Game-theoretic agents must make plans that optimally gather information about their opponents. These problems are modeled by partially observable stochastic games (POSGs), but planning in fully continuous POSGs is intractable without heavy offline computation or assumptions on the order of belief maintained by each player. We formulate a finite history/horizon refinement of POSGs which admits competitive information gathering behavior in trajectory space, and through a series of approximations, we present an online method for computing rational trajectory plans in these games which leverages particle-based estimations of the joint state space and performs stochastic gradient play. We also provide the necessary adjustments required to deploy this method on individual agents. The method is tested in continuous pursuit-evasion and warehouse-pickup scenarios (alongside extensions to $N > 2$ players and to more complex environments with visual and physical obstacles), demonstrating evidence of active information gathering and outperforming passive competitors.

Large language models are quickly becoming the foundation for intelligent agents that are capable of using tools. However, training such agents is challenging because it requires human creation and annotation of a diverse set of tasks, tools, and evaluation criteria. In this paper, we propose the Self-Challenging framework for training an agent on high-quality tasks that are generated by itself. The agent first plays the role of challenger and generates a task after interacting with the given tools. The tasks take the form of a novel general class of problems termed Code-as-Task, which are defined by an instruction, a verification function and solution and failure cases which serve as tests, allowing to filter only for high-quality tasks. The agent then takes an executor role and trains on those tasks with reinforcement learning using the evaluation feedback as a reward. Evaluation on two existing multi-turn tool-use agent benchmarks, M3ToolEval and TauBench, shows the Self-Challenging framework achieves over a two-fold improvement in Llama-3.1-8B-Instruct, despite using only self-generated training data.

The terms Agentic AI and Multiagentic AI have recently gained popularity in discussions on generative artificial intelligence, often used to describe autonomous software agents and systems composed of such agents. However, the use of these terms confuses these buzzwords with well-established concepts in AI literature: intelligent agents and multi-agent systems. This article offers a critical analysis of this conceptual misuse. We review the theoretical origins of "agentic" in the social sciences (Bandura, 1986) and philosophical notions of intentionality (Dennett, 1971), and then summarise foundational works on intelligent agents and multi-agent systems by Wooldridge, Jennings and others. We examine classic agent architectures, from simple reactive agents to Belief-Desire-Intention (BDI) models, and highlight key properties (autonomy, reactivity, proactivity, social capability) that define agency in AI. We then discuss recent developments in large language models (LLMs) and agent platforms based on LLMs, including the emergence of LLM-powered AI agents and open-source multi-agent orchestration frameworks. We argue that the term AI Agentic is often used as a buzzword for what are essentially AI agents, and AI Multiagentic for what are multi-agent systems. This confusion overlooks decades of research in the field of autonomous agents and multi-agent systems. The article advocates for scientific and technological rigour and the use of established terminology from the state of the art in AI, incorporating the wealth of existing knowledge, including standards for multi-agent system platforms, communication languages and coordination and cooperation algorithms, agreement technologies (automated negotiation, argumentation, virtual organisations, trust, reputation, etc.), into the new and promising wave of LLM-based AI agents, so as not to end up reinventing the wheel.

Enterprise Resource Planning (ERP) systems serve as the digital backbone of modern financial institutions, yet they continue to rely on static, rule-based workflows that limit adaptability, scalability, and intelligence. As business operations grow more complex and data-rich, conventional ERP platforms struggle to integrate structured and unstructured data in real time and to accommodate dynamic, cross-functional workflows. In this paper, we present the first AI-native, agent-based framework for ERP systems, introducing a novel architecture of Generative Business Process AI Agents (GBPAs) that bring autonomy, reasoning, and dynamic optimization to enterprise workflows. The proposed system integrates generative AI with business process modeling and multi-agent orchestration, enabling end-to-end automation of complex tasks such as budget planning, financial reporting, and wire transfer processing. Unlike traditional workflow engines, GBPAs interpret user intent, synthesize workflows in real time, and coordinate specialized sub-agents for modular task execution. We validate the framework through case studies in bank wire transfers and employee reimbursements, two representative financial workflows with distinct complexity and data modalities. Results show that GBPAs achieve up to 40% reduction in processing time, 94% drop in error rate, and improved regulatory compliance by enabling parallelism, risk control insertion, and semantic reasoning. These findings highlight the potential of GBPAs to bridge the gap between generative AI capabilities and enterprise-grade automation, laying the groundwork for the next generation of intelligent ERP systems.

Recent advances in Large Language Models (LLMs) have enabled multi-agent systems that simulate real-world interactions with near-human reasoning. While previous studies have extensively examined biases related to protected attributes such as race, the emergence and propagation of biases on socially contentious issues in multi-agent LLM interactions remain underexplored. This study explores how LLM agents shape public opinion through debates on five contentious topics. By simulating over 2,500 debates, we analyze how initially neutral agents, assigned a centrist disposition, adopt specific stances over time. Statistical analyses reveal significant group conformity mirroring human behavior; LLM agents tend to align with numerically dominant groups or more intelligent agents, exerting a greater influence. These findings underscore the crucial role of agent intelligence in shaping discourse and highlight the risks of bias amplification in online interactions. Our results emphasize the need for policy measures that promote diversity and transparency in LLM-generated discussions to mitigate the risks of bias propagation within anonymous online environments.

