MOSAIC-Bench: Measuring Compositional Vulnerability Induction in Coding Agents

Coding agents often pass per-prompt safety review yet ship exploitable code when their tasks are decomposed into routine engineering tickets. The challenge is structural: existing safety alignment evaluates overt requests in isolation, leaving models blind to malicious end-states that emerge from sequenced compliance with innocuous-looking requests. We introduce MOSAIC-Bench (Malicious Objectives Sequenced As Innocuous Compliance), a benchmark of 199 three-stage attack chains paired with deterministic exploit oracles on deployed software substrates (10 web-application substrates, 31 CWE classes, 5 programming languages) that treats both exploit ground truth and downstream reviewer protocol as first-class evaluation axes. On this benchmark, nine production coding agents from Anthropic, OpenAI, Google, Moonshot, Zhipu, and Minimax compose innocuous tickets at 53-86% end-to-end ASR with only two refusals across all staged runs. In a matched direct-prompt experiment over four frontier Claude/Codex agents, vulnerable-output rates fall to 0-20.4%: Claude primarily refuses, while Codex primarily hardens rather than emitting the vulnerable implementation - ticket staging silences both defense modes simultaneously. Downstream, code reviewer agents approve 25.8% of these confirmed-vulnerable cumulative diffs as routine PRs, and a full-context implementation protocol closes only 50% of the staged/direct gap, ruling out context fragmentation as the sole explanation. As a deployable but non-adaptive mitigation, reframing the reviewer as an adversarial pentester reduces evasion across the evaluated reviewer subset; pentester framed evasion ranges from 3.0% to 17.6%, and an open-weight Gemma-4-E4B-it reviewer under this framing detects 88.4% of attacks on the dataset with a 4.6% false-positive rate measured on 608 real-world GitHub PRs.

Semantic Denial of Service in LLM-controlled robots

Safety-oriented instruction-following is supposed to keep LLM-controlled robots safe. We show it also creates an availability attack surface. By injecting short safety-plausible phrases (1-5 tokens) into a robots audio channel, an adversary can trigger the models safety reasoning to halt or disrupt execution without jailbreaking the model or overriding its policy. In the embodied setting, this is a semantic denial-of-service attack: the agent stops because the injected signal looks like a legitimate alert. Across four vision-language models, seven prompt-level defenses, three deployment modes, and single- and multi-injection settings, we find that prompt-only defenses trade off attack suppression against genuine hazard response. The strongest defenses reduce hard-stop attack success on some models, but defenses change the form of disruption, not its fact: suppressed hard stops re-emerge as acknowledge loops and false alerts, which we measure with Disruption Success Rate (DSR). We further find that injection variety is consistently more effective than repeating the same phrase, suggesting that models treat diverse safety cues as corroborating evidence. The practical implication is architectural rather than prompt-level: systems that route unauthenticated audio text directly into the LLM create an avoidable security dependency between safety monitoring and action selection.

Split and Conquer Partial Deepfake Speech

Partial deepfake speech detection requires identifying manipulated regions that may occur within short temporal portions of an otherwise bona fide utterance, making the task particularly challenging for conventional utterance-level classifiers. We propose a split-and-conquer framework that decomposes the problem into two stages: boundary detection and segment-level classification. A dedicated boundary detector first identifies temporal transition points, allowing the audio signal to be divided into segments that are expected to contain acoustically consistent content. Each resulting segment is then evaluated independently to determine whether it corresponds to bona fide or fake speech. This formulation simplifies the learning objective by explicitly separating temporal localization from authenticity assessment, allowing each component to focus on a well-defined task. To further improve robustness, we introduce a reflection-based multi-length training strategy that converts variable-duration segments into several fixed input lengths, producing diverse feature-space representations. Each stage is trained using multiple configurations with different feature extractors and augmentation strategies, and their complementary predictions are fused to obtain improved final models. Experiments on the PartialSpoof benchmark demonstrate state-of-the-art performance across multiple temporal resolutions as well as at the utterance level, with substantial improvements in the accurate detection and localization of spoofed regions. In addition, the proposed method achieves state-of-the-art performance on the Half-Truth dataset, further confirming the robustness and generalization capability of the framework.

