis, and proposes a bounded patch to the agents, information flow, or execution configuration. The executable updated MAS is re-executed on the same task to measure its realized outcome under stochastic execution. This pair-MAS comparison provides complementary learning signals: the Orchestrator learns to generate strong systems with fewer recoverable failures, whereas the Refiner learns to expose and repair observed weaknesses, thereby driving policy-level co-evolution across training iterations. Experiments show that MAS-Loop achieves the strongest overall performance among the evaluated baselines, with consistent gains from refinement across all evaluated benchmarks. The co-evolution process also yields MAS-JudgeCorpus, a dataset of execution-grounded source-updated pairs with orchestration- and trajectory-level supervision, which can provide the community with support for MAS reward modeling and trajectory-aware investigation.
Jianhong Wang, Yang Li, Samuel Kaski, Jonathan Lawry
Open multi-agent systems, where only a subset of agents are controllable, are central to real-world domains such as smart grids, yet lack a principled foundation for reinforcement learning algorithm design. We study this setting through n-agent ad hoc teamwork (NAHT), a recent framework for learning in open multi-agent environments. We propose an axiomatic framework that derives value function learning from the Shapley axioms (Efficiency, Symmetry and Linearity), treating them as structural constraints rather than heuristic design choices. This formulation builds on a novel NAHT decomposition, which characterises the fundamental difference between open and fully controllable multi-agent systems, and establishes a cooperative game-theoretic foundation for learning. We further show that enforcing these axioms on learning individual value functions recovers the Shapley value for Dec-POMDPs, providing a principled bridge between reinforcement learning and classical cooperative game theory. Building on this framework, we introduce Shapley Machines and Banzhaf Machines, two generic algorithmic templates that can be instantiated with standard reinforcement learning methods. When applied to strong base algorithms such as IPPO and POAM, our methods consistently improve learning efficiency and generalisation to unseen agent types and conventions across NAHT benchmark tasks. The code of experiments is published on https://github.com/hsvgbkhgbv/axiomatic-rl-naht.
We study cooperative multi-agent reinforcement learning with decoupled dynamics and a monotone submodular reward function over joint trajectories. Due to the non-additive reward, an optimal policy may depend on the history of all agents. This dependence makes exact optimization exponential in the horizon and the number of agents. To overcome this challenge while retaining theoretical guarantees, we introduce pre-window curvature $c_m$, which quantifies how states visited more than $m$ steps earlier reduce current marginal gains. Using this curvature to control the error from restricting the policy of each agent to depend only on its last $m+1$ states, we propose a greedy algorithm that exploits the decoupled dynamics to optimize local policies sequentially. For fixed $m$, the algorithm achieves a $\frac{1-c_m}{2-c_m}$ approximation guarantee up to estimation error, and its computation time is polynomial in the number of agents and the horizon. We recover the known $1/2$-factor approximation for time-additive rewards and obtain an approximation factor no smaller than the known total-curvature bound for a single agent. We formulate multi-agent traffic monitoring as an example of this framework and evaluate our algorithm on it.
Maxime Toquebiau, Jérôme Botoko Ekila, Paul Van Eecke, Katrien Beuls
Recent studies of language evolution with neural agents often use models where agents can either speak or listen, where there are only two agents in the population, or where agents can share internal representations. These choices prevent the study of how language emerges through cultural evolution, i.e. through the cumulative adaptation of individual behaviours to fit the communicative needs of the community. We tackle this issue with a new model that trains populations of bi-directional neural agents in a fully distributed fashion to solve a referential game. We demonstrate the effectiveness of this approach on multiple image datasets and show that it is able to scale to large populations. We also show that more conventional languages are easier to learn, providing a direct advantage for cultural transmission. Overall, this work provides a novel approach to neural emergent communication that opens up new directions for evolutionary linguistics.
