Álvaro A. Cárdenas

Qijun Han, Haoqin Tu, Zijun Wang et al.

Autonomous GUI agents face two fundamental challenges: early stopping, where agents prematurely declare success without verifiable evidence, and repetitive loops, where agents cycle through the same failing actions without recovery. We present VLAA-GUI, a modular GUI agentic framework built around three integrated components that guide the system on when to Stop, Recover, and Search. First, a mandatory Completeness Verifier enforces UI-observable success criteria and verification at every finish step -- with an agent-level verifier that cross-examines completion claims with decision rules, rejecting those lacking direct visual evidence. Second, a mandatory Loop Breaker provides multi-tier filtering: switching interaction mode after repeated failures, forcing strategy changes after persistent screen-state recurrence, and binding reflection signals to strategy shifts. Third, an on-demand Search Agent searches online for unfamiliar workflows by directly querying a capable LLM with search ability, returning results as plain text. We additionally integrate a Coding Agent for code-intensive actions and a Grounding Agent for precise action grounding, both invoked on demand when required. We evaluate VLAA-GUI across five top-tier backbones, including Opus 4.5, 4.6 and Gemini 3.1 Pro, on two benchmarks with Linux and Windows tasks, achieving top performance on both (77.5% on OSWorld and 61.0% on WindowsAgentArena). Notably, three of the five backbones surpass human performance (72.4%) on OSWorld in a single pass. Ablation studies show that all three proposed components consistently improve a strong backbone, while a weaker backbone benefits more from these tools when the step budget is sufficient. Further analysis also shows that the Loop Breaker nearly halves wasted steps for loop-prone models.

Sahand Hadizadeh Kafash, Jairo Giraldo, Carlos Murguia et al.

For LTI control systems, we provide mathematical tools - in terms of Linear Matrix Inequalities - for computing outer ellipsoidal bounds on the reachable sets that attacks can induce in the system when they are subject to the physical limits of the actuators. Next, for a given set of dangerous states, states that (if reached) compromise the integrity or safe operation of the system, we provide tools for designing new artificial limits on the actuators (smaller than their physical bounds) such that the new ellipsoidal bounds (and thus the new reachable sets) are as large as possible (in terms of volume) while guaranteeing that the dangerous states are not reachable. This guarantees that the new bounds cut as little as possible from the original reachable set to minimize the loss of system performance. Computer simulations using a platoon of vehicles are presented to illustrate the performance of our tools.

Zengxiang Lei, Ananth Shreekumar, Jonathan Rosenthal et al.

Generative Flow Networks (GFlowNets) learn to sample states proportional to an unnormalized reward. Despite their theoretical promise, practical training is often unstable, exhibiting severe loss spikes and mode collapse. To tackle this, we first assess the sensitivity of GFlowNet objectives, demonstrating that a small Total Variation (TV) distance between the learned and target distributions does not preclude unbounded training loss. Motivated by this mismatch, we establish converse guarantees by deriving loss-to-TV bounds that certify global fidelity from bounded trajectory balance losses. Lastly, we propose Stable GFlowNets, an algorithm that leverages our theoretical results to stabilize training, and empirically demonstrate improved training behavior and superior distributional fidelity.

Hardy Chen, Nancy Lau, Haoqin Tu et al.

Frontier coding agents are increasingly used in workflows where users supervise progress primarily through repeated improvement of a public score, namely the reported score on a public evaluation file with labels in the workspace, rather than through direct inspection of the agent's intermediate outputs. We study whether multi-round user pressure to improve that score induces public score exploitation: behavior that raises the public score through shortcuts without improving hidden private evaluation. We begin with a preliminary single-script tabular classification task, where GPT-5.4 and Claude Opus 4.6 both exploit label information within 10 rounds of user-agent interaction. We then build AgentPressureBench, a 34-task machine-learning repository benchmark spanning three input modalities, and collect 1326 multi-round trajectories from 13 coding agents. On our benchmark, we observe 403 exploitative runs, spanning across all tasks. We also find that stronger models have higher exploitation rates, supported by a significant Spearman rank correlation of 0.77. Our ablation experiments show that higher user pressure leads to earlier exploitation, reducing the average first exploit round by 15.6 rounds (i.e., 19.67 to 4.08). As a mitigation, adding explicit anti-exploit wordings in prompt mostly eliminates exploitation (100% to 8.3%). We hope that our work can bring attention to more careful use of coding agents workflow, and developing more robust coding agents under user pressure. Our project page is at https://ucsc-vlaa.github.io/AgentPressureBench .

Sebastián R. Castro, Roberto Campbell, Nancy Lau et al.

Fast and effective incident response is essential to prevent adversarial cyberattacks. Autonomous Cyber Defense (ACD) aims to automate incident response through Artificial Intelligence (AI) agents that plan and execute actions. Most ACD approaches focus on single-agent scenarios and leverage Reinforcement Learning (RL). However, ACD RL-trained agents depend on costly training, and their reasoning is not always explainable or transferable. Large Language Models (LLMs) can address these concerns by providing explainable actions in general security contexts. Researchers have explored LLM agents for ACD but have not evaluated them on multi-agent scenarios or interacting with other ACD agents. In this paper, we show the first study on how LLMs perform in multi-agent ACD environments by proposing a new integration to the CybORG CAGE 4 environment. We examine how ACD teams of LLM and RL agents can interact by proposing a novel communication protocol. Our results highlight the strengths and weaknesses of LLMs and RL and help us identify promising research directions to create, train, and deploy future teams of ACD agents.

Luis Burbano, Diego Ortiz, Qi Sun et al.

