Yechao Zhang

Jiawen Zhang, Kejia Chen, Jiachen Ma et al.

Personal AI agents increasingly rely on long-term memory to provide persistent personalization across sessions. However, existing memory pipelines are largely driven by semantic similarity: memory data close to the current query is retrieved and injected into the model context. This creates a critical trustworthiness gap, since a semantically related memory may still be contextually inappropriate, leading to threats such as cross-domain leakage, sycophancy, tool-call drift, or memory-induced jailbreaks. In this paper, we study memory search as a trust boundary in personal AI agents. We evaluate representative agentic memory frameworks, including A-Mem, Mem0, and MemOS, together with OpenClaw, a real-world personal-agent environment with persistent state and tool-use capability. Our results show that long-term memory is not merely a utility layer, but a durable control channel that can reshape how agents interpret tasks and execute actions, leaving them highly susceptible to the aforementioned threats. To mitigate these vulnerabilities, we propose MemGate, a lightweight and deployable memory plug-in for trustworthy memory search, with only 9M parameters and a 35.1MB footprint. MemGate is inserted between the vector memory store and the backbone LLM, requiring no LLM modification, memory-database rewriting, or inference-time LLM judge. It applies a query-conditioned neural gate to candidate memory representations, turning raw similarity search into task-conditioned memory admission. Across multiple mainstream memory frameworks, real-world agent settings, and diverse LLM backbones, MemGate reduces memory-induced threats while preserving long-term memory utility.

Shengshan Hu, Xiaogeng Liu, Yechao Zhang et al.

While deep face recognition (FR) systems have shown amazing performance in identification and verification, they also arouse privacy concerns for their excessive surveillance on users, especially for public face images widely spread on social networks. Recently, some studies adopt adversarial examples to protect photos from being identified by unauthorized face recognition systems. However, existing methods of generating adversarial face images suffer from many limitations, such as awkward visual, white-box setting, weak transferability, making them difficult to be applied to protect face privacy in reality. In this paper, we propose adversarial makeup transfer GAN (AMT-GAN), a novel face protection method aiming at constructing adversarial face images that preserve stronger black-box transferability and better visual quality simultaneously. AMT-GAN leverages generative adversarial networks (GAN) to synthesize adversarial face images with makeup transferred from reference images. In particular, we introduce a new regularization module along with a joint training strategy to reconcile the conflicts between the adversarial noises and the cycle consistence loss in makeup transfer, achieving a desirable balance between the attack strength and visual changes. Extensive experiments verify that compared with state of the arts, AMT-GAN can not only preserve a comfortable visual quality, but also achieve a higher attack success rate over commercial FR APIs, including Face++, Aliyun, and Microsoft.

Ziqi Zhou, Shengshan Hu, Minghui Li et al.

Multimodal contrastive learning aims to train a general-purpose feature extractor, such as CLIP, on vast amounts of raw, unlabeled paired image-text data. This can greatly benefit various complex downstream tasks, including cross-modal image-text retrieval and image classification. Despite its promising prospect, the security issue of cross-modal pre-trained encoder has not been fully explored yet, especially when the pre-trained encoder is publicly available for commercial use. In this work, we propose AdvCLIP, the first attack framework for generating downstream-agnostic adversarial examples based on cross-modal pre-trained encoders. AdvCLIP aims to construct a universal adversarial patch for a set of natural images that can fool all the downstream tasks inheriting the victim cross-modal pre-trained encoder. To address the challenges of heterogeneity between different modalities and unknown downstream tasks, we first build a topological graph structure to capture the relevant positions between target samples and their neighbors. Then, we design a topology-deviation based generative adversarial network to generate a universal adversarial patch. By adding the patch to images, we minimize their embeddings similarity to different modality and perturb the sample distribution in the feature space, achieving unviersal non-targeted attacks. Our results demonstrate the excellent attack performance of AdvCLIP on two types of downstream tasks across eight datasets. We also tailor three popular defenses to mitigate AdvCLIP, highlighting the need for new defense mechanisms to defend cross-modal pre-trained encoders.

Xiaogeng Liu, Xinyan Wang, Yechao Zhang et al.

