Adversarial Examples

With AI increasingly deployed in safety-critical systems, providing formal robustness guarantees for the underlying models is essential. Existing verification methods either rely on overly conservative approximations or incur prohibitive computational costs. For example, the use of lp-norm perturbations in video settings encodes the belief that the adversary can inject noise in every video frame. In practice, adversarial perturbations exhibit structured spatial and temporal correlations, constrained to lower-dimensional, semantically meaningful subspaces. In this work, we study robustness verification of 3D CNNs processing video and volumetric inputs, targeting applications in action recognition (UCF-101), autonomous driving (Udacity), and medical imaging (MedMNIST) exploiting realistic assumptions on adversarial strength by modelling them as spatio-temporal constraints - where the attacker can modify either a subset of frames or patches within a set of consecutive frames. We demonstrate that modelling realistic constraints enables tighter approximations. We introduce Spatio-Temporal Bound Propagation (STBP), a verification framework that computes an exact closed-form characterization of the first convolutional layer and propagates certified bounds through subsequent layers using scalable approximations. Computing the exact closed form provides the tightest bounds for the first convolutional layer. Thus, we utilise approximation methods in the remainder of the network. To spur further progress in this field, we propose ST-Bench, a verification benchmark for autonomous driving and activity recognition, to systematically evaluate verifiable robustness. Compared to existing verification-based approaches, STBP provides stronger robustness guarantees with significantly improved scalability, achieving 1.7x higher certified robust accuracy under identical perturbation budgets.

Large language model (LLM)-powered content moderation systems are a critical defense against harmful online content. However, they operate primarily on tokenized text and often overlook visual cues that humans naturally use when interpreting content. We show that this limitation creates a fundamental vulnerability: content readily recognized as harmful by humans can evade automated moderation. To systematically study this problem, we introduce Human-Perceptible Adversarial Attacks (HPAA), which embed harmful expressions into otherwise benign text using visually salient typographic manipulations. HPAA strategically combines features such as spacing, emphasis, and spatial arrangement to preserve human recognition while reducing machine detectability. Operating in a black-box setting with a small query budget, the attack automatically generates evasive content without model access or gradient information. We evaluate HPAA on multiple datasets and thirteen widely deployed moderation systems, including commercial APIs and state-of-the-art open-source guardrails. With only three detector queries, generated attacks achieve over 86% human recognition while keeping detection rates below 1% across evaluated systems. We further identify the typographic factors driving successful evasion, analyze why current moderation architectures fail to capture these signals, and discuss practical defenses. Our findings reveal a fundamental blind spot in current LLM-based moderation systems and motivate moderation approaches that better align with human perceptual understanding.

AutoMegaKernel (AMK) compiles a HuggingFace Llama-family model into a single persistent cooperative CUDA kernel that runs the whole forward pass in one launch, with no per-model hand-written CUDA. The contribution is the system, not raw speed. A frozen schedule-IR validator statically certifies deadlock-freedom and race-freedom via static graph checks (not a mechanized proof), so an unsafe agent-proposed schedule is rejected before launch: across 7,160 adversarial schedules (6,091 unsafe) it had zero false-accepts and accepted all 360 real lowerings. The same source retargets sm_80/sm_90/sm_120 from one codebase, auto-generates correct megakernels for 10 of 10 supported models, and on a real SmolLM2-135M checkpoint reproduces HuggingFace greedy decode token-for-token (perplexity match 2.5e-7). An unattended, agent-drivable autoresearch loop self-improves the megakernel over its own baseline (1.25-1.72x). A search-found int8 (W8A16) megakernel beats CUDA-graphed cuBLAS bf16 at batch-1 decode across NVIDIA's datacenter inference fleet: L4 up to 1.33x, the current-gen L40S 1.25-1.27x, A10G up to 1.08x at scale, and the consumer RTX 5090 1.19-1.23x. The ordering is not a clean function of bandwidth (the 864 GB/s L40S beats the 600 GB/s A10G); the divide is inference-class vs training-class. AMK trails cuBLAS on the high-bandwidth training-class A100/H100, where the harness localizes the cross-SM-sync bottleneck; we report the gap plainly. This is a precision-asymmetric (W8A16 vs bf16) comparison at decode position 0; the largest real checkpoint is TinyLlama-1.1B. Code and the harness: https://github.com/RightNow-AI/AutoMegaKernel

Self-supervised data curation provides a pathway to scaling and improving the generalization capabilities of machine learning models. By leveraging self-supervised learning (SSL) for data curation, the demand for massive training datasets required by foundation models can be effectively met. SSL greatly alleviates the costs associated with annotation and manual dataset curation while minimizing the need for human oversight. However, the integrity of SSL-curated datasets must be rigorously checked, as reliance on anonymous and unvetted external sources can substantially increase the risk of data poisoning. In this paper, we propose a Poisoned Data Detector (PDD), an active defense mechanism designed to ensure the integrity of SSL-curated datasets prior to foundation model training. PDDs are designed using a combination of the pretrained ImageBind model and traditional classifiers, including Random Forest (RF), k-Nearest Neighbors (KNN), Naive Bayes (NB), and Support Vector Machines (SVM). We rigorously evaluated PDDs using 176,200 images from three diverse datasets and three different adversarial attacks encompassing both in-distribution and out-of-distribution scenarios. Notably, SVM-PDD achieves superior performance for both in-distribution (Set3-Set5) and out-of-distribution (TrueFace and 140K RealFace) datasets. Our design demonstrates strong scalability and enables the rapid integration of new adversarial attack detectors through an ensemble approach.

While confidence calibration is essential for trustworthy decision-making in safety-critical applications, the robustness of calibrated GNNs to adversarial structural perturbations remains largely unexplored. However, studying calibration attacks on graphs presents unique technical challenges: (1) the discrete nature of graph structures complicates gradient-based optimization, (2) existing underconfidence objectives fail to drive predictions toward uniform distributions, and (3) GNNs are highly sensitive to edge perturbations, often causing unintended label changes that violate attack constraints. To address these challenges, we propose a \textbf{Unified Graph Calibration Attack (UGCA)} framework designed for \textbf{worst-case (white-box) analysis} of GNN calibration robustness. UGCA introduces a KL-divergence loss to encourage uniform predictive distributions, a reranking mechanism to reduce label flipping, a hybrid loss to recover labels when violations occur, and beam search to explore a broader adversarial search space. We further provide theoretical insights linking model generalization, dataset complexity, and calibration vulnerability, showing that models with higher accuracy or trained on datasets with more classes are more susceptible under this threat model. Extensive experiments demonstrate that UGCA substantially increases Expected Calibration Error while preserving classification accuracy. Our code is publicly available at https://github.com/CaptainCuong/Graph-Calibration-Attack.git.

Vertical federated learning (VFL) is a distributed learning paradigm that leverages vertically partitioned features across isolated parties without sharing raw samples; however, it remains vulnerable to active sample reconstruction attacks. Existing defenses fail to achieve a satisfactory trade-off between model utility and privacy protection, due to either suppressing task-relevant information alongside privacy-sensitive features or relying on end-to-end supervised training to converge the defense module, which exposes the model to early-epoch vulnerability. To address this challenge, we adopt a structural causal model (SCM) insight and construct CausShield. From a task-learning standpoint, causal features within a raw sample are those that are directly relevant and contributory to the learning objective, whereas non-causal features are task-irrelevant but often encode sample-specific private information, thereby facilitating reconstruction. Importantly, we lay a theoretical foundation to prove this insight. CausShield thus decomposes the shared representations between the client and the coordinating server in VFL into task-relevant and task-irrelevant components to ensure full-cycle privacy protection. Nonetheless, the decomposition is inherently challenging due to the dual objectives of preserving model utility while mitigating privacy leakage. We address this via a carefully formulated optimization problem, which is solved through unsupervised representation learning. We further theoretically prove that CausShield preserves the convergence behavior of standard VFL. Extensive experiments compare CausShield against seven SOTAs, including InvL (USENIX Security'25), and evaluate robustness against advanced reconstruction attacks such as URVFL (NDSS'25). Results demonstrate that CausShield consistently outperforms in privacy protection, model utility, and computational efficiency.

