As large language models (LLMs) become integral to multi-agent systems, new privacy risks emerge that extend beyond memorization, direct inference, or single-turn evaluations. In particular, seemingly innocuous responses, when composed across interactions, can cumulatively enable adversaries to recover sensitive information, a phenomenon we term compositional privacy leakage. We present the first systematic study of such compositional privacy leaks and possible mitigation methods in multi-agent LLM systems. First, we develop a framework that models how auxiliary knowledge and agent interactions jointly amplify privacy risks, even when each response is benign in isolation. Next, to mitigate this, we propose and evaluate two defense strategies: (1) Theory-of-Mind defense (ToM), where defender agents infer a questioner's intent by anticipating how their outputs may be exploited by adversaries, and (2) Collaborative Consensus Defense (CoDef), where responder agents collaborate with peers who vote based on a shared aggregated state to restrict sensitive information spread. Crucially, we balance our evaluation across compositions that expose sensitive information and compositions that yield benign inferences. Our experiments quantify how these defense strategies differ in balancing the privacy-utility trade-off. We find that while chain-of-thought alone offers limited protection to leakage (~39% sensitive blocking rate), our ToM defense substantially improves sensitive query blocking (up to 97%) but can reduce benign task success. CoDef achieves the best balance, yielding the highest Balanced Outcome (79.8%), highlighting the benefit of combining explicit reasoning with defender collaboration. Together, our results expose a new class of risks in collaborative LLM deployments and provide actionable insights for designing safeguards against compositional, context-driven privacy leakage.

Vision-Language-Action models (VLAs) mark a major shift in robot learning. They replace specialized architectures and task-tailored components of expert policies with large-scale data collection and setup-specific fine-tuning. In this machine learning-focused workflow that is centered around models and scalable training, traditional robotics software frameworks become a bottleneck, while robot simulations offer only limited support for transitioning from and to real-world experiments. In this work, we close this gap by introducing Robot Control Stack (RCS), a lean ecosystem designed from the ground up to support research in robot learning with large-scale generalist policies. At its core, RCS features a modular and easily extensible layered architecture with a unified interface for simulated and physical robots, facilitating sim-to-real transfer. Despite its minimal footprint and dependencies, it offers a complete feature set, enabling both real-world experiments and large-scale training in simulation. Our contribution is twofold: First, we introduce the architecture of RCS and explain its design principles. Second, we evaluate its usability and performance along the development cycle of VLA and RL policies. Our experiments also provide an extensive evaluation of Octo, OpenVLA, and Pi Zero on multiple robots and shed light on how simulation data can improve real-world policy performance. Our code, datasets, weights, and videos are available at: https://robotcontrolstack.github.io/

Stochastic forecasting is critical for efficient decision-making in uncertain systems, such as energy markets and finance, where estimating the full distribution of future scenarios is essential. We propose Diffusion Scenario Tree (DST), a general framework for constructing scenario trees for multivariate prediction tasks using diffusion-based probabilistic forecasting models. DST recursively samples future trajectories and organizes them into a tree via clustering, ensuring non-anticipativity (decisions depending only on observed history) at each stage. We evaluate the framework on the optimization task of energy arbitrage in New York State's day-ahead electricity market. Experimental results show that our approach consistently outperforms the same optimization algorithms that use scenario trees from more conventional models and Model-Free Reinforcement Learning baselines. Furthermore, using DST for stochastic optimization yields more efficient decision policies, achieving higher performance by better handling uncertainty than deterministic and stochastic MPC variants using the same diffusion-based forecaster.

In mixed-traffic environments, autonomous vehicles must adapt to human-controlled vehicles and other unusual driving situations. This setting can be framed as a multi-agent reinforcement learning (MARL) environment with full cooperative reward among the autonomous vehicles. While methods such as Multi-agent Proximal Policy Optimization can be effective in training MARL tasks, they often fail to resolve local conflict between agents and are unable to generalize to stochastic events. In this paper, we propose a Local State Attention module to assist the input state representation. By relying on the self-attention operator, the module is expected to compress the essential information of nearby agents to resolve the conflict in traffic situations. Utilizing a simulated highway merging scenario with the priority vehicle as the unexpected event, our approach is able to prioritize other vehicles' information to manage the merging process. The results demonstrate significant improvements in merging efficiency compared to popular baselines, especially in high-density traffic settings.

Web agents powered by Large Language Models (LLMs) show promise for next-generation AI, but their limited reasoning in uncertain, dynamic web environments hinders robust deployment. In this paper, we identify key reasoning skills essential for effective web agents, i.e., reflection & lookahead, branching, and rollback, and curate trajectory data that exemplifies these abilities by reconstructing the agent's (inference-time) reasoning algorithms into chain-of-thought rationales. We conduct experiments in the agent self-improving benchmark, OpenWebVoyager, and demonstrate that distilling salient reasoning patterns into the backbone LLM via simple fine-tuning can substantially enhance its performance. Our approach yields significant improvements across multiple benchmarks, including WebVoyager, Mind2web-live, and SimpleQA (web search), highlighting the potential of targeted reasoning skill enhancement for web agents.

