Adversarial Robustness

Artificial neural networks (NNs) and machine learning (ML) algorithms are poorly understood from a theoretical perspective, which makes it difficult to fully realize their potential and overcome their weaknesses. For instance, ML algorithms train NN weights by moving them along a low-dimensional subspace of their allowed values, but this implicitly low-dimensional learning structure is not properly exploited to improve training because its nature is not well understood. Moreover, trained NNs are easily confused by pervasive adversarial attacks whose theoretical underpinnings are still unclear. This thesis aims to improve our theoretical understanding of NNs and ML, with a particular focus on adversarial attacks and implicitly low-dimensional learning. For this purpose, we use mathematical tools from statistical mechanics to study different types of NNs and ways in which they can fit the data. In particular, we study two classes of models that fit the data with various degrees of learning and memorization: dense associative memory (DAM) and restricted Boltzmann machines (RBM). In the process, we investigate connections between different versions of these models that are useful to make analytical investigations more efficient.

UTMOS has become one of the most commonly used deep neural network-based speech quality assessment (SQA) metrics in speech processing research. In this paper, we attack UTMOS to probe its robustness. Starting from high-quality speech samples, we optimize the input in two directions: a score-preserving attack, which degrades perceived quality while maintaining the predicted score, and a quality-preserving attack, which lowers the predicted score while maintaining perceived quality. We consider three input spaces: raw waveform, mel spectrogram with a HiFi-GAN vocoder, and the latent space of EnCodec, a neural audio codec. Experimental results show that score-preserving attacks are effective against UTMOS. Although perfect quality-preserving attacks are more difficult, optimization in the EnCodec latent space provides the best chance of success. These results reveal failure modes of UTMOS and highlight the importance of robustness analysis for DNN-based SQA metrics.

Generative models have recently seen rapid adoption in End-to-End (E2E) autonomous driving (AD), with diffusion-based denoising and vocabulary-based retrieval becoming the dominant trajectory-decoding paradigms. Despite their architectural diversity, current generative AD planners share a common inference pattern: a fixed set of candidate trajectories (anchors, vocabulary entries, or proposal queries) is scored by one or more learned heads conditioned on the Bird's-Eye-View (BEV) features, and the highest-scored candidate is returned as the final trajectory. Under this design, the scoring head is the only barrier between perception and the motion command, and its decision margins between competing candidates are often small. We introduce Derail, an adversarial framework that exploits this scoring-head attack surface. Evaluated on various generative planners, Derail flips the trajectory selection from a safe to an unsafe candidate, with score drops of $39$--$80\%$ and collision rates of up to $50\%$, consistently outperforming generic loss-maximization and feature-divergence attacks. Our analysis suggests that safety-violating objectives govern attack effectiveness against generative AD planners, and that the scoring-head inference pattern itself is a recurring attack surface worth explicit defensive consideration.

The structural vulnerabilities of point cloud-based 3D object detectors remain poorly understood. Prior work has studied adversarial robustness primarily on isolated 3D object models, while recent LiDAR spoofing attacks target richer and more realistic driving scenes but focus mainly on physical realizability rather than understanding detector behavior or attack efficiency. In this work, we investigate how LiDAR-based detectors rely on spatial evidence in complex scenes and whether these reliance patterns can be exploited to induce failures more efficiently. To this end, we propose an explainability-guided adversarial analysis methodology. We introduce the Saliency-LiDAR (SALL) method, which aggregates Integrated Gradient attributions across scenes to produce universal saliency maps for LiDAR-based 3D object detectors. Guided by these maps, we design the Explainability-aware Frustum Attack (EFA), which selectively perturbs only the most influential frustums rather than uniformly attacking entire object regions. Experiments on KITTI and nuScenes, across detectors such as PointPillars and SECOND, show that EFA reduces detection recall by more than 15 percentage points while requiring 25-50% fewer perturbed frustums than the state-of-the-art non-saliency-aware baseline. These findings reveal that modern 3D detectors concentrate discriminative evidence in a small subset of spatial regions, exposing a structural robustness vulnerability in current LiDAR perception systems. Our code is released at https://github.com/SecMindLab/Saliency_LiDAR.