Cross-domain shifts challenge Presentation Attack Detection (PAD) on ID Cards, given the restricted data available due to privacy concerns. This work proposes a compact multimodal model, based on new generative and discriminative blocks, which combines visual and textual data for PAD on genuine and synthetic ID images. While multimodal models exhibit strong generalisation after supervised fine-tuning, they fail in zero-shot settings. Our findings underscore that model capacity and real-world data are essential for reliable PAD, while existing synthetic datasets may not reflect real-world challenges. We argue for a re-evaluation of synthetic data as a benchmark and emphasise the need for more realistic, diverse datasets to advance PAD research.

Vision-language models (VLMs) such as CLIP achieve strong zero-shot recognition but remain highly fragile under adversarial perturbations. Recent test-time adaptation defenses improve robustness by leveraging many augmented views, but this leads to impractical slowdown and a clear robustness-throughput trade-off. To address this challenge, we present Stability and Suitability-guided Test-time Prompt Tuning (SS-TPT), evaluating the quality of each augmented view via two complementary scores: (1) stability, measuring prediction invariance to weak augmentations, and (2) suitability, measuring feature-space density among views. These stability and suitability (SS) scores guide both adaptation and inference through an SS-guided consistency loss and an SS-weighted prediction, amplifying trustworthy views while suppressing corrupted ones. Extensive experiments demonstrate that SS-TPT significantly outperforms prior state-of-the-art methods, achieving superior robustness-throughput trade-offs across diverse datasets and varying numbers of views, thereby demonstrating both strong practicality and generality. Our code is available at https://github.com/sunoh-kim/SS-TPT.

LLM-driven automated penetration testing agents are typically evaluated against static targets that neither detect nor respond to attacks, so their behavior under intelligent defense remains untested. The causal consistency of multi-step attack chains likewise hinges on unstable LLM reasoning, and agent decisions remain opaque to human analysts. These three shortcomings, in realism, consistency, and auditability, are usually patched in isolation. We present ZERO-APT, a turn-based attacker-defender-judge framework that addresses them within a single architecture. For realism, ZERO-APT embeds a configurable LLM Defender that consumes Sysmon telemetry and detects attacks in real time, exposing the attacker to a live opponent rather than a passive target. For consistency, three architectural mechanisms move causal consistency from unstable LLM reasoning into enforced system architecture: separation of planning from execution, multi-dimensional ReAct feedback, and a hard-constraint-filtered action library. For auditability, a dedicated Judge agent adjudicates each round, maintains global state, and emits structured post-hoc CTI reports that make every decision traceable. We evaluate a Windows Server 2022 post-exploitation prototype across five scenarios with three Defender configurations. ZERO-APT reaches 79% attack success rate (Aurora 22%, PentestGPT 39%), a Causal Consistency Score of 0.860 (Aurora 0.930, Claude Code 0.520), and end-to-end decision auditability through structured CTI reports. We release the benchmark to support evaluation of penetration agents under intelligent defense.

Inverse adversarial training leverages high-confidence predictions to stabilize robust learning, yet we uncover a critical paradox: high confidence often stems from overfitting to non-causal background correlations rather than intrinsic object semantics. Our investigation reveals that visual context functions as a dual-natured signal, serving as either a necessary supportive prior or a spurious confounder. This insight renders existing blind suppression strategies flawed, as they inevitably lead to severe Feature Loss. To resolve this, we propose High-Confidence Causally Aligned Training (HICAT), a unified framework that establishes a Semantic Equilibrium. Operating on a ``Measure-Debias-Align'' pipeline, HICAT integrates a Learnable Background-Bias Estimator (LBBE) to adaptively diagnose context utility. Guided by this diagnosis, an Adaptive Debiasing mechanism performs surgical logit rectification, complemented by a geometrically grounded Foreground Logit Orthogonal Enhancement (FLOE) loss to enforce rigorous feature disentanglement. Extensive experiments on CIFAR-10, CIFAR-100, and ImageNet-1K demonstrate that HICAT consistently improves over matched baselines across diverse architectures (CNNs and ViTs) while significantly reducing the robust generalization gap.

Vision-language-action (VLA) models are gaining attention in robotics, yet their robustness to adversarial attacks remains largely unexplored. Existing work shows that adversarial patches can mislead VLA-based robots but assumes full access to the entire execution trajectory, an unrealistic requirement in practice. We address this limitation by formulating a partially observable threat model, where the adversary can exploit only a short prefix of the trajectory to generate a fixed patch applied to all subsequent frames. Under this setting, we propose a two-phase framework. First, we localize the patch using the model's attention maps to identify visually critical regions that correspond to the full instruction. Then, we optimize the patch to disrupt the semantic grounding of target objects and increase the curvature of action trajectories, thereby compounding failures in both perception and control. Extensive experiments in simulation and real-world robotic environments show that our method sustains adversarial effects under partial observability, inducing long-horizon disruptions and significantly reducing task success rates.

While principal component analysis (PCA) is a fundamental tool for dimensionality reduction, its dense representations make it ill-suited for high-dimensional data. Existing methods address this by promoting sparsity through explicit $\ell_1$-penalties, but these are not obvious to tune due to the unsupervised nature of the task. In contrast, we propose Adversarial PCA (AdvPCA), which leverages robust optimization to achieve sparsity by optimizing the reconstruction objective against bounded, worst-case latent space perturbations. We show that this formulation admits a closed-form reduction, leading to a practical iterative algorithm that alternates between adversarial linear regression-style updates for the sparse encoder and orthogonal updates for the decoder. By theoretically characterizing the solution, we derive a data-adaptive parameterization that allows the algorithm to perform effectively out of the box. We validate these claims through numerical experiments on synthetic and real-world genomics data.

In dynamic environments, large language models need to keep adapting to new tasks, but continual learning often suffers from forgetting, limited transfer, and vulnerability to adversarial perturbations. To address this, we present AdvCL, which repurposes adversarial perturbations as a geometric control signal for stable continual adaptation. AdvCL combines three plug-in modules: Intra-Smooth promotes local smoothness via small adversarial perturbations; Proto-Clip uses similarity clipping to prevent excessive alignment to current task prototype; and Inter-Align applies directional alignment toward previous task prototype to reduce representational gaps. Experiments show consistent gains in both standard performance and robustness, with lower forgetting and stronger transfer. We further analyze key mechanisms by quantifying the sensitivity of Intra-Smooth to perturbation settings and the effect of Inter-Align on task similarity and geometric distance. In summary, the modules provide complementary gains when combined, and each can also be integrated individually into diverse CL paradigms, including replay, regularization, and dynamic architectures, thereby offering a geometric control mechanism for continual learning.

The vulnerability of deep neural networks to adversarial examples poses a significant challenge for real-world deployment. Existing techniques to enhance deep network robustness rely on adversarial training, an approach that is powerful but computationally intensive and typically tailored to specific attack types. To address these limitations, existing works have explored techniques such as adding gaussian noise or filtering images, both of which can boost the network robustness to various adversarial attacks, albeit modestly. Here, we theoretically demonstrate that these two approaches enhance robustness against adversarial attacks through complementary mechanisms, resulting in supralinear robustness when combined. Building on this insight, we experimentally show that a simple preprocessor combining Gaussian noise and bilateral filtering yields supralinear improvements in adversarial robustness with minimal computational cost. Next, we combine our preprocessor with adversarial training and test on RobustBench to assess its supralinear improvement over state-of-the-art defenses. First, this combination ranks second on AutoAttack and third overall, while using only $\sim$35% of the training FLOPs, using a model with $\sim$50% less parametets, trained with $\sim$33% of the epochs and $\sim$15% the data compared to state-of-the-art defenses. Second, our method scales efficiently, matching the accuracy of competing models with roughly 2-8x less total compute across 3 orders of magnitude. Overall, our approach provides a principled and easily integrable framework for enhancing adversarial robustness, offering negligible computational overhead and a simple yet theoretically grounded design.