Substantial progress has recently been made in the understanding of the cutoff phenomenon for Markov processes, using an information-theoretic statistics known as varentropy [Sal23; Sal24; Sal25a; PS25]. In the present paper, we propose an alternative approach which bypasses the use of varentropy and exploits instead a new W-TV transport inequality, combined with a classical parabolic regularization estimate [BGL01; OV01]. While currently restricted to non-negatively curved processes on smooth spaces, our argument no longer requires the chain rule, nor any approximate version thereof. As applications, we recover the main result of [Sal25a] establishing cutoff for the log-concave Langevin dynamics, and extend the conclusion to a widely-used discrete-time sampling algorithm known as the Proximal Sampler.

This position paper contends that modern AI research must adopt an antifragile perspective on safety -- one in which the system's capacity to guarantee long-term AI safety such as handling rare or out-of-distribution (OOD) events expands over time. Conventional static benchmarks and single-shot robustness tests overlook the reality that environments evolve and that models, if left unchallenged, can drift into maladaptation (e.g., reward hacking, over-optimization, or atrophy of broader capabilities). We argue that an antifragile approach -- Rather than striving to rapidly reduce current uncertainties, the emphasis is on leveraging those uncertainties to better prepare for potentially greater, more unpredictable uncertainties in the future -- is pivotal for the long-term reliability of open-ended ML systems. In this position paper, we first identify key limitations of static testing, including scenario diversity, reward hacking, and over-alignment. We then explore the potential of antifragile solutions to manage rare events. Crucially, we advocate for a fundamental recalibration of the methods used to measure, benchmark, and continually improve AI safety over the long term, complementing existing robustness approaches by providing ethical and practical guidelines towards fostering an antifragile AI safety community.

Vision-language-action (VLA) models finetuned from vision-language models (VLMs) hold the promise of leveraging rich pretrained representations to build generalist robots across diverse tasks and environments. However, direct fine-tuning on robot data often disrupts these representations and limits generalization. We present a framework that better preserves pretrained features while adapting them for robot manipulation. Our approach introduces three components: (i) a dual-encoder design with one frozen vision encoder to retain pretrained features and another trainable for task adaptation, (ii) a string-based action tokenizer that casts continuous actions into character sequences aligned with the model's pretraining domain, and (iii) a co-training strategy that combines robot demonstrations with vision-language datasets emphasizing spatial reasoning and affordances. Evaluations in simulation and on real robots show that our method improves robustness to visual perturbations, generalization to novel instructions and environments, and overall task success compared to baselines.

Collaborative transportation of cable-suspended payloads by teams of Unmanned Aerial Vehicles (UAVs) has the potential to enhance payload capacity, adapt to different payload shapes, and provide built-in compliance, making it attractive for applications ranging from disaster relief to precision logistics. However, multi-UAV coordination under disturbances, nonlinear payload dynamics, and slack--taut cable modes remains a challenging control problem. To our knowledge, no prior work has addressed these cable mode transitions in the multi-UAV context, instead relying on simplifying rigid-link assumptions. We propose CrazyMARL, a decentralized Reinforcement Learning (RL) framework for multi-UAV cable-suspended payload transport. Simulation results demonstrate that the learned policies can outperform classical decentralized controllers in terms of disturbance rejection and tracking precision, achieving an 80% recovery rate from harsh conditions compared to 44% for the baseline method. We also achieve successful zero-shot sim-to-real transfer and demonstrate that our policies are highly robust under harsh conditions, including wind, random external disturbances, and transitions between slack and taut cable dynamics. This work paves the way for autonomous, resilient UAV teams capable of executing complex payload missions in unstructured environments.

V alue iteration (VI) [4] is the practically most used method for reliable analysis of probabilistic systems, in particular Markov decision processes (MDPs) [21] and stochastic games (SGs) [8]. It is used in the state-of-the-art model checkers such as Prism [18] and Storm [11] as the default method due to its better practical scalability, compared to strategy iteration or linear/quadratic programming [14, 19]. The price to pay are issues with precision. Firstly, while other methods yield precise results in theory (omitting floating-point issues), VI converges to the exact result only in the limit. Secondly, the precision of the intermediate iterations was until recently an open question. Given the importance of reliable precision in verification, many recent works focused on modifying VI so that the imprecision can be bounded, yielding a stopping criterion. Consequently, (i) the computed result is reliable, and (ii) the procedure can even terminate earlier whenever the desired precision is achieved.