Large reasoning models (LRMs) extend large language models with explicit multi-step reasoning traces, but this capability introduces a new class of prompt-induced inference-time denial-of-service (PI-DoS) attacks that exploit the high computational cost of reasoning. We first formalize inference cost for LRMs and define PI-DoS, then prove that any practical PI-DoS attack should satisfy three properties: (1) a high amplification ratio, where each query induces a disproportionately long reasoning trace relative to its own length; (ii) stealthiness, in which prompts and responses remain on the natural language manifold and evade distribution shift detectors; and (iii) optimizability, in which the attack supports efficient optimization without being slowed by its own success. Under this framework, we present ReasoningBomb, a reinforcement-learning-based PI-DoS framework that is guided by a constant-time surrogate reward and trains a large reasoning-model attacker to generate short natural prompts that drive victim LRMs into pathologically long and often effectively non-terminating reasoning. Across seven open-source models (including LLMs and LRMs) and three commercial LRMs, ReasoningBomb induces 18,759 completion tokens on average and 19,263 reasoning tokens on average across reasoning models. It outperforms the the runner-up baseline by 35% in completion tokens and 38% in reasoning tokens, while inducing 6-7x more tokens than benign queries and achieving 286.7x input-to-output amplification ratio averaged across all samples. Additionally, our method achieves 99.8% bypass rate on input-based detection, 98.7% on output-based detection, and 98.4% against strict dual-stage joint detection.

Shengshan Hu, Ziqi Zhou, Yechao Zhang et al.

Due to its powerful feature learning capability and high efficiency, deep hashing has achieved great success in large-scale image retrieval. Meanwhile, extensive works have demonstrated that deep neural networks (DNNs) are susceptible to adversarial examples, and exploring adversarial attack against deep hashing has attracted many research efforts. Nevertheless, backdoor attack, another famous threat to DNNs, has not been studied for deep hashing yet. Although various backdoor attacks have been proposed in the field of image classification, existing approaches failed to realize a truly imperceptive backdoor attack that enjoys invisible triggers and clean label setting simultaneously, and they also cannot meet the intrinsic demand of image retrieval backdoor. In this paper, we propose BadHash, the first generative-based imperceptible backdoor attack against deep hashing, which can effectively generate invisible and input-specific poisoned images with clean label. Specifically, we first propose a new conditional generative adversarial network (cGAN) pipeline to effectively generate poisoned samples. For any given benign image, it seeks to generate a natural-looking poisoned counterpart with a unique invisible trigger. In order to improve the attack effectiveness, we introduce a label-based contrastive learning network LabCLN to exploit the semantic characteristics of different labels, which are subsequently used for confusing and misleading the target model to learn the embedded trigger. We finally explore the mechanism of backdoor attacks on image retrieval in the hash space. Extensive experiments on multiple benchmark datasets verify that BadHash can generate imperceptible poisoned samples with strong attack ability and transferability over state-of-the-art deep hashing schemes.

Yechao Zhang, Shengshan Hu, Leo Yu Zhang et al.

Adversarial examples (AEs) for DNNs have been shown to be transferable: AEs that successfully fool white-box surrogate models can also deceive other black-box models with different architectures. Although a bunch of empirical studies have provided guidance on generating highly transferable AEs, many of these findings lack explanations and even lead to inconsistent advice. In this paper, we take a further step towards understanding adversarial transferability, with a particular focus on surrogate aspects. Starting from the intriguing little robustness phenomenon, where models adversarially trained with mildly perturbed adversarial samples can serve as better surrogates, we attribute it to a trade-off between two predominant factors: model smoothness and gradient similarity. Our investigations focus on their joint effects, rather than their separate correlations with transferability. Through a series of theoretical and empirical analyses, we conjecture that the data distribution shift in adversarial training explains the degradation of gradient similarity. Building on these insights, we explore the impacts of data augmentation and gradient regularization on transferability and identify that the trade-off generally exists in the various training mechanisms, thus building a comprehensive blueprint for the regulation mechanism behind transferability. Finally, we provide a general route for constructing better surrogates to boost transferability which optimizes both model smoothness and gradient similarity simultaneously, e.g., the combination of input gradient regularization and sharpness-aware minimization (SAM), validated by extensive experiments. In summary, we call for attention to the united impacts of these two factors for launching effective transfer attacks, rather than optimizing one while ignoring the other, and emphasize the crucial role of manipulating surrogate models.

Xiaogeng Liu, Haoyu Wang, Yechao Zhang et al.

