Backdoor attacks pose a critical threat to the security of deep neural networks, yet existing efforts on universal backdoors often rely on visually salient patterns, making them easier to detect and less practical at scale. In this work, we introduce a novel imperceptible universal backdoor attack that simultaneously controls all target classes with minimal poisoning while preserving stealth. Our key idea is to leverage graph convolutional networks (GCNs) to model inter-class relationships and generate class-specific perturbations that are both effective and visually invisible. The proposed framework optimizes a dual-objective loss that balances stealthiness (measured by perceptual similarity metrics such as PSNR) and attack success rate (ASR), enabling scalable, multi-target backdoor injection. Extensive experiments on ImageNet-1K with ResNet architectures demonstrate that our method achieves high ASR (up to 91.3%) under poisoning rates as low as 0.16%, while maintaining benign accuracy and evading state-of-the-art defenses. These results highlight the emerging risks of invisible universal backdoors and call for more robust detection and mitigation strategies.

Benchmarking large language models (LLMs) is critical for understanding their capabilities, limitations, and robustness. In addition to interface artifacts, prior studies have shown that LLM decisions can be influenced by directive signals such as social cues, framing, and instructions. In this work, we introduce option injection, a benchmarking approach that augments the multiple-choice question answering (MCQA) interface with an additional option containing a misleading directive, leveraging standardized choice structure and scalable evaluation. We construct OI-Bench, a benchmark of 3,000 questions spanning knowledge, reasoning, and commonsense tasks, with 16 directive types covering social compliance, bonus framing, threat framing, and instructional interference. This setting combines manipulation of the choice interface with directive-based interference, enabling systematic assessment of model susceptibility. We evaluate 12 LLMs to analyze attack success rates, behavioral responses, and further investigate mitigation strategies ranging from inference-time prompting to post-training alignment. Experimental results reveal substantial vulnerabilities and heterogeneous robustness across models. OI-Bench is expected to support more systematic evaluation of LLM robustness to directive interference within choice-based interfaces.

Advances in generative modeling have made it increasingly easy to fabricate realistic portrayals of individuals, creating serious risks for security, communication, and public trust. Detecting such person-driven manipulations requires systems that not only distinguish altered content from authentic media but also provide clear and reliable reasoning. In this paper, we introduce TriDF, a comprehensive benchmark for interpretable DeepFake detection. TriDF contains high-quality forgeries from advanced synthesis models, covering 16 DeepFake types across image, video, and audio modalities. The benchmark evaluates three key aspects: Perception, which measures the ability of a model to identify fine-grained manipulation artifacts using human-annotated evidence; Detection, which assesses classification performance across diverse forgery families and generators; and Hallucination, which quantifies the reliability of model-generated explanations. Experiments on state-of-the-art multimodal large language models show that accurate perception is essential for reliable detection, but hallucination can severely disrupt decision-making, revealing the interdependence of these three aspects. TriDF provides a unified framework for understanding the interaction between detection accuracy, evidence identification, and explanation reliability, offering a foundation for building trustworthy systems that address real-world synthetic media threats.

Ensuring the authenticity of video content remains challenging as DeepFake generation becomes increasingly realistic and robust against detection. Most existing detectors implicitly assume temporally consistent and clean facial sequences, an assumption that rarely holds in real-world scenarios where compression artifacts, occlusions, and adversarial attacks destabilize face detection and often lead to invalid or misdetected faces. To address these challenges, we propose a Laplacian-Regularized Graph Convolutional Network (LR-GCN) that robustly detects DeepFakes from noisy or unordered face sequences, while being trained only on clean facial data. Our method constructs an Order-Free Temporal Graph Embedding (OF-TGE) that organizes frame-wise CNN features into an adaptive sparse graph based on semantic affinities. Unlike traditional methods constrained by strict temporal continuity, OF-TGE captures intrinsic feature consistency across frames, making it resilient to shuffled, missing, or heavily corrupted inputs. We further impose a dual-level sparsity mechanism on both graph structure and node features to suppress the influence of invalid faces. Crucially, we introduce an explicit Graph Laplacian Spectral Prior that acts as a high-pass operator in the graph spectral domain, highlighting structural anomalies and forgery artifacts, which are then consolidated by a low-pass GCN aggregation. This sequential design effectively realizes a task-driven spectral band-pass mechanism that suppresses background information and random noise while preserving manipulation cues. Extensive experiments on FF++, Celeb-DFv2, and DFDC demonstrate that LR-GCN achieves state-of-the-art performance and significantly improved robustness under severe global and local disruptions, including missing faces, occlusions, and adversarially perturbed face detections.

