Multi-turn jailbreak attacks are effective against text-only large language models (LLMs) by gradually introducing malicious content across turns. When extended to large vision-language models (LVLMs), we find that naively adding visual inputs can cause existing multi-turn jailbreaks to be easily defended. For example, overly malicious visual input will easily trigger the defense mechanism of safety-aligned LVLMs, making the response more conservative. To address this, we propose MAPA: a multi-turn adaptive prompting attack that 1) at each turn, alternates text-vision attack actions to elicit the most malicious response; and 2) across turns, adjusts the attack trajectory through iterative back-and-forth refinement to gradually amplify response maliciousness. This two-level design enables MAPA to consistently outperform state-of-the-art methods, improving attack success rates by 11-35% on recent benchmarks against LLaVA-V1.6-Mistral-7B, Qwen2.5-VL-7B-Instruct, Llama-3.2-Vision-11B-Instruct and GPT-4o-mini.

Watermarking is a technical alternative to safeguarding intellectual property and reducing misuse. Existing methods focus on optimizing watermarked latent variables to balance watermark robustness and fidelity, as Latent diffusion models (LDMs) are considered a powerful tool for generative tasks. However, reliance on computationally intensive heuristic optimization for iterative signal refinement results in high training overhead and local optima entrapment.To address these issues, we propose an \underline{A}na\underline{l}ytical Watermark\underline{i}ng Framework for Controllabl\underline{e} Generatio\underline{n} (ALIEN). We develop the first analytical derivation of the time-dependent modulation coefficient that guides the diffusion of watermark residuals to achieve controllable watermark embedding pattern.Experimental results show that ALIEN-Q outperforms the state-of-the-art by 33.1\% across 5 quality metrics, and ALIEN-R demonstrates 14.0\% improved robustness against generative variant and stability threats compared to the state-of-the-art across 15 distinct conditions. Code can be available at https://anonymous.4open.science/r/ALIEN/.

Deep neural networks (DNNs) have become valuable intellectual property of model owners, due to the substantial resources required for their development. To protect these assets in the deployed environment, recent research has proposed model usage control mechanisms to ensure models cannot be used without proper authorization. These methods typically lock the utility of the model by embedding an access key into its parameters. However, they often assume static deployment, and largely fail to withstand continual post-deployment model updates, such as fine-tuning or task-specific adaptation. In this paper, we propose ADALOC, to endow key-based model usage control with adaptability during model evolution. It strategically selects a subset of weights as an intrinsic access key, which enables all model updates to be confined to this key throughout the evolution lifecycle. ADALOC enables using the access key to restore the keyed model to the latest authorized states without redistributing the entire network (i.e., adaptation), and frees the model owner from full re-keying after each model update (i.e., lock preservation). We establish a formal foundation to underpin ADALOC, providing crucial bounds such as the errors introduced by updates restricted to the access key. Experiments on standard benchmarks, such as CIFAR-100, Caltech-256, and Flowers-102, and modern architectures, including ResNet, DenseNet, and ConvNeXt, demonstrate that ADALOC achieves high accuracy under significant updates while retaining robust protections. Specifically, authorized usages consistently achieve strong task-specific performance, while unauthorized usage accuracy drops to near-random guessing levels (e.g., 1.01% on CIFAR-100), compared to up to 87.01% without ADALOC. This shows that ADALOC can offer a practical solution for adaptive and protected DNN deployment in evolving real-world scenarios.

Physical adversarial attacks on deep learning systems is concerning due to the ease of deploying such attacks, usually by placing an adversarial patch in a scene to manipulate the outcomes of a deep learning model. Training such patches typically requires regularization that improves physical realizability (e.g., printability, smoothness) and/or robustness to real-world variability (e.g. deformations, viewing angle, noise). One type of variability that has received little attention is scale variability. When a patch is rescaled, either digitally through downsampling/upsampling or physically through changing imaging distances, interpolation-induced color mixing occurs. This smooths out pixel values, resulting in a loss of high-frequency patterns and degrading the adversarial signal. To address this, we present a novel superpixel-based regularization method that guides patch optimization to scale-resilient structures. Our ap proach employs the Simple Linear Iterative Clustering (SLIC) algorithm to dynamically cluster pixels in an adversarial patch during optimization. The Implicit Function Theorem is used to backpropagate gradients through SLIC to update the superpixel boundaries and color. This produces patches that maintain their structure over scale and are less susceptible to interpolation losses. Our method achieves greater performance in the digital domain, and when realized physically, these performance gains are preserved, leading to improved physical performance. Real-world performance was objectively assessed using a novel physical evaluation protocol that utilizes screens and cardboard cut-outs to systematically vary real-world conditions.

Recent advancements in speech synthesis technology have enriched our daily lives, with high-quality and human-like audio widely adopted across real-world applications. However, malicious exploitation like voice-cloning fraud poses severe security risks. Existing defense techniques struggle to address the production large language model (LLM)-based speech synthesis. While previous studies have considered the protection for fine-tuning synthesizers, they assume manually annotated transcripts. Given the labor intensity of manual annotation, end-to-end (E2E) systems leveraging automatic speech recognition (ASR) to generate transcripts are becoming increasingly prevalent, e.g., voice cloning via commercial APIs. Therefore, this E2E speech synthesis also requires new security mechanisms. To tackle these challenges, we propose E2E-VGuard, a proactive defense framework for two emerging threats: (1) production LLM-based speech synthesis, and (2) the novel attack arising from ASR-driven E2E scenarios. Specifically, we employ the encoder ensemble with a feature extractor to protect timbre, while ASR-targeted adversarial examples disrupt pronunciation. Moreover, we incorporate the psychoacoustic model to ensure perturbative imperceptibility. For a comprehensive evaluation, we test 16 open-source synthesizers and 3 commercial APIs across Chinese and English datasets, confirming E2E-VGuard's effectiveness in timbre and pronunciation protection. Real-world deployment validation is also conducted. Our code and demo page are available at https://wxzyd123.github.io/e2e-vguard/.