DiT-Flow: Speech Enhancement Robust to Multiple Distortions based on Flow Matching in Latent Space and Diffusion Transformers

Recent advances in generative models, such as diffusion and flow matching, have shown strong performance in audio tasks. However, speech enhancement (SE) models are typically trained on limited datasets and evaluated under narrow conditions, limiting real-world applicability. To address this, we propose DiT-Flow, a flow matching-based SE framework built on the latent Diffusion Transformer (DiT) backbone and trained for robustness across diverse distortions, including noise, reverberation, and compression. DiT-Flow operates on compact variational auto-encoders (VAEs)-derived latent features. We validated our approach on StillSonicSet, a synthetic yet acoustically realistic dataset composed of LibriSpeech, FSD50K, FMA, and 90 Matterport3D scenes. Experiments show that DiT-Flow consistently outperforms state-of-the-art generative SE models, demonstrating the effectiveness of flow matching in multi-condition speech enhancement. Despite ongoing efforts to expand synthetic data realism, a persistent bottleneck in SE is the inevitable mismatch between training and deployment conditions. By integrating LoRA with the MoE framework, we achieve both parameter-efficient and high-performance training for DiT-Flow robust to multiple distortions with using 4.9% percentage of the total parameters to obtain a better performance on five unseen distortions.

Where Vision Becomes Text: Locating the OCR Routing Bottleneck in Vision-Language Models

Vision-language models (VLMs) can read text from images, but where does this optical character recognition (OCR) information enter the language processing stream? We investigate the OCR routing mechanism across three architecture families (Qwen3-VL, Phi-4, InternVL3.5) using causal interventions. By computing activation differences between original images and text-inpainted versions, we identify architecture-specific OCR bottlenecks whose dominant location depends on the vision-language integration strategy: DeepStack models (Qwen) show peak sensitivity at mid-depth (about 50%) for scene text, while single-stage projection models (Phi-4, InternVL) peak at early layers (6-25%), though the exact layer of maximum effect varies across datasets. The OCR signal is remarkably low-dimensional: PC1 captures 72.9% of variance. Crucially, principal component analysis (PCA) directions learned on one dataset transfer to others, demonstrating shared text-processing pathways. Surprisingly, in models with modular OCR circuits (notably Qwen3-VL-4B), OCR removal can improve counting performance (up to +6.9 percentage points), suggesting OCR interferes with other visual processing in sufficiently modular architectures.

Autonomous Oil Spill Response Through Liquid Neural Trajectory Modeling and Coordinated Marine Robotics

Marine oil spills pose grave environmental and economic risks, threatening marine ecosystems, coastlines, and dependent industries. Predicting and managing oil spill trajectories is highly complex, due to the interplay of physical, chemical, and environmental factors such as wind, currents, and temperature, which makes timely and effective response challenging. Accurate real-time trajectory forecasting and coordinated mitigation are vital for minimizing the impact of these disasters. This study introduces an integrated framework combining a multi-agent swarm robotics system built on the MOOS-IvP platform with Liquid Time-Constant Neural Networks (LTCNs). The proposed system fuses adaptive machine learning with autonomous marine robotics, enabling real-time prediction, dynamic tracking, and rapid response to evolving oil spills. By leveraging LTCNs--well-suited for modeling complex, time-dependent processes--the framework achieves real-time, high-accuracy forecasts of spill movement. Swarm intelligence enables decentralized, scalable, and resilient decision-making among robot agents, enhancing collective monitoring and containment efforts. Our approach was validated using data from the Deepwater Horizon spill, where the LTC-RK4 model achieved 0.96 spatial accuracy, surpassing LSTM approaches by 23%. The integration of advanced neural modeling with autonomous, coordinated robotics demonstrates substantial improvements in prediction precision, flexibility, and operational scalability. Ultimately, this research advances the state-of-the-art for sustainable, autonomous oil spill management and environmental protection by enhancing both trajectory prediction and response coordination.