Deep Reinforcement Learning (DRL) has achieved significant success in robotics and autonomous systems, yet remains vulnerable to adversarial perturbations that can severely degrade performance. Research in adversarial reinforcement learning is often limited by fragmented implementations, inconsistent evaluation protocols, and poor reproducibility. To address these challenges, we present RoAd-RL, an open-source benchmarking framework that provides unified abstractions for policies, attacks, defenses, and robustness metrics, together with reproducible evaluation pipelines and seamless integration with Stable-Baselines3 and Gymnasium. We evaluate DQN, PPO, and SAC agents in LunarLander and Highway-v0 under 192 attack-defense configurations. Results reveal substantial variations in robustness across environments and show that some commonly used defenses can be more detrimental than the attacks they aim to mitigate, while temporal smoothing consistently achieves strong performance. RoAd-RL establishes a standardized benchmark for adversarial reinforcement learning research and is publicly available at https://pypi.org/project/road-rl.

Continual learning (CL), where a model is trained on a sequence of data tasks, is increasingly being adopted across key fields such as large language models and image recognition, yet it remains highly vulnerable to data poisoning that triggers learning divergence or severe excess risk. Despite these threats, a principled theoretical foundation in CL for understanding attack and defense remains lacking. In this paper, we develop a theoretical framework to analyze strategic attacks and defenses in regularization-based CL, a cornerstone of recent CL theory. By framing the adversary-defender interaction as an online zero-sum game, we first establish a fundamental performance limit: no defense succeeds when an adversary poisons a linear proportion of tasks by injecting unbounded noise or pattern shifts in regularization-based CL. We then analyze two possibly defensible scenarios: infrequent attacks and bounded noise per attack. For the former regime, we propose a task-to-task verification mechanism to detect data poisoning and reduce cumulative bias for learning convergence. For the latter regime, we derive a robust defense that minimizes the model's sensitivity to poisoned features, provably accelerating the convergence rate. Extensive experiments on realistic tasks further validate our theoretical results.

Text-to-image diffusion models remain vulnerable to adversarial prompts that elicit disallowed content, motivating reliable inference-time controls. A popular approach is negative guidance, which subtracts a negative prompt direction with a fixed weight. However, it often forces a safety-fidelity trade-off, causing artifacts or prompt drift when over-applied and failing under attacks when under-applied. Dynamic variants reweight guidance using posterior-odds signals, which can be brittle for open-vocabulary compositional prompts, while lightweight similarity-based methods ignore the evolving image evidence along the denoising trajectory. We introduce Concept Removal Guidance (CRG), a training-free method that estimates unwanted-concept presence at each diffusion step from the model's noise predictions, and adaptively calibrates negative guidance via a closed-form constrained update enforcing a target presence threshold while minimally perturbing the conditional trajectory. Across red-teaming benchmarks, CRG reduces attack success rates while preserving benign fidelity, and extends to additional suppression targets such as artist style and violence without fine-tuning or external classifiers.

We present Proteus, a framework developed at Resemble AI for automated robustness testing of our audio deepfake detection system. Given a detector, Proteus systematically searches over sequences of everyday audio transformations (codec transcoding, additive noise, reverberation, dynamic-range compression, and VoIP simulation) to find combinations that fool the detector while preserving speech quality. We propose two complementary search strategies: (1) a breadth-first search that exhaustively maps augmentation effectiveness across the parameter space, and (2) a Q-learning agent designed to efficiently discover deeper attack chains by exploiting structural patterns in the BFS data. We report findings from continuous deployment of Proteus against our production detector, showing that specific augmentation chains can reliably flip detection verdicts while preserving speech intelligibility and speaker identity. We discuss how these findings are used to harden the detector through targeted retraining.

Heterogeneous graph neural networks (HGNNs) perform well on graphs with multiple node and relation types, but their robustness to attacks with restricted structural access remains insufficiently understood. We propose Blackknife, a query-free, strict black-box evasion attack that observes only the one-hop structures of the attacked nodes, their ground-truth labels, and raw features within the permitted action space. No victim-model outputs are used to train the surrogate, generate perturbations, or select attacks.Blackknife trains a shared surrogate using only observable edges and relation types, then optimizes structural perturbations through continuous relaxation and projected gradient ascent. Each rewiring operation deletes one edge and adds another of the same relation type, preserving the target node's total degree and degree for each relation type. A cosine-similarity penalty and an adaptive structural classifier further regularize the perturbations through bilevel optimization. Experiments on different datasets demonstrate competitive attack success rates across five victim architectures and three defense settings. The results expose HGNN vulnerabilities under jointly restricted graph access and query-free attack generation.