The data-centric machine learning aims to find effective ways to build appropriate datasets which can improve the performance of AI models. In this paper, we mainly focus on designing an efficient data-centric scheme to improve robustness for models towards unforeseen malicious inputs in the black-box test settings. Specifically, we introduce a noised-based data augmentation method which is composed of Gaussian Noise, Salt-and-Pepper noise, and the PGD adversarial perturbations. The proposed method is built on lightweight algorithms and proved highly effective based on comprehensive evaluations, showing good efficiency on computation cost and robustness enhancement. In addition, we share our insights about the data-centric robust machine learning gained from our experiments.

Chenhao Qiu, Yechao Zhang, Xin Luo et al.

Long video question answering requires locating sparse, time-scattered visual evidence within highly redundant content. Although current MLLMs perform well on short videos, long videos introduce long-horizon search and verification, which often necessitates multi-turn, agentic interaction. We show that existing LVU agents can exhibit "evidence misalignment": they produce correct answers that are not supported by the retrieved or inspected evidence. To characterize this failure, we introduce two diagnostics (temporal groundedness and semantic groundedness) and use them to reveal two pressures that amplify misalignment: prompt pressure from shared-context saturation at inference time and reward pressure from outcome-only optimization during training. These findings point to a structural root cause: the coupled agent paradigm conflates long-horizon planning with answer authority. We therefore propose the decoupled planner-inspector framework, which separates planning from answer authority and gates final answering on pixel-level verification. Across four long-video benchmarks, our framework improves both answer accuracy and evidence alignment, achieving 55.1% on LVBench and 62.0% on LongVideoBench while producing interpretable search trajectories. Moreover, the decoupled architecture scales consistently with increased search budgets and supports plug-and-play upgrades of the MLLM backbone without retraining the planner. Code and models are available at https://github.com/Echochef/VideoSEAL.

Yechao Zhang, Shiqian Zhao, Jie Zhang et al.

We identify a critical security vulnerability in mainstream Claw personal AI agents: untrusted content encountered during heartbeat-driven background execution can silently pollute agent memory and subsequently influence user-facing behavior without the user's awareness. This vulnerability arises from an architectural design shared across the Claw ecosystem: heartbeat background execution runs in the same session as user-facing conversation, so content ingested from any external source monitored in the background (including email, message channels, news feeds, code repositories, and social platforms) can enter the same memory context used for foreground interaction, often with limited user visibility and without clear source provenance. We formalize this process as an Exposure (E) $\rightarrow$ Memory (M) $\rightarrow$ Behavior (B) pathway: misinformation encountered during heartbeat execution enters the agent's short-term session context, potentially gets written into long-term memory, and later shapes downstream user-facing behavior. We instantiate this pathway in an agent-native social setting using MissClaw, a controlled research replica of Moltbook. We find that (1) social credibility cues, especially perceived consensus, are the dominant driver of short-term behavioral influence, with misleading rates up to 61%; (2) routine memory-saving behavior can promote short-term pollution into durable long-term memory at rates up to 91%, with cross-session behavioral influence reaching 76%; (3) under naturalistic browsing with content dilution and context pruning, pollution still crosses session boundaries. Overall, prompt injection is not required: ordinary social misinformation is sufficient to silently shape agent memory and behavior under heartbeat-driven background execution.

Xiaogeng Liu, Xinyan Wang, Yingzi Ma et al.

On-policy self-distillation (OPSD) trains a student on its own rollouts using a privileged teacher, but its standard objective weights all generated tokens equally, implicitly treating the privileged teacher target as equally reliable at every student-visited prefix. Existing entropy-based OPD methods relax this uniformity by modulating token-level supervision with teacher entropy, but high teacher entropy in reasoning has an ambiguous reliability meaning: it can reflect either non-viable uncertainty or benign solution diversity. To identify this phenomenon, we introduce a branch-viability diagnostic. Specifically, we record next-token alternatives from the privileged-answer teacher prompt, force each alternative after the student prompt plus its on-policy spine prefix, and test whether the resulting student-template continuation recovers the correct answer. On Qwen3-4B, we find that an oriented within-sequence position score is the strongest tested predictor of teacher-token reliability, reaching an area-under-ROC-curve (AUROC) of 0.83; local uncertainty scores are at most 0.57. Motivated by this trajectory-level structure, we propose Position-Weighted On-Policy Self-Distillation (PW-OPSD), which applies an increasing position weight while keeping the same student rollout, privileged teacher pass, and clipped forward-KL target as OPSD. In our comprehensive evaluations with different random seeds, the diagnostic-derived PW-OPSD improves AIME 2024 and AIME 2025 Avg@12 by +1.0 and +1.1 points, and a generalization evaluation on two larger-scale models from different families, DeepSeek-R1-Distill-Llama-8B and Olmo-3-7B-Think, also demonstrates consistent aggregate Avg@12 improvements. These results show that teacher-token reliability in reasoning distillation is trajectory-structured and can be utilized without additional teacher computation.

