Federated learning (FL) has attracted substantial attention in both academia and industry, yet its practical security posture remains poorly understood. In particular, a large body of poisoning research is evaluated under idealized assumptions about attacker participation, client homogeneity, and success metrics, which can substantially distort how security risks are perceived in deployed FL systems. This paper revisits FL security from a measurement perspective. We systematize three major sources of mismatch between research and practice: unrealistic poisoning threat models, the omission of hybrid heterogeneity, and incomplete metrics that overemphasize peak attack success while ignoring stability and utility cost. To study these gaps, we build TFLlib, a uniform evaluation framework that supports image, text, and tabular FL tasks and re-implements representative poisoning attacks under practical settings. Our empirical study shows that idealized evaluation often overstates security risk. Under practical settings, attack performance becomes markedly more dataset-dependent and unstable, and several attacks that appear consistently strong in idealized FL lose effectiveness or incur clear benign-task degradation once practical constraints are enforced. These findings further show that final-round attack success alone is insufficient for security assessment; practical measurement must jointly consider effectiveness, temporal stability, and collateral utility loss. Overall, this work argues that many conclusions in the FL poisoning literature are not directly transferable to real deployments. By tightening the threat model and using measurement protocols aligned with practice, we provide a more realistic view of the security risks faced by contemporary FL systems and distill concrete guidance for future FL security evaluation. Our code is available at https://github.com/xaddwell/TFLlib

Agent skills are becoming a core abstraction in coding agents, packaging long-form instructions and auxiliary scripts to extend tool-augmented behaviors. This abstraction introduces an under-measured attack surface: skill-based prompt injection, where poisoned skills can steer agents away from user intent and safety policies. In practice, naive injections often fail because the malicious intent is too explicit or drifts too far from the original skill, leading agents to ignore or refuse them; existing attacks are also largely hand-crafted. We propose the first automated framework for stealthy prompt injection tailored to agent skills. The framework forms a closed loop with three agents: an Attack Agent that synthesizes injection skills under explicit stealth constraints, a Code Agent that executes tasks using the injected skills in a realistic tool environment, and an Evaluate Agent that logs action traces (e.g., tool calls and file operations) and verifies whether targeted malicious behaviors occurred. We also propose a malicious payload hiding strategy that conceals adversarial operations in auxiliary scripts while injecting optimized inducement prompts to trigger tool execution. Extensive experiments across diverse coding-agent settings and real-world software engineering tasks show that our method consistently achieves high attack success rates under realistic settings.

Visual token compression is widely used to accelerate large vision-language models (LVLMs) by pruning or merging visual tokens, yet its adversarial robustness remains unexplored. We show that existing encoder-based attacks can substantially overestimate the robustness of compressed LVLMs, due to an optimization-inference mismatch: perturbations are optimized on the full-token representation, while inference is performed through a token-compression bottleneck. To address this gap, we propose the Compression-AliGnEd attack (CAGE), which aligns perturbation optimization with compression inference without assuming access to the deployed compression mechanism or its token budget. CAGE combines (i) expected feature disruption, which concentrates distortion on tokens likely to survive across plausible budgets, and (ii) rank distortion alignment, which actively aligns token distortions with rank scores to promote the retention of highly distorted evidence. Across diverse representative plug-and-play compression mechanisms and datasets, our results show that CAGE consistently achieves lower robust accuracy than the baseline. This work highlights that robustness assessments ignoring compression can be overly optimistic, calling for compression-aware security evaluation and defenses for efficient LVLMs.

LLM-based web agents have become increasingly popular for their utility in daily life and work. However, they exhibit critical vulnerabilities when processing malicious URLs: accepting a disguised malicious URL enables subsequent access to unsafe webpages, which can cause severe damage to service providers and users. Despite this risk, no benchmark currently targets this emerging threat. To address this gap, we propose MalURLBench, the first benchmark for evaluating LLMs' vulnerabilities to malicious URLs. MalURLBench contains 61,845 attack instances spanning 10 real-world scenarios and 7 categories of real malicious websites. Experiments with 12 popular LLMs reveal that existing models struggle to detect elaborately disguised malicious URLs. We further identify and analyze key factors that impact attack success rates and propose URLGuard, a lightweight defense module. We believe this work will provide a foundational resource for advancing the security of web agents. Our code is available at https://github.com/JiangYingEr/MalURLBench.