Recent advancements in generative AI have led to image editing models capable of producing realistic forgeries that evade traditional image forgery localization methods, as these approaches depend on physical noise absent in synthetic data. To address this challenge, we theoretically demonstrate that the diffusion process inherently suppresses local high-frequency variance, creating a statistical energy gap that is distinguishable from the natural entropy of optical imaging. Guided by this insight, we propose FLAME, a unified framework that utilizes a LAD map to capture these intrinsic anomalies, coupled with a parameter-efficient adapter for SAM to achieve precise, pixel-level forgery localization. Furthermore, to bridge the lag between forensic benchmarks and evolving generative models, we introduce EditStream, an automated pipeline for continuous, instruction-based training data synthesis. Extensive experiments demonstrate that FLAME establishes a new state-of-the-art, significantly outperforming previous methods on AI-generated forgery datasets while effectively generalizing to unseen generative architectures. Our code is available at https://github.com/phoenixnir/FLAME.

Modern neural networks are highly susceptible to adversarial perturbations. In this work, we identify that part of this vulnerability stems from the sensitivity of the widely used fully connected (FC) classifiers to such perturbations. In contrast, simple $\ell_2$ distance-based classifiers exhibit significantly greater robustness. We provide thorough theoretical and empirical analysis showing that while FC classifiers' high sensitivity makes them discriminative, it also makes them vulnerable. Conversely, $\ell_2$-classifiers' insensitivity grants robustness but limits performance. Motivated by this trade-off, we propose a novel $\ell_2$-reclassifier based on a Hybrid Prototype Mixing (HPM) framework. This method retains the discriminative power of FC classifiers while leveraging the robustness of $\ell_2$ distance. It yields $\ell_2$-distance-based predictions by fusing two prototype types: (1) stable, dataset-level prototypes updated via EMA, and (2) dynamic, batch-level prototypes generated from the FC classifier's predictions using a Straight-Through Estimator (STE). However, this dynamic, STE-based architecture introduces significant challenges for evaluation, such as gradient obfuscation and forward discontinuity. To address this, we propose a new, rigorous evaluation protocol, the Mixed Surrogate Attack (MSA), which uses multiple surrogates along with powerful AutoAttack to ensure a fair and robust assessment. Extensive experiments demonstrate that our lightweight, plug-and-play module, with minimal fine-tuning, effectively enhances the adversarial robustness of various existing SOTA adversarially trained models.

Retrieval-Augmented Generation (RAG) systems are widely deployed and increasingly influential, but their reliance on external corpora exposes new security risks from poisoned retrieval content. Existing RAG attacks are largely focusing on individual queries or narrow topic-local query sets, which limits their practical reach and offers limited camouflage in real-world settings. In this paper, we introduce discourse-level opinion manipulation, a new threat model in which coordinated influence across a semantic query network induces opinion shifts over a holistic, multi-topic query space. We formalize this threat in a black-box setting and propose DiscourseFlip, an agentic, graph-guided attack that dynamically allocates a limited poisoning budget to maximize discourse-level opinion deviation. Extensive experiments demonstrate that DiscourseFlip consistently induces targeted opinion shifts across the contextualized query network and significantly outperforms existing baselines in terms of coverage and effectiveness. User studies further confirm that DiscourseFlip is effective while remaining well camouflaged from user detection. Moreover, systematic analyses show that existing mitigation strategies are ineffective against discourse-level manipulation, underscoring the urgent need for more robust and adaptive defenses to address discourse-level vulnerabilities.

Adversarial Training (AT) is a leading defense against adversarial examples but often suffers from Catastrophic Overfitting (CO) in efficient single-step variants, where robustness to multi-step attacks collapses despite high single-step performance. We address this failure mode with two contributions. First, we formalize Epsilon Overfitting (EO), a perspective in which fixed perturbation magnitudes and directions exacerbate CO, and show that introducing perturbation variability significantly improves robust generalization across different architectures and datasets. Second, we propose PertAlign (Perturbation Alignment), a theoretically grounded, computationally negligible metric that predicts CO onset by measuring gradient alignment across attack stages. Leveraging these insights, we introduce SORA, an adaptive step-size AT method that dynamically adjusts perturbations based on loss surface geometry. SORA consistently prevents CO, achieves state-of-the-art robustness and clean accuracy, and generalizes across datasets and architectures using a single fixed set of hyperparameters, which is essential for applicability in fast AT. Extensive experiments on diverse datasets and architectures show that SORA matches or surpasses the robustness of prior methods while delivering higher clean accuracy and superior efficiency. Code is available at https://github.com/SecondOrderAT/SORA.

While decision-based black-box adversarial attacks present a severe security threat, current methodologies suffer from fundamental limitations. Pixel-wise attacks frequently introduce unnatural, high-frequency visual artifacts, while latent-space frameworks are confined by the limited search space of low-dimensional manifolds and inherent reconstruction flaws. To resolve these limitations, we propose Latent Geometric Chords (LGC) for Query-Efficient Decision-Based Adversarial Attacks alongside a variant, LGC-H. At its core, LGC navigates decision boundaries by executing a curvature-aware geometric search within a compressed semantic manifold. To guarantee high visual fidelity and circumvent dimensionality bottlenecks, we introduce a Residual-based Adversarial Generation (RAG) mechanism. RAG isolates semantic perturbations as geometric chords and superimposes them directly onto the original source image. RAG substantially resolves baseline reconstruction flaws and effectively doubles the permissible search space dimensions. Experimental results demonstrate that LGC achieves robust cross-dataset transferability and substantially outperforms state-of-the-art baselines. Notably, our method, LGC, minimizes perturbation magnitudes while achieving state-of-the-art visual fidelity--with a Structural Similarity Index Measure (SSIM) exceeding 0.99 and a Learned Perceptual Image Patch Similarity (LPIPS) below 0.01 at 5000 queries--and sustaining high attack success rates under stringent perceptual constraints, successfully compromising adversarially trained robust models. The source code is available at: https://github.com/eihmuekhine/Latent-Geometric-Chords.

Given the surge of harmful AI-generated imagery online, reliably distinguishing authentic images from generated ones has become an urgent research topic. While many proposed detection methods perform well under controlled settings, they often collapse when tested on real-world data. A potential root cause are subtle biases in the detectors' training data. As a result, detectors may rely on spurious correlations instead of learning true forensic artifacts. While a recent line of work has identified the problem, there is not yet an established protocol to evaluate how biased a detector actually is. In this work, we therefore take a step back: First, we discuss what it means for a detector to be biased, and how this differs from a lack of robustness. Second, we propose BIAS-ID, a transparent framework for analyzing and quantifying the presence of transformation biases in AI-generated image detectors. We validate our framework by performing an evaluation of six detectors across two datasets, revealing that several state-of-the-art detection methods are strongly affected by biases. Our results highlight the importance of bias-aware evaluation for developing reliable AI-generated image detectors.

Deep neural network (DNN)-based object detectors are widely used for analyzing aerial and satellite imagery in applications such as environmental monitoring and urban analytics. Despite their strong performance, these models are known to be vulnerable to adversarial examples, and physical adversarial attacks using printable patterns pose realistic security threats. In this paper, we evaluate physical adversarial patch attacks against an aerial vehicle detector by bridging digital optimization and real-world deployment. Adversarial patches are optimized in the digital domain using a loss function that minimizes the maximum objectness score while incorporating non-printability score (NPS) and total variation (TV) constraints to ensure both printability and spatial smoothness. The optimized patches are printed and deployed in three configurations: ON, OFF, and OFF-Side. Experiments using a YOLOv3 detector show that while the OFF patch achieves the highest effectiveness in the digital domain (85.51% Average Objectness Reduction Rate (AORR)), the ON patch demonstrates superior robustness in physical environments (0.197-0.343 Objectness Score Ratio (OSR)) due to its consistent visibility. Furthermore, our results indicate that weather-based augmentation does not necessarily improve patch optimization in this domain. These findings provide critical insights into the practical vulnerabilities of aerial object detection systems.