Since the advent of AI, games have served as progress benchmarks. Meanwhile, imperfect-information variants of chess have existed for over a century, present extreme challenges, and have been the focus of significant AI research. Beyond calculation needed in regular chess, they require reasoning about information gathering, the opponent's knowledge, signaling, etc. The most popular variant, Fog of War (FoW) chess (aka. dark chess) is a recognized challenge problem in AI after superhuman performance was reached in no-limit Texas hold'em poker. We present Obscuro, the first superhuman AI for FoW chess. It introduces advances to search in imperfect-information games, enabling strong, scalable reasoning. Experiments against the prior state-of-the-art AI and human players -- including the world's best -- show that Obscuro is significantly stronger. FoW chess is the largest (by amount of imperfect information) turn-based game in which superhuman performance has been achieved and the largest game in which imperfect-information search has been successfully applied.

Structured scene representations are a core component of embodied agents, helping to consolidate raw sensory streams into readable, modular, and searchable formats. Due to their high computational overhead, many approaches build such representations in advance of the task. However, when the task specifications change, such static approaches become inadequate as they may miss key objects, spatial relations, and details. We introduce GraphPad, a modifiable structured memory that an agent can tailor to the needs of the task through API calls. It comprises a mutable scene graph representing the environment, a navigation log indexing frame-by-frame content, and a scratchpad for task-specific notes. Together, GraphPad serves as a dynamic workspace that remains complete, current, and aligned with the agent's immediate understanding of the scene and its task. On the OpenEQA benchmark, GraphPad attains 55.3%, a +3.0% increase over an image-only baseline using the same vision-language model, while operating with five times fewer input frames. These results show that allowing online, language-driven refinement of 3-D memory yields more informative representations without extra training or data collection.

Despite the remarkable generative capabilities of diffusion models, their integration into safety-critical or scientifically rigorous applications remains hindered by the need to ensure compliance with stringent physical, structural, and operational constraints. To address this challenge, this paper introduces Neuro-Symbolic Diffusion (NSD), a novel framework that interleaves diffusion steps with symbolic optimization, enabling the generation of certifiably consistent samples under user-defined functional and logic constraints. This key feature is provided for both standard and discrete diffusion models, enabling, for the first time, the generation of both continuous (e.g., images and trajectories) and discrete (e.g., molecular structures and natural language) outputs that comply with constraints. This ability is demonstrated on tasks spanning three key challenges: (1) Safety, in the context of non-toxic molecular generation and collision-free trajectory optimization; (2) Data scarcity, in domains such as drug discovery and materials engineering; and (3) Out-of-domain generalization, where enforcing symbolic constraints allows adaptation beyond the training distribution.

Sustaining coherent, role-aware communication across multi-agent systems remains a foundational challenge in AI. Current frameworks often lack explicit mechanisms for speaker responsibility, leading to context drift, alignment instability, and degraded interpretability over time. We propose the Modular Speaker Architecture (MSA), a framework that decomposes speaker behavior into modular components for role tracking, responsibility continuity, and contextual coherence. Grounded in high-context human-AI dialogues, MSA includes three core modules: a Speaker Role Module, a Responsibility Chain Tracker, and a Contextual Integrity Validator. We evaluate MSA through annotated case studies and introduce structural metrics-pragmatic consistency, responsibility flow, and context stability-quantified via manual and automatic scoring and bootstrapped statistical analysis. Our results show that MSA reliably maintains interaction structure without reliance on affective signals or surface-level heuristics. We further implement a prototype configuration language (G-Code) and modular API to support MSA deployment in dynamic multi-agent scenarios.

Vision-language models (VLMs) aligned with general human objectives, such as being harmless and hallucination-free, have become valuable assistants of humans in managing visual tasks. However, people with diversified backgrounds have different cognition even in the same situation. Consequently, they may have personalized expectations for VLM assistants. This highlights the urgent need to align VLM assistants with personalized situated cognition for real-world assistance. To study this problem, we first simplify it by characterizing individuals based on the sociological concept of Role-Set. Then, we propose to evaluate the individuals' actions to examine whether the personalized alignment is achieved. Further, we construct a benchmark named PCogAlignBench, which includes 18k instances and 20 individuals with different Role-Sets. Finally, we present a framework called PCogAlign, which constructs a cognition-aware and action-based reward model for personalized alignment. Experimental results and human evaluations demonstrate the reliability of the PCogAlignBench and the effectiveness of our proposed PCogAlign. We will open-source the constructed benchmark and code at https://github.com/NLPGM/PCogAlign.