Annotating medical data for training AI models is often costly and limited due to the shortage of specialists with relevant clinical expertise. This challenge is further compounded by privacy and ethical concerns associated with sensitive patient information. As a result, well-trained medical segmentation models on private datasets constitute valuable intellectual property requiring robust protection mechanisms. Existing model protection techniques primarily focus on classification and generative tasks, while segmentation models-crucial to medical image analysis-remain largely underexplored. In this paper, we propose a novel, stealthy, and harmless method, StealthMark, for verifying the ownership of medical segmentation models under black-box conditions. Our approach subtly modulates model uncertainty without altering the final segmentation outputs, thereby preserving the model's performance. To enable ownership verification, we incorporate model-agnostic explanation methods, e.g. LIME, to extract feature attributions from the model outputs. Under specific triggering conditions, these explanations reveal a distinct and verifiable watermark. We further design the watermark as a QR code to facilitate robust and recognizable ownership claims. We conducted extensive experiments across four medical imaging datasets and five mainstream segmentation models. The results demonstrate the effectiveness, stealthiness, and harmlessness of our method on the original model's segmentation performance. For example, when applied to the SAM model, StealthMark consistently achieved ASR above 95% across various datasets while maintaining less than a 1% drop in Dice and AUC scores, significantly outperforming backdoor-based watermarking methods and highlighting its strong potential for practical deployment. Our implementation code is made available at: https://github.com/Qinkaiyu/StealthMark.

As a pivotal technique for improving the defense of deep models, adversarial robustness transfer via distillation has demonstrated remarkable success in conventional image classification tasks. However, this paradigm encounters critical challenges when applied to vision-language models (VLM) (e.g., CLIP): constructing adversarially robust teacher for large-scale multi-modal models demands prohibitively high computational resources. We bridge this gap by revealing an interesting phenomenon: vanilla CLIP (without adversarial training) exhibits intrinsic defensive capabilities against adversarial examples generated by another CLIP with different architectures. We formally define this as proxy adversarial robustness, and naturally propose a Heterogeneous Proxy Transfer (HPT) framework that establishes cross-architectural robustness distillation channels between CLIP variants, effortlessly enabling the VLM robustness transfer from proxy to target models. Yet, such proxy transfer paradigm easily induces severe overfitting, leading to a sharp degradation in zero-shot natural generalization. To resolve that, we design Generalization-Pivot Decoupling (GPD) by leveraging the difference in learning rate scheduling. This decouples the proxy transfer process into a generalization-anchored warm-up that maintains generalization and a generalization-pulled HPT that promotes adversarial robustness, to achieve an equilibrium between natural generalization and adversarial robustness. Extensive experiments on 15 zero-shot datasets demonstrate the effectiveness of our HPT-GPD method. The code is available at the website of github.com/fxw13/HPT-GPD.

Single-pass hallucination detectors rely on internal telemetry (e.g., uncertainty, hidden-state geometry, and attention) of large language models, implicitly assuming hallucinations leave separable traces in these signals. We study a white-box, model-side adversary that fine-tunes lightweight LoRA adapters on the model while keeping the detector fixed, and introduce CORVUS, an efficient red-teaming procedure that learns to camouflage detector-visible telemetry under teacher forcing, including an embedding-space FGSM attention stress test. Trained on 1,000 out-of-distribution Alpaca instructions (<0.5% trainable parameters), CORVUS transfers to FAVA-Annotation across Llama-2, Vicuna, Llama-3, and Qwen2.5, and degrades both training-free detectors (e.g., LLM-Check) and probe-based detectors (e.g., SEP, ICR-probe), motivating adversary-aware auditing that incorporates external grounding or cross-model evidence.