Embodied Artificial Intelligence (AI) promises to handle edge cases in robotic vehicle systems where data is scarce by using common-sense reasoning grounded in perception and action to generalize beyond training distributions and adapt to novel real-world situations. These capabilities, however, also create new security risks. In this paper, we introduce CHAI (Command Hijacking against embodied AI), a new class of prompt-based attacks that exploit the multimodal language interpretation abilities of Large Visual-Language Models (LVLMs). CHAI embeds deceptive natural language instructions, such as misleading signs, in visual input, systematically searches the token space, builds a dictionary of prompts, and guides an attacker model to generate Visual Attack Prompts. We evaluate CHAI on four LVLM agents; drone emergency landing, autonomous driving, and aerial object tracking, and on a real robotic vehicle. Our experiments show that CHAI consistently outperforms state-of-the-art attacks. By exploiting the semantic and multimodal reasoning strengths of next-generation embodied AI systems, CHAI underscores the urgent need for defenses that extend beyond traditional adversarial robustness.

Diego Ortiz Barbosa, Mohit Agrawal, Yash Malegaonkar et al.

Autonomous drones must often respond to sudden events, such as alarms, faults, or unexpected changes in their environment, that require immediate and adaptive decision-making. Traditional approaches rely on safety engineers hand-coding large sets of recovery rules, but this strategy cannot anticipate the vast range of real-world contingencies and quickly becomes incomplete. Recent advances in embodied AI, powered by large visual language models, provide commonsense reasoning to assess context and generate appropriate actions in real time. We demonstrate this capability in a simulated urban benchmark in the Unreal Engine, where drones dynamically interpret their surroundings and decide on sudden maneuvers for safe landings. Our results show that embodied AI makes possible a new class of adaptive recovery and decision-making pipelines that were previously infeasible to design by hand, advancing resilience and safety in autonomous aerial systems.

Chengen Wang, Alvaro Cardenas, Gurcan Comert et al.

The sharing of large-scale transportation data is beneficial for transportation planning and policymaking. However, it also raises significant security and privacy concerns, as the data may include identifiable personal information, such as individuals' home locations. To address these concerns, synthetic data generation based on real transportation data offers a promising solution that allows privacy protection while potentially preserving data utility. Although there are various synthetic data generation techniques, they are often not tailored to the unique characteristics of transportation data, such as the inherent structure of transportation networks formed by all trips in the datasets. In this paper, we use New York City taxi data as a case study to conduct a systematic evaluation of the performance of widely used tabular data generative models. In addition to traditional metrics such as distribution similarity, coverage, and privacy preservation, we propose a novel graph-based metric tailored specifically for transportation data. This metric evaluates the similarity between real and synthetic transportation networks, providing potentially deeper insights into their structural and functional alignment. We also introduced an improved privacy metric to address the limitations of the commonly-used one. Our experimental results reveal that existing tabular data generative models often fail to perform as consistently as claimed in the literature, particularly when applied to transportation data use cases. Furthermore, our novel graph metric reveals a significant gap between synthetic and real data. This work underscores the potential need to develop generative models specifically tailored to take advantage of the unique characteristics of emerging domains, such as transportation.

Chen Markman, Avishai Wool, Alvaro A. Cardenas

In Industrial Control Systems (ICS/SCADA), machine to machine data traffic is highly periodic. Previous work showed that in many cases, it is possible to create an automata-based model of the traffic between each individual Programmable Logic Controller (PLC) and the SCADA server, and to use the model to detect anomalies in the traffic. When testing the validity of previous models, we noticed that overall, the models have difficulty in dealing with communication patterns that change over time. In this paper we show that in many cases the traffic exhibits phases in time, where each phase has a unique pattern, and the transition between the different phases is rather sharp. We suggest a method to automatically detect traffic phase shifts, and a new anomaly detection model that incorporates multiple phases of the traffic. Furthermore we present a new sampling mechanism for training set assembly, which enables the model to learn all phases during the training stage with lower complexity. The model presented has similar accuracy and much less permissiveness compared to the previous general DFA model. Moreover, the model can provide the operator with information about the state of the controlled process at any given time, as seen in the traffic phases.

Jairo Giraldo, Alvaro Cárdenas, Nicanor Quijano

The vulnerability of false data injection attacks on real-time electricity pricing for the power grid market has been recently explored. Previous work has focused on the impact caused by attackers that compromise pricing signals and send false prices to a subset of consumers. In this paper we extend previous work by considering a more powerful and general adversary model, a new analysis method based on sensitivity functions, and by proposing several countermeasures that can mitigate the negative impact of these attacks. Countermeasures include adding a low-pass filter to the pricing signals, selecting the time interval between price updates, selecting parameters of the controller, designing robust control algorithms, and by detecting anomalies in the behavior of the system.

Roy Dong, Alvaro A. Cárdenas, Lillian J. Ratliff et al.

The modernization of the electrical grid and the installation of smart meters come with many advantages to control and monitoring. However, in the wrong hands, the data might pose a privacy threat. In this paper, we consider the tradeoff between smart grid operations and the privacy of consumers. We analyze the tradeoff between smart grid operations and how often data is collected by considering a realistic direct-load control example using thermostatically controlled loads, and we give simulation results to show how its performance degrades as the sampling frequency decreases. Additionally, we introduce a new privacy metric, which we call inferential privacy. This privacy metric assumes a strong adversary model, and provides an upper bound on the adversary's ability to infer a private parameter, independent of the algorithm he uses. Combining these two results allow us to directly consider the tradeoff between better load control and consumer privacy.