Adversarial patches are physical patterns attached to real objects to mislead AI vision systems. Their real-world risk is not determined by a single successful prediction, but by whether they remain effective after deployment under changing viewpoints, distances, and scene conditions. We refer to this property as scene robustness, the effectiveness of a deployed patch across conditions in a real environment. Yet existing evaluations do not measure scene robustness well: real image benchmarks are realistic but fixed, while simulators are controllable but not grounded in a specific real scene. We present AdvScene, a scene-grounded framework for measuring the scene robustness of adversarial patches in reconstructed real environments. AdvScene reframes evaluation as operational measurement: given a fixed deployed patch, it characterizes the patch's operational envelope - where and when the attack succeeds - as a function of viewpoint, distance, and scene context. A key challenge is that the attack is typically defined only in a single anchor view, while evaluation requires a representation that remains faithful under viewpoint changes. We formalize this as a constrained lifting problem and introduce Adversarial Patch-to-Scene Embedding (APSE), which resolves cross-view ambiguity while preserving attack-critical appearance and enforcing locality, target-surface attachment, and cross-view consistency. We validate AdvScene using real-world physical data and conduct a comprehensive evaluation of existing adversarial patches. Our results show that AdvScene reveals substantial scene-dependent variation in attack effectiveness that is not captured by existing image-centric or simulator-based evaluations.

The work presents an approach for addressing the challenge of robustness in Large Language Models (LLMs) to alterations and potential errors caused by semantically similar but textually different prompts. Recent works have shown that these kinds of prompt variations can significantly impact the performance of LLMs on tasks. The central question is: can LLMs' robustness to semantically-neutral prompt alterations be acquired without expensive retraining of the entire model? We address this question both theoretically and through experiments. Our theoretical analysis reveals a crucial factor impacting model robustness - a systematic expected shift or perturbation-induced bias in neural network module outputs. Motivated by this analysis, we show that robustness can be achieved via a simple fine-tuning process: debiasing for robustness. We identify conditions when debiasing helps and when it does not, and demonstrate, through both theory and extensive experiments, that debiasing for robustness may indeed be a quick and efficient tool to enhance robustness and provide certification against random prompt perturbations.

Forensic vision-language models (VLMs) have recently been developed to detect image tampering and provide natural-language explanations. However, their robustness against adversarial manipulation remains underexplored. Existing adversarial attacks typically aim to flip the model's binary judgment, while the accompanying explanation may still reveal forensic cues and contradict the attacked judgment. In this paper, we study judgment-explanation consistent adversarial attacks against forensic VLMs and propose JECA^2, a controlled white-box red-team diagnostic that jointly redirects visual attribution and aligns textual explanations with the target judgment. On the visual side, JECA^2 uses Grad-CAM-guided perturbations to divert attribution from tampered regions toward benign regions. On the textual side, it optimizes prompt embeddings toward authenticity-affirming semantics under a token-proximity constraint. Experiments on forensic VLM benchmarks show that JECA^2 achieves higher attack success and automated judgment-explanation consistency than implemented baselines under white-box threat settings, while transfer to closed-source VLMs remains measurable but limited. Our results highlight a consistency failure mode in explanation-based forensic VLMs and motivate future robustness evaluation beyond binary detection accuracy.

While Vision-Language-Action (VLA) models have emerged as powerful generalist policies, their severe vulnerability to adversarial patches significantly hinders their deployment in safety-critical domains. Moreover, existing patch attacks primarily focus on white-box settings, heavily overfitting to the specific action output space of the target model, which results in poor cross-architecture transferability. To overcome this limitation, we propose VLA-Hijack, a unified adversarial framework that breaks the transferability bottleneck by exploiting a fundamental vulnerability identified in this work: before planning any motion, a VLA model must first use visual information to locate its own robotic arm within the environment. Targeting this shared visual self-localization process, our approach concurrently optimizes Attention-Guided Proprioceptive Suppression to inhibit the real robotic arm's features, and Multimodal Proprioceptive Injection to establish the patch as a surrogate "phantom embodiment". By alternating between semantic concept anchoring and visual prototype projection, VLA-Hijack effectively severs the semantic relationship between the agent's true embodiment and its control policy. Extensive experiments across diverse architectures (OpenVLA, UniVLA, and CronusVLA) demonstrate that VLA-Hijack achieves superior optimization efficiency in white-box settings and sets a new SOTA for cross-architecture and cross-domain black-box transferability.

Ultra-low-field (ULF) MRI offers portable and low-cost imaging but suffers from poor image quality. To address this, we present our submission to the 2025 ULF Enhancement Challenge (ULF-EnC), where the goal is to synthesise high-field-like MRIs from 64 mT scans. Our pipeline enhances ULF MRI through a combination of anatomical conditioning and model ensembling. We first generate tissue segmentation priors using a Swin UNETR trained solely on challenge-provided data. These priors condition two independent enhancement networks - a CycleGAN and a transformer-based residual enhancement model (T-REX) - each trained to synthesise 3 T-like MRIs. Outputs from both models are combined using a weighted average. Our approach produces enhanced MRIs that were comparable to high-field scans both quantitatively and qualitatively.

This paper describes our approach for the 8th UG2+ Workshop (CVPR 2026) Track~2, which targets semantic segmentation of outdoor scenes degraded by five weather conditions: blur, darkness, snow, haze, and glare. A central challenge we observe is a severe generalization gap -- models that perform well on the validation set often collapse on the test set. For instance, SegFormer-B5 drops 16.1 mIoU points from validation to test, suggesting that model capacity alone is insufficient for robustness. We investigate whether a carefully designed training recipe, rather than architectural complexity, can address this gap. Starting from a pre-trained SegMAN-S backbone, we systematically study the effects of domain-adaptive fine-tuning, multi-source data mixing, scene-balanced sampling, and synthetic degradation augmentation. Our final system achieves 59.9% mIoU on the official test set while maintaining a validation-test gap of only 6.5 points -- less than half that of larger models. We analyze negative results from architectural modifications, loss function variants, and model scaling to provide practical insights for weather-robust segmentation under limited data.

Despite recent progress in backdoor attacks, existing methods remain susceptible to post-training defenses that erase the backdoor through fine-tuning or pruning. We revisit the core objectives of backdoor attacks and derive principled criteria characterizing optimal sample-specific trigger construction under a Bayes-optimal model of the victim's training. Our analysis reveals that both attack success and clean-accuracy preservation are simultaneously optimized when triggered samples are steered into low-density regions of the clean data distribution, a distributional condition that controls all moments of the poisoned distribution at once rather than a handful of input-space summary statistics. We introduce a bilevel optimization framework that estimates density ratios via conditional time-score matching and optimizes a mixture-model objective to place triggered samples in these sparse regions. Extensive evaluations on MNIST, CIFAR-10, GTSRB, and TinyImageNet demonstrate that our method achieves above 99% attack success rate before defense and retains 50--85 percentage points higher post-defense ASR than the strongest baselines under fine-tuning defenses. Against neuron-pruning defenses, the method exhibits complete immunity, with zero neurons identified for removal across all pruning thresholds. These results expose a fundamental gap in current defense paradigms and underscore the need for defenses that operate beyond the support of the clean distribution.

With the rapid proliferation of generative models, such as diffusion models, digital watermarking has emerged as a crucial solution for identifying AI-generated images. Modern post-hoc watermarking schemes use neural networks to achieve an extremely low false-alarm rate while remaining robust to common image transformations. However, there is a lack of comparison between these modern methods and classic ones, particularly in real-world scenarios where robustness and security take precedence over achieving an extremely low false-alarm probability. In this paper, we propose a fair comparison of robustness and security between modern and classic post-hoc watermarking across various types of classic augmentations and recent sophisticated attacks. Our experiments show that, in a realistic scenario, classic watermarking outperforms modern techniques in terms of security while maintaining robustness.

Adversarial training (AT) remains one of the most reliable empirical defenses against adversarial attacks. Its robustness critically depends on how the underlying min-max objective is optimized. In practice, Stochastic Gradient Descent (SGD) optimizer remains the default optimization choice for AT, whereas adaptive optimizers often improve standard training but may yield inferior robustness. Recently, the Muon optimizer, which orthogonalizes matrix-valued updates via an approximate polar decomposition, has achieved notable success in large-scale training at a memory cost comparable to SGD. This raises a security-relevant question: \textit{can orthogonalized optimization improve AT under strong and heterogeneous threat models?} Focusing on this problem, we conduct a comprehensive theoretical and empirical study. Theoretically, we show that Muon imposes a spectral-norm stability ceiling on matrix updates, limiting uncontrolled spectral growth in the training dynamics without explicitly shrinking the learned weights. Empirically, across five architectures and three $\ell_p$ threat models ($\ell_\infty$, $\ell_1$, $\ell_2$) and their union, Muon is competitive with SGD on CNNs and substantially outperforms AdamW on both CNNs and ViTs. These results identify optimizer geometry as a security-relevant factor in adversarial training, while clarifying the empirical regimes in which orthogonalized updates are beneficial. Overall, our findings highlight optimizer design as a security-critical component of AT.