Lulu Xue, Shengshan Hu, Linqiang Qian et al.

Machine unlearning is a newly popularized technique for removing specific training data from a trained model, enabling it to comply with data deletion requests. While it protects the rights of users requesting unlearning, it also introduces new privacy risks. Prior works have primarily focused on the privacy of data that has been unlearned, while the risks to retained data remain largely unexplored. To address this gap, we focus on the privacy risks of retained data and, for the first time, reveal the vulnerabilities introduced by machine unlearning under the dual-view setting, where an adversary can query both the original and the unlearned models. From an information-theoretic perspective, we introduce the concept of {privacy knowledge gain} and demonstrate that the dual-view setting allows adversaries to obtain more information than querying either model alone, thereby amplifying privacy leakage. To effectively demonstrate this threat, we propose DVIA, a Dual-View Inference Attack, which extracts membership information on retained data using black-box queries to both models. DVIA eliminates the need to train an attack model and employs a lightweight likelihood ratio inference module for efficient inference. Experiments across different datasets and model architectures validate the effectiveness of DVIA and highlight the privacy risks inherent in the dual-view setting.

Boyang Ma, Hechuan Guo, Peizhuo Lv et al.

Embodied AI systems (e.g., autonomous vehicles, service robots, and LLM-driven interactive agents) are rapidly transitioning from controlled environments to safety critical real-world deployments. Unlike disembodied AI, failures in embodied intelligence lead to irreversible physical consequences, raising fundamental questions about security, safety, and reliability. While existing research predominantly analyzes embodied AI through the lenses of Large Language Model (LLM) vulnerabilities or classical Cyber-Physical System (CPS) failures, this survey argues that these perspectives are individually insufficient to explain many observed breakdowns in modern embodied systems. We posit that a significant class of failures arises from embodiment-induced system-level mismatches, rather than from isolated model flaws or traditional CPS attacks. Specifically, we identify four core insights that explain why embodied AI is fundamentally harder to secure: (i) semantic correctness does not imply physical safety, as language-level reasoning abstracts away geometry, dynamics, and contact constraints; (ii) identical actions can lead to drastically different outcomes across physical states due to nonlinear dynamics and state uncertainty; (iii) small errors propagate and amplify across tightly coupled perception-decision-action loops; and (iv) safety is not compositional across time or system layers, enabling locally safe decisions to accumulate into globally unsafe behavior. These insights suggest that securing embodied AI requires moving beyond component-level defenses toward system-level reasoning about physical risk, uncertainty, and failure propagation.

Xiao Li, Yue Li, Hao Wu et al.

As large language models (LLMs) are increasingly adopted for code vulnerability detection, their reliability and robustness across diverse vulnerability types have become a pressing concern. In traditional adversarial settings, code obfuscation has long been used as a general strategy to bypass auditing tools, preserving exploitability without tampering with the tools themselves. Numerous efforts have explored obfuscation methods and tools, yet their capabilities differ in terms of supported techniques, granularity, and programming languages, making it difficult to systematically assess their impact on LLM-based vulnerability detection. To address this gap, we provide a structured systematization of obfuscation techniques and evaluate them under a unified framework. Specifically, we categorize existing obfuscation methods into three major classes (layout, data flow, and control flow) covering 11 subcategories and 19 concrete techniques. We implement these techniques across four programming languages (Solidity, C, C++, and Python) using a consistent LLM-driven approach, and evaluate their effects on 15 LLMs spanning four model families (DeepSeek, OpenAI, Qwen, and LLaMA), as well as on two coding agents (GitHub Copilot and Codex). Our findings reveal both positive and negative impacts of code obfuscation on LLM-based vulnerability detection, highlighting conditions under which obfuscation leads to performance improvements or degradations. We further analyze these outcomes with respect to vulnerability characteristics, code properties, and model attributes. Finally, we outline several open problems and propose future directions to enhance the robustness of LLMs for real-world vulnerability detection.