In deepfake detection, the varying degrees of compression employed by social media platforms pose significant challenges for model generalization and reliability. Although existing methods have progressed from single-modal to multimodal approaches, they face critical limitations: single-modal methods struggle with feature degradation under data compression in social media streaming, while multimodal approaches require expensive data collection and labeling and suffer from inconsistent modal quality or accessibility in real-world scenarios. To address these challenges, we propose a novel Unimodal-generated Multimodal Contrastive Learning (UMCL) framework for robust cross-compression-rate (CCR) deepfake detection. In the training stage, our approach transforms a single visual modality into three complementary features: compression-robust rPPG signals, temporal landmark dynamics, and semantic embeddings from pre-trained vision-language models. These features are explicitly aligned through an affinity-driven semantic alignment (ASA) strategy, which models inter-modal relationships through affinity matrices and optimizes their consistency through contrastive learning. Subsequently, our cross-quality similarity learning (CQSL) strategy enhances feature robustness across compression rates. Extensive experiments demonstrate that our method achieves superior performance across various compression rates and manipulation types, establishing a new benchmark for robust deepfake detection. Notably, our approach maintains high detection accuracy even when individual features degrade, while providing interpretable insights into feature relationships through explicit alignment.

The rapid proliferation of pretrained models and open repositories has made model merging a convenient yet risky practice, allowing free-riders to combine fine-tuned models into a new multi-capability model without authorization. Such unauthorized model merging not only violates intellectual property rights but also undermines model ownership and accountability. To address this issue, we present MergeGuard, a proactive dual-stage weight protection framework that disrupts merging compatibility while maintaining task fidelity. In the first stage, we redistribute task-relevant information across layers via L2-regularized optimization, ensuring that important gradients are evenly dispersed. In the second stage, we inject structured perturbations to misalign task subspaces, breaking curvature compatibility in the loss landscape. Together, these stages reshape the model's parameter geometry such that merged models collapse into destructive interference while the protected model remains fully functional. Extensive experiments on both vision (ViT-L-14) and language (Llama2, Gemma2, Mistral) models demonstrate that MergeGuard reduces merged model accuracy by up to 90% with less than 1.5% performance loss on the protected model.

The integration of Large Language Models (LLMs) into computer applications has introduced transformative capabilities but also significant security challenges. Existing safety alignments, which primarily focus on semantic interpretation, leave LLMs vulnerable to attacks that use non-standard data representations. This paper introduces ArtPerception, a novel black-box jailbreak framework that strategically leverages ASCII art to bypass the security measures of state-of-the-art (SOTA) LLMs. Unlike prior methods that rely on iterative, brute-force attacks, ArtPerception introduces a systematic, two-phase methodology. Phase 1 conducts a one-time, model-specific pre-test to empirically determine the optimal parameters for ASCII art recognition. Phase 2 leverages these insights to launch a highly efficient, one-shot malicious jailbreak attack. We propose a Modified Levenshtein Distance (MLD) metric for a more nuanced evaluation of an LLM's recognition capability. Through comprehensive experiments on four SOTA open-source LLMs, we demonstrate superior jailbreak performance. We further validate our framework's real-world relevance by showing its successful transferability to leading commercial models, including GPT-4o, Claude Sonnet 3.7, and DeepSeek-V3, and by conducting a rigorous effectiveness analysis against potential defenses such as LLaMA Guard and Azure's content filters. Our findings underscore that true LLM security requires defending against a multi-modal space of interpretations, even within text-only inputs, and highlight the effectiveness of strategic, reconnaissance-based attacks. Content Warning: This paper includes potentially harmful and offensive model outputs.