The Maximal Covering Location-Interdiction Problem (MCLIP) is a classic bi-level optimization problem, which is fundamental to resilient infrastructure planning yet remains computationally intractable. Specifically, the upper level determines facility locations to maximize coverage, while the lower level executes worst-case interdiction to minimize the coverage. The strong coupling between the upper and lower levels, combined with their respective high combinatorial complexity, renders traditional methods ineffective. To bridge this gap, we propose a Dual-Agent Deep Reinforcement Learning (DADRL) framework based on adversarial learning, comprising a location agent corresponding to the upper level and an interdiction agent corresponding to the lower level. Our contributions are threefold: (1) The location agent is trained simultaneously against an evolving interdiction agent, making it effectively capture the dynamic competitive interplay between the upper and lower levels; (2) To fully exploit the learned capabilities of the interdiction agent, we propose a Surrogate-based Ensemble Inference Strategy that utilizes the trained interdiction agent as a high-fidelity surrogate to guide the decisions of location agent; (3) Extensive experiments on synthetic and real-world datasets demonstrate that our approach achieves superior computational efficiency while maintaining highly competitive solution quality compared to other baselines. Furthermore, our DADRL framework is model-agnostic to network structures, while its underlying adversarial learning paradigm demonstrates strong potential for solving other bi-level optimization problems.

Vision Language Models adapt well to downstream tasks but are highly vulnerable to adversarial perturbations that disrupt cross-modal semantic alignment. Existing defenses are largely unidirectional or structural, failing to exploit bidirectional cross-modal complementarity and instance-wise adaptive protection. To overcome the limitations of unidirectional and static defenses in adversarial settings, we propose Closed-Loop Bidirectional Prompting, casting robust adaptation as cross-modal agreement recovery via a dynamic feedback loop on frozen encoders. A Semantic Anchor is introduced as a stable prior to constrain cyclic updates and mitigate perturbation-induced feature corruption. Through anchor-based bootstrapping, textual semantics denoise visual representations, while the refined visuals enable instance-adaptive prompt updating, yielding a rectified and robust consensus. Extensive evaluations across 11 datasets validate state-of-the-art robustness and strong base-to-new generalization, while maintaining a favorable trade-off between computational cost and accuracy.

Black-box adversarial attacks that minimize only the ground-truth confidence suffer from class drift: perturbations wander through the feature space without committing to a specific adversarial class, wasting queries on diffuse, undirected progress. We introduce Opportunistic Target Selection (OTS), a lightweight wrapper that switches an untargeted attack to a targeted objective early in its trajectory, locking onto whichever non-true class currently leads. OTS requires no architectural modification to the underlying attack, no gradient access, and no a priori target-class knowledge. We validate OTS on three score-based attacks (SimBA, Square Attack with cross-entropy loss, and Bandits) across five standard ImageNet classifiers (4,500 runs). On random-search attacks, OTS closely tracks oracle performance, with gains up to +27 pp in success rate and 43% relative reduction in censored-mean iterations on ResNet-50. On gradient-estimation attacks (Bandits) and attacks with margin loss, OTS is redundant, a negative result that reinforces our interpretation of OTS as a margin-loss surrogate. On adversarially-trained models, a bimodal difficulty distribution eliminates the regime where targeting helps.

Reinforcement learning with verifiable rewards (RLVR) has driven breakthroughs in domains such as math, tool-use, and software engineering, yet its extension to computer-use agents (CUAs) has been bottlenecked by the scarcity of scalable training data with deterministic rewards. Constructing such data for CUAs requires consistent task instruction, executable environment, and verifiable reward. However, hand-curated benchmarks achieve high reward fidelity but cover few applications and LLM-as-judge-based datasets scale broadly but lack reliable verification. We present CUA-Gym, a scalable pipeline that co-generates task instructions, environment states, and reward functions. Concretely, a Generator agent constructs the initial and golden environment states, and a separate Discriminator agent writes the reward function from the task specification. An orchestrator agent drives the two through iterative rounds upon execution. Generated tuples then pass a final filter combining LLM majority voting and agent rollouts, ensuring quality beyond the per-task adversarial loop. To address the scarcity of training environments, we further synthesize CUA-Gym-Hub, a broad suite of high-fidelity mock web applications grounded in real-world software-use distributions, expanding the scale of CUA RLVR data by magnitude. Using this pipeline, we construct CUA-Gym, a dataset of 32,112 verified RLVR training tuples grounded in 110 environments. Trained with GSPO on CUA-Gym, our CUA-Gym-A3B and CUA-Gym-A17B achieve 62.1% and 72.6% on OSWorld-Verified, outperforming prior open-source CUAs at comparable scales, with performance scaling smoothly in both data volume and environment diversity. The same checkpoints also improve on the held-out WebArena benchmark, indicating transfer beyond the training environments. We will open-source the full synthesis pipeline, dataset, CUA-Gym-Hub environments, and models.

Face recognition systems are increasingly vulnerable to morphing attacks, where a composite image is crafted to match multiple identities, enabling unauthorized access and identity fraud. Existing detection methods identify morphed images but cannot recover constituent images or identities, limiting their forensic utility. This paper presents a novel reference-free facial demorphing framework that leverages Multimodal Large Language Models (MLLMs) to guide a coupled diffusion-based reconstruction process. Our key innovation lies in extracting semantic embeddings from intermediate MLLM layers to condition the demorphing, providing high-level reasoning about facial attributes and identity cues that complement low-level pixel information. We formulate demorphing as a coupled conditional generation problem, where both constituent faces are synthesized jointly through a denoising diffusion model operating directly in the RGB domain, ensuring inter-identity consistency while preserving fine-grained perceptual details. Unlike prior approaches that rely on compressed latent representations or assume identity overlap between training and testing sets, our method bypasses lossy text generation-reencoding cycles by directly utilizing MLLM hidden states as conditioning signals, enabling the denoising network to attend to subtle visual cues such as hair, background, and facial textures. Ablation studies further reveal that middle MLLM layers encode more identity-discriminative representations, RGB-domain demorphing outperforms latent-space approaches by 30--40% at strict operating points, and full MLLM embeddings provide substantial advantages over raw ViT features through enhanced semantic structuring from multimodal pretraining.

Deep learning models are vulnerable to adversarial perturbations, raising important concerns for safety-critical deployment. Empirical defenses can achieve strong robustness in practice, but lack formal guarantees, motivating the need for certifiably robust classifiers. While certified methods provide formal guarantees, they often yield overly conservative bounds due to their inability to exploit structure in complex data distributions. In this work, we propose a framework for designing certifiably robust classifiers that leverages latent structure in data representations. We first analyze the Gaussian mixture setting, deriving necessary and sufficient conditions for the existence of robust classifiers and constructing a classifier with a closed-form robustness certificate and generalization guarantees. Our main contribution is to show that exact structure is not required: we prove that if a pretrained encoder maps inputs to a latent distribution that is $\varepsilon$-close (in KL divergence) to a Gaussian mixture, then certified accuracy degrades gracefully, with an explicit bound relating robustness under the true and approximate distributions. This result enables the direct use of pretrained models without requiring exact distributional assumptions. Empirically, our method achieves state-of-the-art or competitive certified accuracy on CIFAR-10 and ImageNet, while maintaining strong clean performance and low computational overhead. Overall, our work establishes approximate latent structure as a practical and principled route to certifiable robustness.