The robustness of trajectory prediction models is crucial for developing safe autonomous driving systems. Adversarial attacks on trajectory prediction can significantly impair the accuracy of predicted trajectories, leading to hazardous driving behaviors. While heuristic defense strategies have been implemented to enhance the robustness of trajectory prediction models, these measures often fail against more sophisticated, targeted adversarial attacks. Hence, there is a pressing need to establish verifiable safety assurances for trajectory prediction models. In this paper, we extend the traditional Randomized Smoothing framework to "TrajRS", which provides a certified robust radius for smoothed trajectory predictors. We clarify and expand the formal definitions of robustness in trajectory prediction and tailor the practical TrajRS scheme specifically to "robustness for the optimal prediction" and "robustness for all possible predictions". An extensive set of experiments demonstrates that TrajRS effectively achieves robustness certification for all smoothed pedestrian trajectory predictors in this work.

The existence of adversarial attacks is often attributed to the presence of non-robust features in neural networks. While prior defenses reduce their impact via pruning, masking, or feature recalibration, we instead propose to jointly learn to amplify and attenuate these signals through a simple activation scaling mechanism. To this end, we introduce Activation Amplification and Attenuation (A3), a lightweight plug-in module that enhances adversarial robustness with minimal modifications of the activations. A3 dynamically rescales the activations using a learnable mask and a scaling factor derived from the original activation magnitudes. The influence of adversarial perturbations can be amplified or attenuated using the same learnable parameters by simply flipping the sign of the scaling operation. The amplified signals serve as negative references to construct novel contrastive and ranking loss functions. Experimental analysis shows that learning to degrade the predictions in amplification mode simultaneously improves adversarial robustness in attenuation mode. Moreover, A3 relies on only a small number of learnable parameters, with most of its behavior being determined by the scaling mechanism rather than additional network capacity. Extensive experiments demonstrate that integrating A3 into different backbones, datasets, and training methods consistently improves adversarial robustness while introducing negligible computational and memory overhead compared to existing plug-in modules. Code is available at: https://github.com/tgoncalv/A3.

Pervasive data contamination -- stemming from measurement errors, outliers, or adversarial corruption -- has motivated the development of robust statistical methods. In this context, we propose a two-stage Adversarial Contamination-resistant Iterative Hard Thresholding (AC-IHT) algorithm for high-dimensional regression with contamination. Our nonconvex algorithm achieves minimax near-optimal (up to logarithmic terms) estimation by iteratively updating the coefficient vector and the contamination vector with different thresholding scales. We further demonstrate that our AC-IHT estimator is signal-adaptive: under proper signal conditions, it adaptively attains a sharper estimation rate and more accurate support recovery. Moreover, it enjoys the strong oracle property, laying a theoretical foundation for asymptotic inference. Numerical experiments confirm its superior finite-sample performance. Finally, we discuss theoretical extensions of the proposed procedure to generalized linear models and to heavy-tailed noise settings.

Adversarial evaluation of AI systems has matured along four largely disconnected tracks: diffusion-based attacks on text and large language models (LLMs), diffusion-based attacks on image classifiers, jailbreak pipelines against vision-language models, and diffusion-based input purification defenses. Each has developed its own vocabulary, threat models, and benchmarks, with denoising diffusion models emerging as a shared generative mechanism whose recipes are now actively ported between communities. This survey performs an information-fusion exercise at the meta-research level: we integrate these four tracks into a single conceptual framework with a unified taxonomy, evaluation criteria, and research agenda, focusing on the LLM-side slice. We catalog fifty published papers across four scope areas (text/LLM, image classifier, vision-language model, defense), plus four diffusion-LLM-as-victim entries and ten non-diffusion baselines against which any new attack must be compared. We propose a six-class taxonomy of diffusion roles in adversarial pipelines, augmented by a threat-model axis recording attacker knowledge, query budget, and target accessibility, and apply a five-dimension framework (attack success rate, transferability, query budget, perplexity, defense-evasion) uniformly across modalities. The review adopts a dual attacker-defender perspective: alongside the attack catalog we cover four diffusion-based defenses that form the natural evaluation backdrop for new attacks. Our critical analysis identifies five recurring weaknesses of the current LLM-side literature, and we close with a research agenda of open questions and concrete experimental designs. The companion catalog and spreadsheet are released with the paper. We are explicit that this is a narrative review with quality assessment, not a PRISMA-compliant systematic review, and discuss the implications for replication.