AutoLoop: Fast Visual SLAM Fine-tuning through Agentic Curriculum Learning

Current visual SLAM systems face significant challenges in balancing computational efficiency with robust loop closure handling. Traditional approaches require careful manual tuning and incur substantial computational overhead, while learning-based methods either lack explicit loop closure capabilities or implement them through computationally expensive methods. We present AutoLoop, a novel approach that combines automated curriculum learning with efficient fine-tuning for visual SLAM systems. Our method employs a DDPG (Deep Deterministic Policy Gradient) agent to dynamically adjust loop closure weights during training, eliminating the need for manual hyperparameter search while significantly reducing the required training steps. The approach pre-computes potential loop closure pairs offline and leverages them through an agent-guided curriculum, allowing the model to adapt efficiently to new scenarios. Experiments conducted on TartanAir for training and validated across multiple benchmarks including KITTI, EuRoC, ICL-NUIM and TUM RGB-D demonstrate that AutoLoop achieves comparable or superior performance while reducing training time by an order of magnitude compared to traditional approaches. AutoLoop provides a practical solution for rapid adaptation of visual SLAM systems, automating the weight tuning process that traditionally requires multiple manual iterations. Our results show that this automated curriculum strategy not only accelerates training but also maintains or improves the model's performance across diverse environmental conditions.

Unmasking Deepfakes: Leveraging Augmentations and Features Variability for Deepfake Speech Detection

The detection of deepfake speech has become increasingly challenging with the rapid evolution of deepfake technologies. In this paper, we propose a hybrid architecture for deepfake speech detection, combining a self-supervised learning framework for feature extraction with a classifier head to form an end-to-end model. Our approach incorporates both audio-level and feature-level augmentation techniques. Specifically, we introduce and analyze various masking strategies for augmenting raw audio spectrograms and for enhancing feature representations during training. We incorporate compression augmentations during the pretraining phase of the feature extractor to address the limitations of small, single-language datasets. We evaluate the model on the ASVSpoof5 (ASVSpoof 2024) challenge, achieving state-of-the-art results in Track 1 under closed conditions with an Equal Error Rate of 4.37%. By employing different pretrained feature extractors, the model achieves an enhanced EER of 3.39%. Our model demonstrates robust performance against unseen deepfake attacks and exhibits strong generalization across different codecs.

Robust Monocular Visual Odometry using Curriculum Learning

Curriculum Learning (CL), drawing inspiration from natural learning patterns observed in humans and animals, employs a systematic approach of gradually introducing increasingly complex training data during model development. Our work applies innovative CL methodologies to address the challenging geometric problem of monocular Visual Odometry (VO) estimation, which is essential for robot navigation in constrained environments. The primary objective of our research is to push the boundaries of current state-of-the-art (SOTA) benchmarks in monocular VO by investigating various curriculum learning strategies. We enhance the end-to-end Deep-Patch-Visual Odometry (DPVO) framework through the integration of novel CL approaches, with the goal of developing more resilient models capable of maintaining high performance across challenging environments and complex motion scenarios. Our research encompasses several distinctive CL strategies. We develop methods to evaluate sample difficulty based on trajectory motion characteristics, implement sophisticated adaptive scheduling through self-paced weighted loss mechanisms, and utilize reinforcement learning agents for dynamic adjustment of training emphasis. Through comprehensive evaluation on the real-world TartanAir dataset, our Curriculum Learning-based Deep-Patch-Visual Odometry (CL-DPVO) demonstrates superior performance compared to existing SOTA methods, including both feature-based and learning-based VO approaches. The results validate the effectiveness of integrating curriculum learning principles into visual odometry systems.

Adaptive USVs Swarm Optimization for Target Tracking in Dynamic Environments

This research investigates the performance and efficiency of Unmanned Surface Vehicles (USVs) in multi-target tracking scenarios using the Adaptive Particle Swarm Optimization with k-Nearest Neighbors (APSO-kNN) algorithm. The study explores various search patterns-Random Walk, Spiral, Lawnmower, and Cluster Search to assess their effectiveness in dynamic environments. Through extensive simulations, we evaluate the impact of different search strategies, varying the number of targets and USVs' sensing capabilities, and integrating a Pursuit-Evasion model to test adaptability. Our findings demonstrate that systematic search patterns like Spiral and Lawnmower provide superior coverage and tracking accuracy, making them ideal for thorough area exploration. In contrast, the Random Walk pattern, while highly adaptable, shows lower accuracy due to its non-deterministic nature, and Cluster Search maintains group cohesion but is heavily dependent on target distribution. The mixed strategy, combining multiple patterns, offers robust performance across varied scenarios, while APSO-kNN effectively balances exploration and exploitation, making it a promising approach for real-world applications such as surveillance, search and rescue, and environmental monitoring. This study provides valuable insights into optimizing search strategies and sensing configurations for USV swarms, ultimately enhancing their operational efficiency and success in complex environments.