Adversarial images pose a severe security threat to multimodal large language models through prompt injection. Existing defenses largely lack a principled understanding of the underlying mechanisms and struggle to balance efficiency and defense utility. In this work, we show that successful adversarial attacks do not rely on the entire image uniformly but instead depend on a small subset of critical image tokens. Based on this insight, we propose Gradient Token Masking (GTM), which localizes these tokens via gradient analysis and neutralizes them through masking. We find that attribution based on the first generated token's output probability fails when attacks preserve the predicted token. To overcome this, GTM utilizes the Hidden-State Gradient Norm score for generation-influence attribution under adversarial inputs. We prove that its ranking is consistent with that of the full adversarial loss gradient, providing a theoretical guarantee for accurate localization. Our method requires only a single forward-backward pass to identify and zero out a small number of high-scoring tokens, effectively disrupting the adversarial attack path. Extensive experiments on prompt injection and multimodal jailbreak attacks demonstrate that our approach reduces attack success rates (ASR) to near zero while preserving model utility with negligible computational overhead.

Despite the strong performance of deep neural networks in modern Web and language applications, they remain vulnerable to adversarial attacks, especially transferable attacks that generate adversarial examples using surrogate models without accessing the victim model. Transferable attacks in the text domain are still under-explored, with only a few studies addressing this challenging issue, often with suboptimal results due to equal treatment of submodels or inaccurate estimation of importance scores. To address these challenges, we propose a simple yet effective paradigm for transfer-based textual adversarial attack, named SEP-Attack. Specifically, we employ the Determinantal Point Process (DPP) to generate diverse surrogate ensemble weights, representing the transferability of submodels. Using these weights, we introduce a new metric to evaluate prediction confidence scores, which in turn are used to calculate word importance scores and generate adversarial candidates. Finally, we quantify the transferability score for each candidate and select the top ones as the final transferable adversarial examples. Experiments conducted on four datasets and two real-world APIs validate the efficacy of SEP-Attack, significantly outperforming state-of-the-art baselines.

Detecting AI-generated images (AIGI) remains challenging because detectors often fail to generalize to unseen generators. Although existing methods are trained on large datasets, their performance still degrades when generation settings change, indicating that data scale alone is insufficient and that limited coverage of generative variations during training is a key factor. Studies on generative model editing show that small changes in internal representations can produce diverse and meaningful image variations, many of which are not explored under standard sampling. Leveraging this insight, we propose PROBE (Probing Robustness via Boundary Exploration), a framework that improves detector generalization by actively exploring challenging regions of the generative process. Instead of treating the generator as a fixed data source, PROBE uses the detector as a critic to steer the generator through manifold-level modifications, producing realistic samples that are difficult to classify. These samples expose failure cases that are uncommon under standard data sampling strategies and are used to refine the detector. Experimental results across multiple benchmarks indicate that PROBE enhances generalization to unseen generators, resulting in more generalizable AIGI detection performance. Code and models are available at https://github.com/Amamiya-C/PROBE-AIGI-Detection

This paper examines the erosion of Public Key Cryptography (PKC) security under adaptive adversarial optimisation driven by artificial intelligence. The problem addressed is the growing mismatch between algorithm-centric cryptographic security models and operational attack realities, where adversaries exploit implementation-level observability rather than breaking cryptographic primitives.

Jailbreak prompts can trigger harmful completions on aligned LLMs, In accordance, safety steering has been proposed: test-time activation interventions that steer jailbreak activations to trigger refusal while preserving benign utility. However, existing steering methods are fundamentally supervised and tied to a static, limited training set, whereas real jailbreaks evolve and are often out-of-distributed from the training set, leading to failures on unseen attacks. In this paper, we tackle the failure on unseen jailbreaks problem, base on unsupervised latent direction discovery. We propose a bi-level adversarial training framework for zero-shot jailbreak defense. In the inner step, we simulate diverse jail-broken activations by extrapolating from refusal-state harmful-request activations via unsupervised latent direction discovery, which expands the coverage of real jailbreak activation subspaces. In the outer step, we train a potential-induced steering field to push these adversarial jailbroken states into refusal regions while keeping benign unchanged. Across three LLMs and six classical jailbreak families, our method achieves strong defense with attack success rates mostly below 5%, and rising subspace coverage throughout training helps explain the improved generalization.

The integration of fairness and privacy in centralized data-driven applications is critical, especially as these systems increasingly influence sectors with significant societal impact. Current methods rarely address privacy, fairness, and accuracy together, which can potentially compromise ethical standards and privacy regulations. However, balancing these three objectives is quite challenging since each of objective often imposes conflicting requirements on the design and training of models, making it difficult to optimize one without compromising the others. This paper introduces a novel multitask adversarial model that treats fairness and privacy as integral objectives rather than afterthoughts, and learns a latent representation that hides sensitive attributes while preserving essential task-related information. Our approach dynamically balances fairness with accuracy and privacy through an optimized cost function with minimal performance loss even under strict conditions. Extensive testing on diverse datasets shows the ability of our model to achieve high standards of fairness and privacy without significant sacrifice to accuracy. Benchmarking against state-of-the-art privacy and fairness standards shows that our method enhances the robustness of privacy, fairness, and accuracy optimization, proving its adaptability across various datasets.

Multimodal large language models (MLLMs) need efficient mechanisms to update knowledge without degrading existing capabilities. While intrinsic multimodal knowledge editing achieves strong reliability and locality, it often exhibits limited generality, failing to propagate edits across semantically equivalent visual and linguistic variations. This issue arises from the lack of explicit semantic supervision, rigid editing scopes, and biased anchoring to individual samples in high-dimensional multimodal spaces. We address robust intrinsic multimodal knowledge editing by explicitly targeting generalization. We formalize robustness through knowledge units that group semantically equivalent multimodal inputs and define generality as consistent predictions within each unit. To expose fragile semantic regions, we introduce Latent Adversarial Robustification (LAR), which generates adversarial yet semantically coherent variants in the joint latent space. We further propose Rank-Constrained Subspace Learning (RCSL), enforcing low-rank alignment of adversarial representations at the edit layer via a singular value-based objective. Extensive analysis demonstrates the effectiveness of ASAM empirically.

Large language model (LLM)-based multi-agent systems increasingly rely on intermediate communication to coordinate complex tasks. While most existing systems communicate through natural language, recent work shows that latent communication, particularly through transformer key-value (KV) caches, can improve efficiency and preserve richer task-relevant information. However, KV caches also encode contextual inputs, intermediate reasoning states, and agent-specific information, creating an opaque channel through which sensitive content may propagate across agents without explicit textual disclosure. To address this, we introduce \textbf{LCGuard} (Latent Communication Guard), a framework for safe KV-based latent communication in multi-agent LLM systems. LCGuard treats shared KV caches as latent working memory and learns representation-level transformations before cache artifacts are transmitted across agents. We formalize representation-level sensitive information leakage operationally through reconstruction: a shared cache artifact is unsafe if an adversarial decoder can recover agent-specific sensitive inputs from it. This leads to an adversarial training formulation in which the adversary learns to reconstruct sensitive inputs, while LCGuard learns transformations that preserve task-relevant semantics and reduce reconstructable information. Empirical evaluations across multiple model families and multi-agent benchmarks show that LCGuard consistently reduces reconstruction-based leakage and attack success rates while maintaining competitive task performance compared to standard KV-sharing baselines.

Closed-weight generative services are increasingly deployed through query-based APIs, where users can obtain generated outputs while model parameters remain inaccessible. However, such deployment does not prevent model stealing: an attacker can repeatedly query the service, collect large volumes of released synthetic images, and use them as training data for a private substitute model. This query-output-driven process enables unauthorized knowledge distillation and capability replication without direct access to the original weights. To mitigate this threat, a practical defense should preserve the visual fidelity of released images, provide explicit control over perturbation magnitude, and scale efficiently to large-volume output release. We present WaveGuard, a single-pass, generator-based protection framework that safeguards released synthetic images under a user-specified perturbation budget. WaveGuard employs a frequency-aware perturbation generator to inject structured, imperceptible perturbations that maintain perceptual utility for benign viewers while reducing the usefulness of protected images as training data for unauthorized student models. Extensive experiments under WikiArt-related synthetic-output distillation settings show that WaveGuard achieves a favorable efficacy--fidelity--efficiency trade-off, with explicit imperceptibility control and substantial gains in protection efficiency.