Stochastic quantum neural networks (SQNNs) encode neuronal activations as qubits, synaptic topology as entanglement, and neural noise through a Lindblad master equation. A recent conference study applied a ring-entangled SQNN to collaborative intrusion detection and reached three conclusions: ring entanglement is essential for non-local anomaly detection; an adversarial-resilience bound holds but is conservative; and the depolarising channel fails to act as a dropout-style regulariser, behaving instead as output noise. It left open whether a per-gate stochastic deactivation (``true quantum dropout'') could regularise where the depolarising channel could not, and whether the loose robustness bound could be replaced by a predictive theory. This paper resolves both and extends the framework to real data and to neutral-atom hardware. We give an $N$-qubit formulation through the stochastic master equation and its vectorised Liouvillian, and prove a decoherence-contraction theorem: a depolarising channel of strength $γ$ over $L$ entangling layers contracts every weight-$w$ Pauli read-out by a factor $(1-4γ/3)^{wL}$ (for the weight-$1$ read-out used here, $(1-4γ/3)^{L}$); building on the general noise-as-defence result of Du et al., we make this quantitative and operational for intrusion detection. On the real NSL-KDD dataset under white-box FGSM and PGD attacks, a depolarising SQNN trained with the channel is, over seven seeds under strong $\ell_\infty$/$\ell_2$ attacks, significantly more robust than the noiseless circuit ($\ell_\infty$ PGD-$20$, $p=0.04$, large effect) and, critically, never suffers the catastrophic robustness collapse that the noiseless model and gradient-trained classical detectors (which fall from $95\%$ to $47\%$) do, cutting robustness variance roughly twofold; we show this robustness arises from a noise-reshaped training boundary rather than from attack-time gradient contraction. For generalisation, we derive an adaptive-penalty formula showing that per-gate dropout implements a curvature-weighted $L_2$ penalty $\tfrac{p(1-p)}{2}\sumθ^2\partial^2_θL$ in weight space, maximised at $p=1/2$, whereas depolarising noise implements an output-space penalty. A $30$-seed study confirms the formula's quantitative prediction: both mechanisms reduce the train-test gap by a small but statistically significant margin ($\approx\!0.01$; $p<10^{-4}$ and $p=0.004$), are statistically indistinguishable from each other, and the effect is concentrated where overfitting is largest; increasing the dropout rate past $1/2$ does not help, as the formula predicts. The single-seed dichotomy of prior work does not survive replication. We close with a neutral-atom realisation and a feasibility-by-$N$ analysis.

Vision-language models (VLMs) achieve strong zero-shot recognition, but they remain highly vulnerable to adversarial perturbations. Recent test-time adaptations improve robustness without retraining, but they do not directly adapt the corrupted visual representation itself. Prompt-based methods adapt the learnable text prompts, while input-space methods optimize pixels or padding at test time. These approaches can improve predictions, but they do so through an indirect and expensive optimization path. We propose Test-time Visual Subspace Steering (T-VSS), a lightweight defense that performs test-time adaptation directly in the visual feature space. T-VSS first builds a sample-specific low-rank subspace from multi-view feature residuals anchored at the attacked image. It then learns a shared feature correction within this subspace using reliability-weighted entropy minimization. By constraining adaptation to a compact visual geometry, T-VSS steers attacked features toward more stable and discriminative predictions while avoiding noisy full-space updates. Experiments on fine-grained, ImageNet, and ImageNet-OOD benchmarks show that T-VSS improves adversarial robustness while maintaining competitive clean accuracy and better efficiency than prior test-time adaptations.