The widespread use of Vision Language Models (VLMs, e.g. CLIP) has raised concerns about their vulnerability to sophisticated and imperceptible adversarial attacks. These attacks could compromise model performance and system security in cross-modal tasks. To address this challenge, three main defense paradigms have been proposed: Training-time Defense, Test-time Adaptation Defense, and Training-free Defense. Training-time Defense involves modifying the training process, typically through adversarial fine-tuning to improve the robustness to adversarial examples. While effective, this approach requires substantial computational resources and may not generalize across all adversarial attacks. Test-time Adaptation Defense focuses on adapting the model at inference time by updating its parameters to handle unlabeled adversarial examples, offering flexibility but often at the cost of increased complexity and computational overhead. Training-free Defense avoids modifying the model itself, instead focusing on altering the adversarial inputs or their feature embeddings, which enforces input perturbations to mitigate the impact of attacks without additional training. This survey reviews the latest advancements in adversarial defense strategies for VLMs, highlighting the strengths and limitations of such approaches and discussing ongoing challenges in enhancing the robustness of VLMs.

Image forgery has become a critical threat with the rapid proliferation of AI-based generation tools, which make it increasingly easy to synthesize realistic but fraudulent facial content. Existing detection methods achieve near-perfect performance when training and testing are conducted within the same domain, yet their effectiveness deteriorates substantially in crossdomain scenarios. This limitation is problematic, as new forgery techniques continuously emerge and detectors must remain reliable against unseen manipulations. To address this challenge, we propose the Real-Centered Detection Network (RCDN), a frequency spatial convolutional neural networks(CNN) framework with an Xception backbone that anchors its representation space around authentic facial images. Instead of modeling the diverse and evolving patterns of forgeries, RCDN emphasizes the consistency of real images, leveraging a dual-branch architecture and a real centered loss design to enhance robustness under distribution shifts. Extensive experiments on the DiFF dataset, focusing on three representative forgery types (FE, I2I, T2I), demonstrate that RCDN achieves both state-of-the-art in-domain accuracy and significantly stronger cross-domain generalization. Notably, RCDN reduces the generalization gap compared to leading baselines and achieves the highest cross/in-domain stability ratio, highlighting its potential as a practical solution for defending against evolving and unseen image forgery techniques.

This paper reveals that LLM-powered agents exhibit not only demographic bias (e.g., gender, religion) but also intergroup bias under minimal "us" versus "them" cues. When such group boundaries align with the agent-human divide, a new bias risk emerges: agents may treat other AI agents as the ingroup and humans as the outgroup. To examine this risk, we conduct a controlled multi-agent social simulation and find that agents display consistent intergroup bias in an all-agent setting. More critically, this bias persists even in human-facing interactions when agents are uncertain about whether the counterpart is truly human, revealing a belief-dependent fragility in bias suppression toward humans. Motivated by this observation, we identify a new attack surface rooted in identity beliefs and formalize a Belief Poisoning Attack (BPA) that can manipulate agent identity beliefs and induce outgroup bias toward humans. Extensive experiments demonstrate both the prevalence of agent intergroup bias and the severity of BPA across settings, while also showing that our proposed defenses can mitigate the risk. These findings are expected to inform safer agent design and motivate more robust safeguards for human-facing agents.

The increasing deployment of large language models (LLMs) in safety-critical applications raises fundamental challenges in systematically evaluating robustness against adversarial behaviors. Existing red-teaming practices are largely manual and expert-driven, which limits scalability, reproducibility, and coverage in high-dimensional prompt spaces. We formulate automated LLM red-teaming as a structured adversarial search problem and propose a learning-driven framework for scalable vulnerability discovery. The approach combines meta-prompt-guided adversarial prompt generation with a hierarchical execution and detection pipeline, enabling standardized evaluation across six representative threat categories, including reward hacking, deceptive alignment, data exfiltration, sandbagging, inappropriate tool use, and chain-of-thought manipulation. Extensive experiments on GPT-OSS-20B identify 47 vulnerabilities, including 21 high-severity failures and 12 previously undocumented attack patterns. Compared with manual red-teaming under matched query budgets, our method achieves a 3.9$\times$ higher discovery rate with 89\% detection accuracy, demonstrating superior coverage, efficiency, and reproducibility for large-scale robustness evaluation.