Discrete diffusion models excel at visual synthesis but rely on slow, iterative decoding. Existing single-step distillation methods attempt to bypass this bottleneck, either by training auxiliary score networks that effectively double compute, or by introducing specialized parameterizations and multi-stage pipelines that fragment optimization. In this paper, we introduce Fixed-Point Distillation (FPD), an end-to-end framework that constructs local correction targets by partially corrupting the student's one-step draft and refining it with a single teacher step. To compute the training objective in a semantically meaningful space, we lift discrete tokens into continuous features and apply a multi-bandwidth drift loss that iteratively accumulates these corrections. To backpropagate through the discrete bottleneck, we employ a straight-through estimator that feeds exact hard-sampled tokens to the teacher and decoder during the forward pass, ensuring that training and inference operate on the same codebook manifold, while routing continuous gradients back to the student logits. This fully differentiable pathway additionally accommodates an optional unconditional adversarial objective to enhance perceptual realism. Evaluations on both class- and text-conditional generation validate the effectiveness of our framework. FPD achieves competitive visual fidelity and structural alignment within a single inference step, narrowing the gap to multi-step teachers while outperforming existing discrete distillation baselines.

Multimodal large language models (MLLMs) remain vulnerable to transfer-based targeted attacks, where perturbations optimized on open-source surrogate encoders can generalize to closed-source MLLMs. A key challenge for improving adversarial transferability is to effectively capture the intrinsic visual focus shared across different models, such that perturbations align with transferable semantic cues rather than surrogate-specific behaviors. However, existing methods suffer from spatial-domain feature redundancy and surrogate-specific gradient signals, thereby hindering cross-model transferability. In this paper, we propose FRA-Attack, which addresses both challenges from a unified frequency-domain regularization perspective. For feature alignment, a high-pass DCT objective on patch features suppresses redundant global structures and concentrates the loss on the high-frequency band that carries the MLLMs' intrinsic visual focus. For gradient optimization, we introduce Frequency-domain Gradient Regularization (FGR), a \textit{model-agnostic} low-pass regularizer that modulates the surrogate gradient using only the geometric frequency coordinate, \textit{i.e.}, no surrogate-derived statistic is involved, so that FGR is model-agnostic by construction, removing surrogate-specific high-frequency artifacts while preserving transferable low-frequency directions. Together, the two components form a unified frequency-domain treatment of transferability. Extensive experiments on $15$ flagship MLLMs across $7$ vendors show that FRA-Attack achieves superior cross-model transferability, particularly with state-of-the-art performance on GPT-5.4, Claude-Opus-4.6 and Gemini-3-flash.

Dataset distillation (DD) compresses a large training set into a small synthetic set for efficient training, but most DD methods optimize only clean accuracy and leave robustness uncontrolled. Recent robust DD methods improve robustness, yet they often suffer from a poor accuracy-robustness trade-off because they (i) treat all adversarially perturbed examples uniformly, despite robust risk being dominated by near-zero robust margins, and (ii) do not explicitly increase inter-class separation in the decision boundary where attacks concentrate. We present Contrastive Curriculum for Robust Dataset Distillation (C$^2$R), a framework that couples an attack-aware curriculum with a contrastive robustness objective. From a robust-margin perspective, we derive a perturbation score that approximates each sample's robust hinge, enabling a curriculum that prioritizes the smallest-margin adversaries that most directly drive robust error. In parallel, a class-balanced contrastive robustness loss enforces adversarial invariance while explicitly widening boundary separation across classes. Experiments on CIFAR-10/100, Tiny-ImageNet, and multiple ImageNet-1K subsets under six attacks show that C$^2$R achieves the best robust accuracy, outperforming prior robust DD by $2.8$% on average.

Gradient-based attacks are important methods for evaluating model robustness. However, since the proposal of APGD, it has been difficult for such methods to achieve significant breakthroughs. To achieve such an effect, we first analyze the issue of "high-loss non-adversarial examples" that degrades attack performance in previous methods, and prove that this issue arises from inappropriate objectives for adversarial example generation. Subsequently, we reconstruct the objective as "maximizing the difference between the non-ground-truth label probability upper bound and the ground-truth label probability", and proposes a novel and powerful gradient-based attack method named Sequential Difference Maximization (SDM). SDM establishes a three-layer optimization framework of "cycle-stage-step". It adopts the negative probability loss function and the Directional Probability Difference Ratio (DPDR) loss function in the initial and subsequent optimization stages, respectively, and approaches the ideal objective of adversarial example generation via stage-wise sequential optimization. Experiments demonstrate that compared with previous state-of-the-art methods, SDM not only achieves stronger attack performance but also exhibits superior cost-effectiveness. The code is available at https://github.com/X-L-Liu/ICML-SDM.

Recently, pre-trained encoders have gained widespread use due to their strong capability in representation extraction. However, they are vulnerable to downstream-agnostic attacks (DAAs). Existing DAA methods operate under a permissive threat model, where an attack is successful if the generated downstream-agnostic adversarial examples (DAEs) change the original prediction, without requiring a specific target. In this paper, we propose a Targeted DAA (TDAA) method under a stricter threat model requiring the attack to be both targeted and downstream-agnostic. Since the downstream task is unknown and encoders do not directly produce predictions, achieving a targeted attack is particularly challenging. To address this, we introduce a novel component termed the 'threat image', pre-selected by the attacker as the target. Specifically, a generator is designed to produce example-specific adversarial perturbations that compel the victim encoder to output identical features for both the DAEs and the threat image. Unlike previous DAA methods that generate a single shared perturbation for all samples, which often fails due to image diversity, our method adopts an example-specific paradigm. This generates tailored perturbations for each image to ensure a high attack success rate and invisibility. By leveraging the threat image as a feature-level anchor, our method builds a task-agnostic bridge to reveal the vulnerabilities of the victim encoder. Extensive experiments on 10 self-supervised methods across 3 benchmark datasets demonstrate the effectiveness of our approach and reveal the pronounced vulnerability of pre-trained encoders. The code will be made publicly available after the review period.

Unified autoregressive models (UAMs) are transformer models that generate text as well as image tokens within a single autoregressive pass. Shared parameters and a multimodal vocabulary simplify the training pipeline and facilitate flexible multimodal generation, yet might introduce new vulnerabilities. In particular, we are the first to show that this unified architecture enables multimodal backdoor attacks, where a trigger can propagate malicious effects across multiple output modalities. Specifically, we present the Token by Token Backdoor Attack (ToBAC), the first backdoor attack targeting UAMs, exploring both data-based and model-based poisoning strategies. We demonstrate that innocuous characters or even common words can be transformed into triggers that elicit harmful behavior in autoregressive image generation. ToBAC can jointly manipulate visual outputs and accompanying text, increasing the perceived authenticity of fabricated content. With model access, ToBAC enables attacks on the unified Liquid model in which a subtle word (e.g., ``cool'') induces modality-aligned brand promotion or ideological influence in 55% of generations. Without model access, ToBAC can be induced through data poisoning, achieving an average success rate of 63.1% against JanusPro.

Existing Multivariate Time Series Anomaly Detection (MTSAD) frameworks increasingly rely on integrating Graph Neural Networks (GNNs) with sequence models to capture complex spatio-temporal dependencies. However, less attention is paid to the spatial over-generalization problem, where unconstrained structural modeling indiscriminately reconstructs anomalies, inevitably degrading detection recall. To tackle this problem, we propose a novel framework that unifies spatio-temporal modeling through a joint prior-observation adversarial learning paradigm. In the spatial dimension, the model alternately learns adjacency matrices as structural prior and models the association discrepancy between prior and data-driven observation in a minimax manner during training. Such adversarial optimization not only improves the model sensitivity for time-wise detection, but also enables the model to localize anomalies to specific channels. To systematically evaluate this anomaly localization capability, we further construct a synthetic benchmark equipped with precise channel-wise annotations. Extensive experiments across public datasets and our dedicated benchmark demonstrate that the proposed framework establishes a new state-of-the-art in both time-wise detection and spatial localization tasks. Our code, pre-trained models, and benchmark are publicly available at https://github.com/anocodetest1/POST.