Radio frequency jamming poses a critical threat to the availability of wireless Industrial Internet of Things (IIoT) networks. Existing detection and classification techniques are poorly suited to this setting: coarse signal-strength and cross-layer features lack information richness, while raw I/Q baseband approaches require hardware and throughput that is impractical at the scale of hundred-node IIoT deployments. This paper presents CITADEL, a lightweight two-stage hierarchical pipeline that uses only Channel State Information (CSI) measurements, which are natively available on commodity IIoT devices, to detect and classify jamming attacks including previously unseen ones. While prior work has shown that jamming leaves observable CSI signatures, CITADEL is the first system to translate this insight into an end-to-end pipeline that jointly achieves closed-set classification of known attacks, open-set detection of zero-day attacks, and resistance to adversarial evasion. Evaluated across 6 known attack types and 15 zero-day scenarios, CITADEL achieves 100% known-attack detection and 97.1% zero-day detection at a 0.4% end-to-end false positive rate. Under adversarial evaluation spanning white-box and black-box threat models, gradient-based evasion remains below 2% across all tested perturbation budgets and the strongest published CSI attack generator achieves less than 5% average evasion. A systematic comparison against eight baselines confirms that no existing method achieves comparable performance on CSI data across all three axes: detection, generalization, and robustness. The full pipeline completes inference in 14.2 ms at 95.9 mJ on an edge GPU, establishing CITADEL as a practical solution for large-scale IIoT network security.

The proliferation of IoT devices has fueled distributed edge systems to collect vast amounts of sensitive data, creating fertile ground for on-device machine learning applications. While federated learning (FL) mitigates privacy concerns by exchanging model parameters instead of raw data, we identify a critical blind spot in current research. We examine the most commonly used personalized federated learning (PFL) methods, which allow clients to maintain private, personalized models to address data heterogeneity across clients. Through systematic analysis, we reveal that PFL methods exhibit heightened vulnerability to transfer-based adversarial attacks compared to centralized learning paradigms. Wherein, malicious clients can exploit local model knowledge to craft adversarial examples that can compromise peer clients' personalized models. We establish this vulnerability through both theoretical analysis and empirical evaluation across multiple benchmark datasets, demonstrating significant accuracy drops across various PFL methods. To address this challenge, we propose a defense framework combining stochastic input noise, input-scaled trace regularization, and parameter sensitivity maximization to improve FL's robustness. Our findings establish the first systematic study of adversarial threats in PFL systems, providing both diagnostic tools and practical countermeasures.

Input Diversity (DI), a random resize and pad applied at each attack iteration, is a near-default ingredient of transfer-based attacks, widely assumed to improve transferability. We show this assumption is regime-dependent and, for adversarially trained surrogates, often reversed. Holding the attack fixed and varying only the surrogate, raising the DI probability improves transfer from standard surrogates but degrades it from robust ones: the two response curves separate like a pair of scissors, a pattern we call the Scissors Effect. On ImageNet, blind DI costs a robust source 10.3 percentage points of attack success across four architecturally diverse targets; the effect is several times smaller at 32x32. Direct measurement supports a bias-variance account: DI displaces the gradient by a comparable amount on both groups but reduces its variance only where the gradient is noisy, and robust surrogates have little noise left to average away. A gradient-consistency probe, frozen and hashed before the runs, predicts the sign of the effect on seven unseen surrogates, and we report where it fails alongside where it works. The practical consequence holds independently of the mechanism: leaving DI enabled by default understates the attack a robust surrogate can mount, and so overstates the robustness of the model being evaluated. Code: https://github.com/Avalon-S/ScissorsEffect.

Test-time adaptation (TTA) can mitigate domain shift without source data, but it is highly brittle under adversarially contaminated test streams, where corrupted inputs also destabilize online updates. We study robust test-time adaptation (RTTA) in the adversarial-stream setting, which remains comparatively underexplored relative to standard TTA, and propose SAFER (Stochastic Augmentation Framework for Enhanced Robustness), a training-free reliability-guided augmentation wrapper for RTTA. SAFER preserves the wrapped TTA objective while replacing brittle single-view predictions with a reliability-guided pooled predictor. For each test sample, SAFER generates stochastic augmentations and aggregates their predictions through correlation-weighted pooling with outlier detection. We further study an adaptive-mixing extension that improves clean-performance retention by adjusting original-versus-augmentation weighting using feature disagreement signals. We evaluate on PACS, VLCS, and OfficeHome under PGD attacks at various attack rates. Across benchmarks, SAFER improves resilience of TTA methods to adversarial attacks while maintaining competitive clean performance.

