Safe reinforcement learning (RL) seeks to mitigate unsafe behaviors that arise from exploration during training by reducing constraint violations while maintaining task performance. Existing approaches typically rely on a single policy to jointly optimize reward and safety, which can cause instability due to conflicting objectives, or they use external safety filters that override actions and require prior system knowledge. In this paper, we propose a modular cost-aware regulator that scales the agent's actions based on predicted constraint violations, preserving exploration through smooth action modulation rather than overriding the policy. The regulator is trained to minimize constraint violations while avoiding degenerate suppression of actions. Our approach integrates seamlessly with off-policy RL methods such as SAC and TD3, and achieves state-of-the-art return-to-cost ratios on Safety Gym locomotion tasks with sparse costs, reducing constraint violations by up to 126 times while increasing returns by over an order of magnitude compared to prior methods.

Social deduction games like Werewolf combine language, reasoning, and strategy, providing a testbed for studying natural language and social intelligence. However, most studies reduce the game to LLM-based self-play, yielding templated utterances and anecdotal cases that overlook the richness of social gameplay. Evaluation further relies on coarse metrics such as survival time or subjective scoring due to the lack of quality reference data. To address these gaps, we curate a high-quality, human-verified multimodal Werewolf dataset containing over 100 hours of video, 32.4M utterance tokens, and 15 rule variants. Based on this dataset, we propose a novel strategy-alignment evaluation that leverages the winning faction's strategies as ground truth in two stages: 1) Speech evaluation, formulated as multiple-choice-style tasks that assess whether the model can adopt appropriate stances across five dimensions of social ability; and 2) Decision evaluation, which assesses the model's voting choices and opponent-role inferences. This framework enables a fine-grained evaluation of models' linguistic and reasoning capabilities, while capturing their ability to generate strategically coherent gameplay. Our experiments show that state-of-the-art LLMs show diverse performance, with roughly half remain below 0.50, revealing clear gaps in deception and counterfactual reasoning. We hope our dataset further inspires research on language, reasoning, and strategy in multi-agent interaction.

Large language models (LLMs) based Agents are increasingly pivotal in simulating and understanding complex human systems and interactions. We propose the AI-Agent School (AAS) system, built around a self-evolving mechanism that leverages agents for simulating complex educational dynamics. Addressing the fragmented issues in teaching process modeling and the limitations of agents performance in simulating diverse educational participants, AAS constructs the Zero-Exp strategy, employs a continuous "experience-reflection-optimization" cycle, grounded in a dual memory base comprising experience and knowledge bases and incorporating short-term and long-term memory components. Through this mechanism, agents autonomously evolve via situated interactions within diverse simulated school scenarios. This evolution enables agents to more accurately model the nuanced, multi-faceted teacher-student engagements and underlying learning processes found in physical schools. Experiment confirms that AAS can effectively simulate intricate educational dynamics and is effective in fostering advanced agent cognitive abilities, providing a foundational stepping stone from the "Era of Experience" to the "Era of Simulation" by generating high-fidelity behavioral and interaction data.

Learning to autonomously execute long-horizon procedures from natural language remains a core challenge for intelligent agents. Free-form instructions such as recipes, scientific protocols, or business workflows encode rich procedural knowledge, but their variability and lack of structure cause agents driven by large language models (LLMs) to drift or fail during execution. We introduce Procedure Aware DynaMic Execution (PADME), an agent framework that produces and exploits a graph-based representation of procedures. Unlike prior work that relies on manual graph construction or unstructured reasoning, PADME autonomously transforms procedural text into executable graphs that capture task dependencies, decision points, and reusable subroutines. Central to PADME is a two-phase methodology; Teach phase, which focuses on systematic structuring, enrichment with executable logic of procedures, followed by Execute phase, which enables dynamic execution in response to real-time inputs and environment feedback. This separation ensures quality assurance and scalability, allowing expert knowledge to be encoded once and reliably reused across varying contexts. The graph representation also provides an inductive bias that reduces error accumulation in long-horizon reasoning, underscoring the importance of structured procedure modeling for reliable agent-driven automation. Empirically, PADME achieves state-of-the-art performance on four diverse benchmarks, including ALFWorld and ScienceWorld. These results demonstrate that agents equipped with graph-based procedure representations offer a powerful intermediate abstraction for robust and generalizable execution.

When AI agents retrieve and reason over external documents, adversaries can manipulate the data they receive to subvert their behaviour. Previous research has studied indirect prompt injection, where the attacker injects malicious instructions. We argue that injection of instructions is not necessary to manipulate agents - attackers could instead supply biased, misleading, or false information. We term this an attack by content. Existing defenses, which focus on detecting hidden commands, are ineffective against attacks by content. To defend themselves and their users, agents must critically evaluate retrieved information, corroborating claims with external evidence and evaluating source trustworthiness. We argue that this is analogous to an existing NLP task, automated fact-checking, which we propose to repurpose as a cognitive self-defense tool for agents.