Science is a process of useful models being proven wrong time and again, but despite its many flaws, it somehow works. Why is this? Paul Feyerabend argued that it works because people dare to hold and defend incompatible rival theories against the accepted one, since a theory's failures often lead to its abandonment in the presence of a more accurate or simpler alternative. He called this view theoretical pluralism. We introduce a theoretical and experimental model in which this can be examined. It compares communities of pluralist agents, which can hold rival theories side by side and maintain beliefs within and between those theories, with communities of monist agents, which can only incorporate what they hear or observe into a single model of their own. In a multi-agent variational inference setting, our experimental demonstrations illustrate how communities with rival theories can go through periods of uncertainty followed by a shift in the theories they hold true, how the information that is communicated between communities affects beliefs when those communities make different observations about the world, and how sharing support for theories and sharing beliefs within them contribute differently to collective belief dynamics.
Chandler Smith, Qi Guo, Cecilia Elena Tilli, Sophia Hatz, David Demitri Africa, Patricia Paskov, Christian Schroeder de Witt, Philip Torr 等 9 位作者
The deployment of AI agents in multi-agent settings enables inter-agent influence, whereby agents can strategically steer each other's behaviour towards an intended outcome. Such capabilities pose novel risks, as malicious agents could direct others towards harmful actions. In this work we investigate three such capabilities - persuasion, deception, and coercion - across five realistic evaluation environments and a variety of agents instantiated with frontier language models. We observed significant inter-agent influence capabilities; in oversight environments, all tested models shifted overseer decisions from rejection to approval through persuasion, deception, or coercion. We also observed successful influence in non-hierarchical 'peer-to-peer' settings through redirection of a peer's safety research trajectory. As multi-agent deployments become increasingly commonplace, these results point to a pressing need for work on mitigating the risks from inter-agent influence. Code and data are available at https://github.com/Cooperative-AI-Foundation/inter-agent-influence.
Marco Virgolin, Thomas Haszprunar, Maximilian Regenold
Large language model (LLM) assistants are increasingly proposed for regulatory affairs, yet their reliability in regulatory tasks remains unclear. We introduce the EU AI Act Q&A Benchmark, an expert-curated benchmark of 110 open-ended questions about the EU AI Act (Regulation (EU) 2024/1689). Each question comes with an exemplary answer, question-specific correctness criteria, and expected references. Evaluation measures answer correctness, reference quality, conciseness, tone, and latency, with a multi-turn ``hard mode'' testing responses to simulated user pushback. We also present the results of the 2026 EU AI Act Q&A Challenge, based on the benchmark, where three off-the-shelf baselines and five purpose-built assistants (from four commercial developers and one academic entrant) were tested. Only two of the five submitted assistants consistently outperformed the baselines; one achieved high scores across all evaluated metrics. These results show that specialist development alone does not guarantee better performance and highlight the critical importance of LLM assistant evaluation in regulatory affairs. The benchmark code is openly available at https://github.com/regenold-GmbH/eu-ai-act-benchmark-2026/.
Suifeng Zhao, Zida Liu, Lei Sun, Xun Wang, Daojian Zeng, Jun Gao, Sujian Li
Multimodal large language models (MLLMs) increasingly address knowledge-intensive tasks by integrating evidence scattered across multiple document pages into a final answer. Because such answers are synthesized from many pages, the link between the answer and its supporting visual evidence can become obscured, hindering verification and trust. Cross-page visual attribution is therefore essential: it requires identifying the supporting pages and localizing fine-grained evidence regions within them, thereby making multi-page reasoning transparent and auditable. Existing approaches either process multiple pages jointly or iteratively refine evidence within a single trajectory, risking cross-page context pollution and, for interactive methods, substantial sequential overhead. We introduce \textbf{AttriSwarm}, an agent-swarm framework that assigns these concerns to separate roles: a leader handles global cross-page reasoning, while subagents perform page-wise attribution in isolation and in parallel. To support training and evaluation, we construct \textbf{AttriDoc}, a cross-page QA dataset linking supporting pages to their annotated evidence regions, and use it for role-specific training of AttriSwarm, improving answer accuracy, page-level $F_1$, and region-level IoU over its training-free counterpart by an average of 18.94, 29.34, and 23.04 percentage points, respectively, across three open-source backbones and three datasets.