The vulnerability of ML models to adversarial examples has recently emerged as a major concern. While adversarial training is one of the most effective countermeasures to this issue, its high computational cost remains an obstacle to practical deployment. Recent progress in reducing this cost has relied, in the case of linear models, on a formal equivalence between the adversarial risk and a simpler form of regularized risk. This enabled significantly more efficient training procedures, which naturally raises the question of whether such an equivalence can be extended beyond linear models. In this work, we formally show that no such equivalence is possible for two-layer networks. Our proofs proceed via a reduction to key properties that fundamentally separate the adversarial risk from any simple regularized risk which would only exhibit a weak form of data dependence. Beyond this setting, we provide empirical evidence on Wide-ResNets indicating that the same type of impossibility persists in deeper and more expressive architectures.

The automated design of agentic systems offers a promising pathway for scaling large language models (LLMs) beyond single-agent reasoning. While prior work has advanced task performance through handcrafted or automatically generated multi-agent workflows, robustness is often treated as an afterthought, leaving systems vulnerable to external adversaries and internal failures. We propose AutoRAS, a framework for the Automated design of Robust Agentic Systems. AutoRAS formulates system design as generating a sequence of symbolic primitives that jointly encode structural connectivity and behavioral actions, and learns to optimize this sequence using execution-derived safety signals and flow-based sequence-level objectives. Extensive experiments show that AutoRAS achieves the best performance in both vanilla and adversarial settings, with the smallest performance degradation under attacks. Further analyses demonstrate strong transferability, stable optimization behavior, stability across primitive sets, and favorable cost trade-offs. Our code is available at this https URL.

Agentic AI systems increasingly rely on language-model components to interpret instructions, process external data, invoke tools, and coordinate with other agents. These capabilities make prompt-injection and jailbreak attacks more consequential, especially as attackers adopt model-guided automation to scale probing, prompt refinement, and response evaluation. This work analyzes the resulting attack-defense setting through a probabilistic model of a target system, its defense mechanism, and the attacker's automated judge. Our analysis shows that conventional detect-and-block defenses can allow attacker success rate (ASR) to approach one as the query budget grows, since predictable refusals provide useful feedback to automated search. We then examine detect-and-misdirect, where detected malicious interactions receive controlled, non-operational responses designed to induce false-positive errors in the attacker's judge. This strategy reduces the positive predictive value of attacker-selected candidates and yields a bounded asymptotic ASR. We evaluate a proof-of-concept realization of this strategy through Contextual Misdirection via Progressive Engagement (CMPE), a lightweight conversational misdirection method designed to replace predictable refusal text with safe but strategically misleading responses in automated jailbreak settings. On jailbreak benchmarks, CMPE reduces estimated ASR upper bounds by up to two orders of magnitude and nearly eliminates verified attack success in end-to-end experiments with PAIR, GPTFuzz, and AutoDAN-Turbo.

Deep Neural Networks DNNs have achieved remarkable accuracy in various tasks including their application in CyberPhysical Systems CPS for detecting False Data Injection Attacks FDIA during critical operations However the unique infrastructure of CPS makes DNNs vulnerable to exploitation by attackers aiming to evade detection Additionally the distinct nature of CPS presents challenges for conventional defense mechanisms against FDIA This paper proposes an innovative defense framework that strengthens DNNs against such attacks by introducing an additional input layer that performs padding in the input samples using pseudofeature values derived from the inputs statistical distribution This padding increases the input dimensionality in a randomized and dataaware manner making adversarial attacks computationally infeasible due to the nontransferable nature of crafted perturbations and the unpredictability of the padded structure Our method is lightweight modelagnostic and requires no modifications to the core architecture making it highly deployable in realworld CPS settings We evaluated our framework on critical power grid applications such as state estimation using the IEEE 14bus 30bus 118bus and 300bus systems Experiments under adversarial settings demonstrate that our padding strategy significantly improves model robustness with negligible impact on performance and effectively mitigates attacks that would otherwise bypass conventional defenses