An agent facing a planning problem can use answers to how-to questions to reduce uncertainty and fill knowledge gaps, helping it solve both current and future tasks. However, their open ended nature, where valid answers to "How do I X?" range from executable actions to high-level descriptions of X's sub-goals, makes them challenging for AI agents to ask, and for AI experts to answer, in ways that support efficient planning. We introduce $How^{2}$, a memory agent framework that enables agents to ask how-to questions, store the answers, and reuse them for lifelong learning in interactive environments. We evaluate our approach in Plancraft, a Minecraft crafting environment, where agents must complete an assembly task by manipulating inventory items. Using teacher models that answer at varying levels of abstraction, from executable action sequences to high-level subgoal descriptions, we show that lifelong learning agents benefit most from answers that are abstracted and decoupled from the current state. $How^{2}$ offers a way for LLM-based agents to improve their planning capabilities over time by asking questions in interactive environments.

The rapid expansion of scientific data has widened the gap between analytical capability and research intent. Existing AI-based analysis tools, ranging from AutoML frameworks to agentic research assistants, either favor automation over transparency or depend on manual scripting that hinders scalability and reproducibility. We present ARIA (Automated Research Intelligence Assistant), a spec-driven, human-in-the-loop framework for automated and interpretable data analysis. ARIA integrates six interoperable layers, namely Command, Context, Code, Data, Orchestration, and AI Module, within a document-centric workflow that unifies human reasoning and machine execution. Through natural-language specifications, researchers define analytical goals while ARIA autonomously generates executable code, validates computations, and produces transparent documentation. Beyond achieving high predictive accuracy, ARIA can rapidly identify optimal feature sets and select suitable models, minimizing redundant tuning and repetitive experimentation. In the Boston Housing case, ARIA discovered 25 key features and determined XGBoost as the best performing model (R square = 0.93) with minimal overfitting. Evaluations across heterogeneous domains demonstrate ARIA's strong performance, interpretability, and efficiency compared with state-of-the-art systems. By combining AI for research and AI for science principles within a spec-driven architecture, ARIA establishes a new paradigm for transparent, collaborative, and reproducible scientific discovery.

While multi-agent debate has been proposed as a promising strategy for improving AI reasoning ability, we find that debate can sometimes be harmful rather than helpful. Prior work has primarily focused on debates within homogeneous groups of agents, whereas we explore how diversity in model capabilities influences the dynamics and outcomes of multi-agent interactions. Through a series of experiments, we demonstrate that debate can lead to a decrease in accuracy over time - even in settings where stronger (i.e., more capable) models outnumber their weaker counterparts. Our analysis reveals that models frequently shift from correct to incorrect answers in response to peer reasoning, favoring agreement over challenging flawed reasoning. We perform additional experiments investigating various potential contributing factors to these harmful shifts - including sycophancy, social conformity, and model and task type. These results highlight important failure modes in the exchange of reasons during multi-agent debate, suggesting that naive applications of debate may cause performance degradation when agents are neither incentivised nor adequately equipped to resist persuasive but incorrect reasoning.

We address the long-horizon gap in large language model (LLM) agents by enabling them to sustain coherent strategies in adversarial, stochastic environments. Settlers of Catan provides a challenging benchmark: success depends on balancing short- and long-term goals amid randomness, trading, expansion, and blocking. Prompt-centric LLM agents (e.g., ReAct, Reflexion) must re-interpret large, evolving game states each turn, quickly saturating context windows and losing strategic consistency. We propose HexMachina, a continual learning multi-agent system that separates environment discovery (inducing an adapter layer without documentation) from strategy improvement (evolving a compiled player through code refinement and simulation). This design preserves executable artifacts, allowing the LLM to focus on high-level strategy rather than per-turn reasoning. In controlled Catanatron experiments, HexMachina learns from scratch and evolves players that outperform the strongest human-crafted baseline (AlphaBeta), achieving a 54% win rate and surpassing prompt-driven and no-discovery baselines. Ablations confirm that isolating pure strategy learning improves performance. Overall, artifact-centric continual learning transforms LLMs from brittle stepwise deciders into stable strategy designers, advancing long-horizon autonomy.

Classical federated learning (FL) assumes that the clients have a limited amount of noisy data with which they voluntarily participate and contribute towards learning a global, more accurate model in a principled manner. The learning happens in a distributed fashion without sharing the data with the center. However, these methods do not consider the incentive of an agent for participating and contributing to the process, given that data collection and running a distributed algorithm is costly for the clients. The question of rationality of contribution has been asked recently in the literature and some results exist that consider this problem. This paper addresses the question of simultaneous parameter learning and incentivizing contribution in a truthful manner, which distinguishes it from the extant literature. Our first mechanism incentivizes each client to contribute to the FL process at a Nash equilibrium and simultaneously learn the model parameters. We also ensure that agents are incentivized to truthfully reveal information in the intermediate stages of the algorithm. However, this equilibrium outcome can be away from the optimal, where clients contribute with their full data and the algorithm learns the optimal parameters. We propose a second mechanism that enables the full data contribution along with optimal parameter learning. Large scale experiments with real (federated) datasets (CIFAR-10, FEMNIST, and Twitter) show that these algorithms converge quite fast in practice, yield good welfare guarantees and better model performance for all agents.