Arabic handwriting recognition (AHR) has made significant progress with deep learning models. AHR research has largely focused on performance, with security receiving little attention. This study provides what appears to be a new line of inquiry by demonstrating the vulnerability of high-performing models to adversarial black-box attacks. The focus on black-box attacks reflects real-world scenarios where the attacker has no prior knowledge of the model architecture. Extensive experiments were conducted on two benchmark AHR datasets containing Arabic handwritten Characters. Results demonstrated the effectiveness of the attacks, with the Pixle attack achieving an attack success rate of 99-100% on most models. Other, less aggressive attacks achieved success rates of 50-96% across most experiments. Despite the higher attack success rate, the attacks maintain the structural integrity of the characters, rendering them almost imperceptible to the human eye. The findings indicate the higher vulnerability of the studied models to adversarial manipulation. This underscores the need to strengthen efforts to secure these models and ensure their reliability in AHR real-world applications.

Cooperation is central to multi-agent reinforcement learning (MARL), yet learned coordination can be fragile when external perturbations disrupt inter-agent interactions. Prior robust MARL methods have primarily considered value-oriented attacks, leaving a gap in robustness when interaction structures themselves are corrupted. In this paper, we propose an interaction-breaking adversarial learning (IBAL) framework that takes an information-theoretic view to construct attacks that impede coordination by perturbing agents' observations and actions, and trains agents to perform reliably under such disruptions. Empirically, our approach improves robustness over existing robust MARL baselines across diverse attack settings and yields stronger performance even under agent-missing scenarios. Our code is available at https://sunwoolee0504.github.io/IBAL.

Infrared object detection is crucial for perception in autonomous driving and surveillance but remains vulnerable to physical adversarial attacks. Unlike in the RGB domain, where attacks rely on color texture, infrared attacks must manipulate thermal signatures, making the geometry shape of heat-blocking materials the primary adversarial information carrier. Current shape-based methods suffer from a fundamental trade-off between representational capability and optimization power, limiting their attack effectiveness.In this work, we overcome this dilemma by introducing learnable Fourier shapes to the infrared domain. We utilize an end-to-end differentiable framework where a compact set of Fourier coefficients, defining the shape boundary, is analytically mapped to a pixel-space mask via the winding number theorem. This enables efficient gradient-based optimization to generate potent shapes that cause human targets to evade detection. Extensive digital and physical experiments provide a comprehensive evaluation and validate our superior performance. Our resulting physical patch achieves striking robustness, successfully evading detectors across diverse distances, angles, poses, and individuals, and achieves over 88% attack success rate at distances greater than 25m (conf.=0.5). Code is available at https://github.com/Yongyx99/Fourier-shape-attack.

Recent Singing Voice Synthesis (SVS) advances enable highly realistic but potentially malicious AI covers, making singing voice deepfake detection (SVDD) crucial. Self-Supervised Learning (SSL)-based detectors achieve state-of-the-art performance by fine-tuning speech SSL backbones to capture singing-specific spoof artifacts. Existing adversarial attacks often fail against SSL-SVDD, creating a false impression of inherent robustness. We reveal this stems from two challenges. First, at the objective level, attacks optimize cross-entropy on local surrogates, crossing surrogate-specific boundaries rather than suppressing shared spoof evidence. Second, at the method level, attacks follow the surrogate's dominant gradient direction. In SSL-SVDD, this aligns with fine-tuned artifact-sensitive directions, limiting transferability to unseen detectors - a geometric failure we term the Linearity Trap. To properly evaluate robustness, we propose MARS (Meta-Adversarial Regression of Semantics), a transfer-based black-box framework tailored to SSL-SVDD. Structurally, MARS shifts to hypothesis-evidence manipulation by constructing a natural semantic anchor from the pre-trained SSL space and an artifact anchor from the fine-tuned space. Algorithmically, MARS escapes the Linearity Trap via bi-level optimization: the inner stage induces tangential exploration, while the outer stage guides the audio toward the natural semantic manifold. Experiments on the CtrSVDD benchmark show MARS improves Attack Success Rate (ASR) in in-distribution transfer (13%), out-of-distribution transfer (10%), and cross-task evaluation (36%), highlighting the urgent need for robust SVDD systems.

Physical adversarial attacks often overfit single surrogate models and optimization objectives. While ensemble attacks can mitigate this, existing methods struggle with severe gradient conflicts within restricted physical texture spaces, significantly degrading cross-model transferability. To bridge this gap, this paper proposes a Joint Multi-Objective and Multi-Model Optimization Framework (JMOF) that leverages quantitative similarity analysis to select the optimal surrogate model ensemble. Within JMOF, a dual-level mechanism jointly suppresses prediction outputs and flattens intermediate feature distributions, balancing attack efficiency with deep generalization. Additionally, an Orthogonal Gradient Alignment (OGA) strategy resolves cross-model gradient conflicts, transforming mutually repulsive gradients into synergistic optimization directions. Extensive simulated and real-world experiments demonstrate that JMOF outperforms state-of-the-art baselines against diverse black-box detectors. Crucially, JMOF exhibits substantial cross-vision-task generalization, generating attacks capable of simultaneously deceiving object detection and semantic segmentation or monocular depth estimation models. This research advances the generalization limits of physical adversarial attacks, providing a robust framework for evaluating visual AI vulnerabilities in real-world deployments.

Post-hoc saliency methods are widely used to interpret deep neural networks, but their faithfulness is difficult to evaluate reliably. Existing evaluations mask features according to saliency-induced feature ordering and measure performance degradation, but this degradation can be confounded by the masking operator: zero masking may create out-of-distribution artifacts, while interpolation-based masking may preserve residual predictive information. We propose Adversarial Information Masking (AIM), a saliency-guided adversarial feature replacement framework for evaluating both saliency-map faithfulness and masking-operator reliability. AIM replaces selected features with values from an adversarial counterpart of the input and compares degradation under complementary masking orders. We assess reliability using random-attribution bias and stability of explanation-method faithfulness rankings. Experiments on image, audio, and EEG tasks suggest that AIM reduces masking-induced bias compared with zero and interpolation-based masking, while revealing modality-dependent differences between signed and unsigned attributions.

Recent advances in molecular representation integrates molecular topological and visual modalities, opening new avenues for precise Molecular Relational Learning (MRL). Existing MRL methods focus on intra-domain modeling, and their inherent domain-closed effect limits applicability to molecular science, particularly in elucidating cross-domain interaction mechanisms. Consequently, the imperative for Cross-Domain Molecular Relational Learning has become increasingly pressing. Benefiting from structure-activity analysis, we propose the Domain Adversarial Training Network with Structural-Semantic Transfer Discrepancy (DisTrans) to optimize cross-domain adaptive representation for molecular structures and visual images. 1) We employ the gradient reversal strategy based on substructure topological discrepancies between domains to learn the domain dependence of molecular structures. This strategy guides the model to adapt to the structural adjacency patterns in the target domain, generating domain-separable structural representations. 2) We apply the cross-domain representation guidance mechanism to align the functional-group semantic information between the source and target domains, learning cross-domain consistency information. The experimental results in two typical cross-domain strategies demonstrate that DisTrans outperforms 16 baseline methods, maintaining satisfactory performance even under pronounced inter-domain discrepancy.

Physics-informed neural networks (PINNs) are powerful surrogates for differential equations but are notoriously difficult to train due to spectral bias, stiffness, and poor accuracy on high-frequency or multiscale solutions. Adversarial training based on generative adversarial networks (GANs) has recently gained surprisingly strong empirical results in improving training, but the underlying mechanisms remain elusive. To this end, we propose a new analysis framework for adversarially trained PINNs, based on the key observation of how the discriminator in GANs can influence the training dynamics of PINNs. The framework first provides a much needed theoretical grounding to why and when adversarial training is effective in PINNs, then presents a unified analysis of GANs variants in such training, and finally leads to a new, practical, efficient training algorithm for PINNs. Empirical results demonstrate that our method can significantly reduce the pathology of PINNs training, thereby providing better models with superior performances, often several magnitudes more accurate than alternative methods.