In this work, we introduce a training procedure for shallow neural networks that promotes robustness against adversarial attacks. We solve a non-convex Lipschitz-regularized training program by introducing a convex restriction that can be efficiently solved to global optimality. Our approach can be employed as a post-processing step by taking a pre-trained network as an initial solution to then solving the convex program whose optimal network is guaranteed to be no worse than the initial one. We illustrate the improvements of our training procedure with experiments using real world datasets for regression tasks under an adversarial setting. We show numerically that solving our proposed convex program yields networks with lower objective values on the Lipschitz-regularized program compared to existing methods. Additionally, we show that on certain datasets, networks obtained using our convex training program are both more accurate and robust with respect to adversarial attacks.

Hardware-enabled monitoring of GPU workloads underpins many proposals for AI compute governance, but if developers can defeat monitoring mechanisms, such schemes are unworkable. We evaluate the adversarial robustness of GPU workload classification using only zero-overhead, privacy-preserving NVML telemetry: content-agnostic signals that observe physical effects of computation without accessing model weights, training data, or hyperparameters. Across 5 rounds of monitor-evader iteration, we evaluate 20 evasion strategy families on 9 GPU models spanning 4 architecture generations. We develop a classifier that achieves 98.2% binary accuracy at identifying training workloads across the whole corpus, and 43-87% accuracy against the most challenging unexpected workloads even when they are adversarially disguised.

Recent empirical work shows that semantically equivalent paraphrases can fool financial sentiment classifiers: although a paraphrase remains close to the original under a strong reference embedding, it may shift the target model's representation enough to change the predicted class. Existing robustness theory either assumes a single-model threat model or focuses mainly on empirical attack algorithms. We develop a continuous local model of semantic paraphrase perturbations that captures this two-model structure. We show that the worst-case local displacement of the target representation, subject to a proxy-model budget, is governed by the largest generalised eigenvalue of a matrix pencil $(A,B)$ constructed from the Jacobians of the two embedding maps. The resulting attackability index $λ^*(x)$ is intrinsic to the local paraphrase geometry and the chosen embedders, yields a closed-form prediction-flip condition for affine readouts, and supports conservative population and finite-sample attackability certificates. For uniform control over classes of affine readouts, we derive a distribution-free VC bound for binary attackability indicators and a scale-sensitive margin bound based on an attackability-adjusted margin that subtracts a local geometric penalty from the standard classifier margin. We also connect the continuous theory to discrete paraphrase search, identify an asymmetry between successful and unsuccessful finite searches, and give a covering condition under which the discrete and continuous settings agree. Finally, we propose an empirical verification framework using soft-token relaxations and generated paraphrase sets to assess the local eigenvalue geometry, prediction-flip condition, and finite-search approximation on a deployed financial-text classifier.

Multi-threat robustness remains a fundamental challenge in deep learning. Although joint adversarial training (JAT) is widely adopted, it suffers from negative transfer under heterogeneous threats, particularly between $\ell_p$-bounded and semantic attacks. Through first-order gradient analysis, we formalize this as gradient incompatibility and theoretically establish the necessity of decoupled optimization. We further reveal that these conflicting threats exhibit separable spectral characteristics in the frequency domain. Motivated by this observation, we propose Threat-aware Frequency Decoupling (TaFD), a two-stage defense framework that reformulates JAT as a frequency-domain divide-and-conquer paradigm. TaFD first discovers latent threat domains via unsupervised clustering of attack spectral prototypes and trains a lightweight classifier for inference-time threat domain identification. Conditioned on the prediction, TaFD employs a Frequency-Conditional Convolution that learns threat-domain-specific spectral masks and routes each sample to the corresponding expert, enforcing structural parameter separation and alleviating optimization conflicts. We validate TaFD on three representative image-classification benchmarks (CIFAR-10, CIFAR-100, and Tiny-ImageNet) and on two representative architectures (the convolutional ResNet and the hybrid-transformer MobileViT). Extensive results demonstrate that TaFD achieves more balanced robustness against heterogeneous attacks than existing JAT and frequency-domain baselines, improving average robust accuracy by approximately 11% over the strongest baseline while maintaining leading clean accuracy.