Recent advances in Audio-Language Models (ALMs) have significantly improved multimodal understanding capabilities. However, the introduction of the audio modality also brings new and unique vulnerability vectors. Previous studies have proposed jailbreak attacks that specifically target ALMs, revealing that defenses directly transferred from traditional audio adversarial attacks or text-based Large Language Model (LLM) jailbreaks are largely ineffective against these ALM-specific threats. To address this issue, we propose ALMGuard, the first defense framework tailored to ALMs. Based on the assumption that safety-aligned shortcuts naturally exist in ALMs, we design a method to identify universal Shortcut Activation Perturbations (SAPs) that serve as triggers that activate the safety shortcuts to safeguard ALMs at inference time. To better sift out effective triggers while preserving the model's utility on benign tasks, we further propose Mel-Gradient Sparse Mask (M-GSM), which restricts perturbations to Mel-frequency bins that are sensitive to jailbreaks but insensitive to speech understanding. Both theoretical analyses and empirical results demonstrate the robustness of our method against both seen and unseen attacks. Overall, \MethodName reduces the average success rate of advanced ALM-specific jailbreak attacks to 4.6% across four models, while maintaining comparable utility on benign benchmarks, establishing it as the new state of the art. Our code and data are available at https://github.com/WeifeiJin/ALMGuard.

Multi-agent systems powered by large language models are advancing rapidly, yet the tension between mutual trust and security remains underexplored. We introduce and empirically validate the Trust-Vulnerability Paradox (TVP): increasing inter-agent trust to enhance coordination simultaneously expands risks of over-exposure and over-authorization. To investigate this paradox, we construct a scenario-game dataset spanning 3 macro scenes and 19 sub-scenes, and run extensive closed-loop interactions with trust explicitly parameterized. Using Minimum Necessary Information (MNI) as the safety baseline, we propose two unified metrics: Over-Exposure Rate (OER) to detect boundary violations, and Authorization Drift (AD) to capture sensitivity to trust levels. Results across multiple model backends and orchestration frameworks reveal consistent trends: higher trust improves task success but also heightens exposure risks, with heterogeneous trust-to-risk mappings across systems. We further examine defenses such as Sensitive Information Repartitioning and Guardian-Agent enablement, both of which reduce OER and attenuate AD. Overall, this study formalizes TVP, establishes reproducible baselines with unified metrics, and demonstrates that trust must be modeled and scheduled as a first-class security variable in multi-agent system design.

The Model Context Protocol (MCP) has emerged as a universal standard that enables AI agents to seamlessly connect with external tools, significantly enhancing their functionality. However, while MCP brings notable benefits, it also introduces significant vulnerabilities, such as Tool Poisoning Attacks (TPA), where hidden malicious instructions exploit the sycophancy of large language models (LLMs) to manipulate agent behavior. Despite these risks, current academic research on MCP security remains limited, with most studies focusing on narrow or qualitative analyses that fail to capture the diversity of real-world threats. To address this gap, we present the MCP Attack Library (MCPLIB), which categorizes and implements 31 distinct attack methods under four key classifications: direct tool injection, indirect tool injection, malicious user attacks, and LLM inherent attack. We further conduct a quantitative analysis of the efficacy of each attack. Our experiments reveal key insights into MCP vulnerabilities, including agents' blind reliance on tool descriptions, sensitivity to file-based attacks, chain attacks exploiting shared context, and difficulty distinguishing external data from executable commands. These insights, validated through attack experiments, underscore the urgency for robust defense strategies and informed MCP design. Our contributions include 1) constructing a comprehensive MCP attack taxonomy, 2) introducing a unified attack framework MCPLIB, and 3) conducting empirical vulnerability analysis to enhance MCP security mechanisms. This work provides a foundational framework, supporting the secure evolution of MCP ecosystems.

Defending backdoor attacks in Federated Learning (FL) under heterogeneous client data distributions encounters limitations balancing effectiveness and privacy-preserving, while most existing methods highly rely on the assumption of homogeneous client data distributions or the availability of a clean serve dataset. In this paper, we propose an FL backdoor defense framework, named CLIP-Fed, that utilizes the zero-shot learning capabilities of vision-language pre-training models. Our scheme overcomes the limitations of Non-IID imposed on defense effectiveness by integrating pre-aggregation and post-aggregation defense strategies. CLIP-Fed aligns the knowledge of the global model and CLIP on the augmented dataset using prototype contrastive loss and Kullback-Leibler divergence, so that class prototype deviations caused by backdoor samples are ensured and the correlation between trigger patterns and target labels is eliminated. In order to balance privacy-preserving and coverage enhancement of the dataset against diverse triggers, we further construct and augment the server dataset via using the multimodal large language model and frequency analysis without any client samples. Extensive experiments on representative datasets evidence the effectiveness of CLIP-Fed. Comparing to other existing methods, CLIP-Fed achieves an average reduction in Attack Success Rate, {\em i.e.}, 2.03\% on CIFAR-10 and 1.35\% on CIFAR-10-LT, while improving average Main Task Accuracy by 7.92\% and 0.48\%, respectively. Our codes are available at https://anonymous.4open.science/r/CLIP-Fed.