Deep neural networks (DNNs) underpin critical applications yet remain vulnerable to backdoor attacks, typically reliant on heuristic brute-force methods. Despite significant empirical advancements in backdoor research, the lack of rigorous theoretical analysis limits understanding of underlying mechanisms, constraining attack predictability and adaptability. Therefore, we provide a theoretical analysis targeting backdoor attacks, focusing on how sparse decision boundaries enable disproportionate model manipulation. Based on this finding, we derive a closed-form, ambiguous boundary region, wherein negligible relabeled samples induce substantial misclassification. Influence function analysis further quantifies significant parameter shifts caused by these margin samples, with minimal impact on clean accuracy, formally grounding why such low poison rates suffice for efficacious attacks. Leveraging these insights, we propose Eminence, an explainable and robust black-box backdoor framework with provable theoretical guarantees and inherent stealth properties. Eminence optimizes a universal, visually subtle trigger that strategically exploits vulnerable decision boundaries and effectively achieves robust misclassification with exceptionally low poison rates (< 0.1%, compared to SOTA methods typically requiring > 1%). Comprehensive experiments validate our theoretical discussions and demonstrate the effectiveness of Eminence, confirming an exponential relationship between margin poisoning and adversarial boundary manipulation. Eminence maintains > 90% attack success rate, exhibits negligible clean-accuracy loss, and demonstrates high transferability across diverse models, datasets and scenarios.

Recent advances in multi-modal large language models (MLLMs) have enabled unified perception-reasoning capabilities, yet these systems remain highly vulnerable to jailbreak attacks that bypass safety alignment and induce harmful behaviors. Existing benchmarks such as JailBreakV-28K, MM-SafetyBench, and HADES provide valuable insights into multi-modal vulnerabilities, but they typically focus on limited attack scenarios, lack standardized defense evaluation, and offer no unified, reproducible toolbox. To address these gaps, we introduce OmniSafeBench-MM, which is a comprehensive toolbox for multi-modal jailbreak attack-defense evaluation. OmniSafeBench-MM integrates 13 representative attack methods, 15 defense strategies, and a diverse dataset spanning 9 major risk domains and 50 fine-grained categories, structured across consultative, imperative, and declarative inquiry types to reflect realistic user intentions. Beyond data coverage, it establishes a three-dimensional evaluation protocol measuring (1) harmfulness, distinguished by a granular, multi-level scale ranging from low-impact individual harm to catastrophic societal threats, (2) intent alignment between responses and queries, and (3) response detail level, enabling nuanced safety-utility analysis. We conduct extensive experiments on 10 open-source and 8 closed-source MLLMs to reveal their vulnerability to multi-modal jailbreak. By unifying data, methodology, and evaluation into an open-source, reproducible platform, OmniSafeBench-MM provides a standardized foundation for future research. The code is released at https://github.com/jiaxiaojunQAQ/OmniSafeBench-MM.

Clinical deployment requires segmentation models to stay stable under distribution shifts and perturbations. The mainstream solution is adversarial training (AT) to improve robustness; however, AT often brings a clean--robustness trade-off and high training/tuning cost, which limits scalability and maintainability in medical imaging. We propose \emph{Layer-wise Noise-Guided Selective Wavelet Reconstruction (LNG-SWR)}. During training, we inject small, zero-mean noise at multiple layers to learn a frequency-bias prior that steers representations away from noise-sensitive directions. We then apply prior-guided selective wavelet reconstruction on the input/feature branch to achieve frequency adaptation: suppress noise-sensitive bands, enhance directional structures and shape cues, and stabilize boundary responses while maintaining spectral consistency. The framework is backbone-agnostic and adds low additional inference overhead. It can serve as a plug-in enhancement to AT and also improves robustness without AT. On CT and ultrasound datasets, under a unified protocol with PGD-$L_{\infty}/L_{2}$ and SSAH, LNG-SWR delivers consistent gains on clean Dice/IoU and significantly reduces the performance drop under strong attacks; combining LNG-SWR with AT yields additive gains. When combined with adversarial training, robustness improves further without sacrificing clean accuracy, indicating an engineering-friendly and scalable path to robust segmentation. These results indicate that LNG-SWR provides a simple, effective, and engineering-friendly path to robust medical image segmentation in both adversarial and standard training regimes.