Generating adversarial driving scenarios is critical for evaluating and improving autonomous vehicle decision-making systems in simulation. Recent approaches rely primarily on the prior knowledge of Large Language Models and Vision-Language Models to generate driving scenarios procedurally. We argue that adversarial scenes should be generated based on the failure diagnosis (e.g., indecisiveness, multi-frame inconsistency) of the driving policy to specifically address the policy's weaknesses instead of relying on prior assumptions. In this paper, we propose SPHINX, a closed-loop framework for adversarial scenario synthesis guided by a simple principle: first explain, then explore. Beyond blindly exploring the scenario space, SPHINX leverages explainable artificial intelligence methods to analyze the policy, identifying key visual concepts and their influence on policy outputs, and the uncertainty of the decisions. Given the interpretable evidence extracted from the policy's own decision process, we use a vision language model to rationalize and criticize failure modes of the current policy. These critics are then used to generate targeted adversarial scenarios for policy retraining and improvement. We demonstrate that SPHINX can highlight an interpretable account of policy failures while other adversarial scene generation cannot. Across the evaluated benchmarks and test suites, SPHINX can be applied to diverse state-of-the-art autonomous vehicle architectures and yields consistent robustness improvements over existing scenario-generation methods.

Time-series forecasting models remain vulnerable to gradient-based adversarial attacks while existing defense mechanisms typically incur a trade-off in robustness for bounded response and compute cost. The problem is pronounced in Moving Target Defense where maintaining multiple randomized model instances substantially exacerbates the training overhead. In this work, we introduce MorphStrata, a student generation strategy with selective, layer-specific stochastic noise injection that extends the traditional Morphence defense. MorphStrata uses a Transformer backbone as the teacher and perturbs randomly selected architectural blocks to create structured heterogeneity across student models in response to varied data distributions and threat models. We evaluate against vanilla Transformer and Morphence backbones on a suite of benchmarks including the Jena Climate, Electricity Load Diagrams, and Appliances Energy Prediction using FGSM, BIM and PGD attacks across multiple attack strengths. Across datasets and attack regimes, the proposed ensemble maintains comparable adversarial RMSE. Specifically, for high entropy, periodic datasets as in the case of the AEP data, MorphStrata achieves the lowest RMSE across all attacks and perturbation budgets, improving over the static baseline by up to 24.11% and 97.97% under FGSM and BIM respectively at an epsilon value of 0.5 over 30 randomized trials. Targeting the layers to generate MorphStrata students accounts for less than 1% increase in train-times over the Morphence MTD baseline for most of the experiments, while accounting for double digit gains in adversarial RMSE reduction. We also observe a positive correlation between higher pairwise L2 distance (among generated students) and overall defense effectiveness. In summary, MorphStrata maintains adversarial robustness as an MTD defense at marginal cost deltas when compared to existing baselines.

World models are widely used in robotic and agentic engineering control systems due to their ability to learn latent dynamics for planning and decision-making. As these systems are increasingly deployed in safety-critical settings, understanding their robustness under adversarial conditions has become essential. However, existing evaluations lack a unified benchmark for testing adversarial threats across the policy, value, and latent-dynamics levels of world-model agents. To fill this gap, we present ARB4WM, a unified evaluation framework for pre-deployment robustness and risk assessment of world-model agents under visual perturbations. ARB4WM defines five white-box loss objectives across these three levels and studies their effects when combined with single-step or multi-step perturbation strategies and temporal attack modes, including full-frame, half-sequence, and sparse-frame exposure. Specifically, we evaluate four Dreamer-style agents across 20 tasks from MetaWorld and the DeepMind Control Suite under different loss objectives, perturbation strategies, and temporal attack modes. Results show that attacks targeting value estimation, latent representations, and RSSM dynamics can be as damaging as direct policy disruption, and that early or frequent perturbations are especially harmful, while input-level defenses provide limited recovery under adaptive attacks. These findings suggest that safety, risk, and reliability assessment for world models should cover multiple component-oriented attack objectives and temporal exposure protocols rather than relying solely on action-space robustness. Source code is available at https://github.com/zaoanguai/ARB4WM.