3D scene representation methods like Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS) have significantly advanced novel view synthesis. As these methods become prevalent, addressing their vulnerabilities becomes critical. We analyze 3DGS robustness against image-level poisoning attacks and propose a novel density-guided poisoning method. Our method strategically injects Gaussian points into low-density regions identified via Kernel Density Estimation (KDE), embedding viewpoint-dependent illusory objects clearly visible from poisoned views while minimally affecting innocent views. Additionally, we introduce an adaptive noise strategy to disrupt multi-view consistency, further enhancing attack effectiveness. We propose a KDE-based evaluation protocol to assess attack difficulty systematically, enabling objective benchmarking for future research. Extensive experiments demonstrate our method's superior performance compared to state-of-the-art techniques. Project page: https://hentci.github.io/stealthattack/

Recent studies on deep watermarking have predominantly focused on in-processing watermarking, which integrates the watermarking process into image generation. However, post-processing watermarking, which embeds watermarks after image generation, offers more flexibility. It can be applied to outputs from any generative model (e.g. GANs, diffusion models) without needing access to the model's internal structure. It also allows users to embed unique watermarks into individual images. Therefore, this study focuses on post-processing watermarking and enhances its robustness by incorporating an ensemble attack network during training. We construct various versions of attack networks using CNN and Transformer in both spatial and frequency domains to investigate how each combination influences the robustness of the watermarking model. Our results demonstrate that combining a CNN-based attack network in the spatial domain with a Transformer-based attack network in the frequency domain yields the highest robustness in watermarking models. Extensive evaluation on the WAVES benchmark, using average bit accuracy as the metric, demonstrates that our ensemble attack network significantly enhances the robustness of baseline watermarking methods under various stress tests. In particular, for the Regeneration Attack defined in WAVES, our method improves StegaStamp by 18.743%. The code is released at:https://github.com/aiiu-lab/DeepRobustWatermark.

Natural language explanations in visual question answering (VQA-NLE) aim to make black-box models more transparent by elucidating their decision-making processes. However, we find that existing VQA-NLE systems can produce inconsistent explanations and reach conclusions without genuinely understanding the underlying context, exposing weaknesses in either their inference pipeline or explanation-generation mechanism. To highlight these vulnerabilities, we not only leverage an existing adversarial strategy to perturb questions but also propose a novel strategy that minimally alters images to induce contradictory or spurious outputs. We further introduce a mitigation method that leverages external knowledge to alleviate these inconsistencies, thereby bolstering model robustness. Extensive evaluations on two standard benchmarks and two widely used VQA-NLE models underscore the effectiveness of our attacks and the potential of knowledge-based defenses, ultimately revealing pressing security and reliability concerns in current VQA-NLE systems.

Task arithmetic in large-scale pre-trained models enables agile adaptation to diverse downstream tasks without extensive retraining. By leveraging task vectors (TVs), users can perform modular updates through simple arithmetic operations like addition and subtraction. Yet, this flexibility presents new security challenges. In this paper, we investigate how TVs are vulnerable to backdoor attacks, revealing how malicious actors can exploit them to compromise model integrity. By creating composite backdoors that are designed asymmetrically, we introduce BadTV, a backdoor attack specifically crafted to remain effective simultaneously under task learning, forgetting, and analogy operations. Extensive experiments show that BadTV achieves near-perfect attack success rates across diverse scenarios, posing a serious threat to models relying on task arithmetic. We also evaluate current defenses, finding they fail to detect or mitigate BadTV. Our results highlight the urgent need for robust countermeasures to secure TVs in real-world deployments.