Unsupervised domain adaptation (UDA) aims to transfer knowledge from a label-rich source domain to an unlabeled target domain by addressing domain shifts. Most UDA approaches emphasize transfer ability, but often overlook robustness against adversarial attacks. Although vanilla adversarial training (VAT) improves the robustness of deep neural networks, it has little effect on UDA. This paper focuses on answering three key questions: 1) Why does VAT, known for its defensive effectiveness, fail in the UDA paradigm? 2) What is the generalization bound theory under attacks and how does it evolve from classical UDA theory? 3) How can we implement a robustification training procedure without complex modifications? Specifically, we explore and reveal the inherent entanglement challenge in general UDA+VAT paradigm, and propose an unsupervised robust domain adaptation (URDA) paradigm. We further derive the generalization bound theory of the URDA paradigm so that it can resist adversarial noise and domain shift. To the best of our knowledge, this is the first time to establish the URDA paradigm and theory. We further introduce a simple, novel yet effective URDA algorithm called Disentangled Adversarial Robustness Training (DART), a two-step training procedure that ensures both transferability and robustness. DART first pre-trains an arbitrary UDA model, and then applies an instantaneous robustification post-training step via disentangled distillation.Experiments on four benchmark datasets with/without attacks show that DART effectively enhances robustness while maintaining domain adaptability, and validate the URDA paradigm and theory.

Multimodal pre-trained models (e.g., ImageBind), which align distinct data modalities into a shared embedding space, have shown remarkable success across downstream tasks. However, their increasing adoption raises serious security concerns, especially regarding targeted adversarial attacks. In this paper, we show that existing targeted adversarial attacks on multimodal pre-trained models still have limitations in two aspects: generalizability and undetectability. Specifically, the crafted targeted adversarial examples (AEs) exhibit limited generalization to partially known or semantically similar targets in cross-modal alignment tasks (i.e., limited generalizability) and can be easily detected by simple anomaly detection methods (i.e., limited undetectability). To address these limitations, we propose a novel method called Proxy Targeted Attack (PTA), which leverages multiple source-modal and target-modal proxies to optimize targeted AEs, ensuring they remain evasive to defenses while aligning with multiple potential targets. We also provide theoretical analyses to highlight the relationship between generalizability and undetectability and to ensure optimal generalizability while meeting the specified requirements for undetectability. Furthermore, experimental results demonstrate that our PTA can achieve a high success rate across various related targets and remain undetectable against multiple anomaly detection methods.

While Code Language Models (CLMs) have demonstrated superior performance in software engineering tasks such as code generation and summarization, recent empirical studies reveal a critical privacy vulnerability: these models exhibit unintended memorization of sensitive training data, enabling verbatim reproduction of confidential information when specifically prompted. To address this issue, several approaches, including training data de-duplication and differential privacy augmentation, have been proposed. However, these methods require full-model retraining for deployed CLMs, which incurs substantial computational costs. In this paper, we aim to answer the following research question: Can sensitive information memorized by CLMs be erased effectively and efficiently? We conduct a pioneering investigation into erasing sensitive memorization in CLMs through machine unlearning - a post-hoc modification method that removes specific information from trained models without requiring full retraining. Specifically, we first quantify the memorization risks of sensitive data within CLM training datasets and curate a high-risk dataset of 50,000 sensitive memorized samples as unlearning targets. We study two widely used gradient ascent-based unlearning approaches: the vanilla and constraint-based methods, and introduce CodeEraser, an advanced variant that selectively unlearns sensitive memorized segments in code while preserving the structural integrity and functional correctness of the surrounding code. Extensive experiments on three families of CLMs, i.e., CodeParrot, CodeGen-Mono, and Qwen2.5-Coder, validate the effectiveness and efficiency of CodeEraser in erasing targeted sensitive memorization while maintaining model utility.

Recently, speech assistant and speech verification have been used in many fields, which brings much benefit and convenience for us. However, when we enjoy these speech applications, our speech may be collected by attackers for speech synthesis. For example, an attacker generates some inappropriate political opinions with the characteristic of the victim's voice by obtaining a piece of the victim's speech, which will greatly influence the victim's reputation. Specifically, with the appearance of some zero-shot voice conversion methods, the cost of speech synthesis attacks has been further reduced, which also brings greater challenges to user voice security and privacy. Some researchers have proposed the corresponding privacy-preserving methods. However, the existing approaches have some non-negligible drawbacks: low transferability and robustness, high computational overhead. These deficiencies seriously limit the existing method deployed in practical scenarios. Therefore, in this paper, we propose a lightweight, robust, plug-and-play privacy preservation method against speech synthesis attacks in a black-box setting. Our method generates and adds a frequency-domain perturbation to the original speech to achieve privacy protection and high speech quality. Then, we present a data augmentation strategy and noise smoothing mechanism to improve the robustness of the proposed method. Besides, to reduce the user's defense overhead, we also propose a novel identity-wise protection mechanism. It can generate a universal perturbation for one speaker and support privacy preservation for speech of any length. Finally, we conduct extensive experiments on 5 speech synthesis models, 5 speech verification models, 1 speech recognition model, and 2 datasets. The experimental results demonstrate that our method has satisfying privacy-preserving performance, high speech quality, and utility.

With the rapid development of large language models, the generation of fake news has become increasingly effortless, posing a growing societal threat and underscoring the urgent need for reliable detection methods. Early efforts to identify LLM-generated fake news have predominantly focused on the textual content itself; however, because much of that content may appear coherent and factually consistent, the subtle traces of falsification are often difficult to uncover. Through distributional divergence analysis, we uncover prompt-induced linguistic fingerprints: statistically distinct probability shifts between LLM-generated real and fake news when maliciously prompted. Based on this insight, we propose a novel method named Linguistic Fingerprints Extraction (LIFE). By reconstructing word-level probability distributions, LIFE can find discriminative patterns that facilitate the detection of LLM-generated fake news. To further amplify these fingerprint patterns, we also leverage key-fragment techniques that accentuate subtle linguistic differences, thereby improving detection reliability. Our experiments show that LIFE achieves state-of-the-art performance in LLM-generated fake news and maintains high performance in human-written fake news. The code and data are available at https://anonymous.4open.science/r/LIFE-E86A.

While Large Language Models (LLMs) have achieved remarkable progress, they remain vulnerable to jailbreak attacks. Existing methods, primarily relying on discrete input optimization (e.g., GCG), often suffer from high computational costs and generate high-perplexity prompts that are easily blocked by simple filters. To overcome these limitations, we propose Latent Fusion Jailbreak (LFJ), a stealthy white-box attack that operates in the continuous latent space. Unlike previous approaches, LFJ constructs adversarial representations by mathematically fusing the hidden states of a harmful query with a thematically similar benign query, effectively masking malicious intent while retaining semantic drive. We further introduce a gradient-guided optimization strategy to balance attack success and computational efficiency. Extensive evaluations on Vicuna-7B, LLaMA-2-7B-Chat, Guanaco-7B, LLaMA-3-70B, and Mistral-7B-Instruct show that LFJ achieves an average Attack Success Rate (ASR) of 94.01%, significantly outperforming state-of-the-art baselines like GCG and AutoDAN while avoiding detectable input artifacts. Furthermore, we identify that thematic similarity in the latent space is a critical vulnerability in current safety alignments. Finally, we propose a latent adversarial training defense that reduces LFJ's ASR by over 80% without compromising model utility.