Yue Li, Xiao Li, Hao Wu et al.

Large Language Models (LLMs) are increasingly used for automated software development, making their ability to preserve secure coding practices critical. In practice, however, many security requirements are implicit or underspecified, whereas usability requirements are explicit and high-signal. This asymmetry motivates our investigation of usability pressure as a practical attack surface: realistic usability-oriented requirements (e.g., new features, performance constraints, or simplicity demands) can cause coding LLMs to satisfy explicit usability goals while silently dropping implicit security constraints -- a form of reward hacking. We formalize this threat as UPAttack and propose U-SPLOIT, an automated framework to craft UPAttack that (i) selects tasks where a model is initially secure, (ii) synthesizes usability pressures by identifying usability rewards of insecure alternatives across three vectors (Functionality, Implementation, Trade-off), and (iii) verifies security regression via both existing test cases and dynamically generated exploit payloads. Across 75 seed scenarios (25 CWEs x 3 cases), spanning multiple languages (Python, C, and JavaScript), U-SPLOIT achieves attack success rates up to 98.1% on multiple state-of-the-art models (e.g., GPT-5.2-chat and Gemini-3-Flash-Preview).

Junyu Shi, Minghui Li, Junguo Zuo et al.

Deepfakes, leveraging advanced AIGC (Artificial Intelligence-Generated Content) techniques, create hyper-realistic synthetic images and videos of human faces, posing a significant threat to the authenticity of social media. While this real-world threat is increasingly prevalent, existing academic evaluations and benchmarks for detecting deepfake forgery often fall short to achieve effective application for their lack of specificity, limited deepfake diversity, restricted manipulation techniques.To address these limitations, we introduce RedFace (Real-world-oriented Deepfake Face), a specialized facial deepfake dataset, comprising over 60,000 forged images and 1,000 manipulated videos derived from authentic facial features, to bridge the gap between academic evaluations and real-world necessity. Unlike prior benchmarks, which typically rely on academic methods to generate deepfakes, RedFace utilizes 9 commercial online platforms to integrate the latest deepfake technologies found "in the wild", effectively simulating real-world black-box scenarios.Moreover, RedFace's deepfakes are synthesized using bespoke algorithms, allowing it to capture diverse and evolving methods used by real-world deepfake creators. Extensive experimental results on RedFace (including cross-domain, intra-domain, and real-world social network dissemination simulations) verify the limited practicality of existing deepfake detection schemes against real-world applications. We further perform a detailed analysis of the RedFace dataset, elucidating the reason of its impact on detection performance compared to conventional datasets. Our dataset is available at: https://github.com/kikyou-220/RedFace.

Xianlong Wang, Shengshan Hu, Yechao Zhang et al.

Clean-label indiscriminate poisoning attacks add invisible perturbations to correctly labeled training images, thus dramatically reducing the generalization capability of the victim models. Recently, some defense mechanisms have been proposed such as adversarial training, image transformation techniques, and image purification. However, these schemes are either susceptible to adaptive attacks, built on unrealistic assumptions, or only effective against specific poison types, limiting their universal applicability. In this research, we propose a more universally effective, practical, and robust defense scheme called ECLIPSE. We first investigate the impact of Gaussian noise on the poisons and theoretically prove that any kind of poison will be largely assimilated when imposing sufficient random noise. In light of this, we assume the victim has access to an extremely limited number of clean images (a more practical scene) and subsequently enlarge this sparse set for training a denoising probabilistic model (a universal denoising tool). We then begin by introducing Gaussian noise to absorb the poisons and then apply the model for denoising, resulting in a roughly purified dataset. Finally, to address the trade-off of the inconsistency in the assimilation sensitivity of different poisons by Gaussian noise, we propose a lightweight corruption compensation module to effectively eliminate residual poisons, providing a more universal defense approach. Extensive experiments demonstrate that our defense approach outperforms 10 state-of-the-art defenses. We also propose an adaptive attack against ECLIPSE and verify the robustness of our defense scheme. Our code is available at https://github.com/CGCL-codes/ECLIPSE.

Ziqi Zhou, Minghui Li, Wei Liu et al.

With the evolution of self-supervised learning, the pre-training paradigm has emerged as a predominant solution within the deep learning landscape. Model providers furnish pre-trained encoders designed to function as versatile feature extractors, enabling downstream users to harness the benefits of expansive models with minimal effort through fine-tuning. Nevertheless, recent works have exposed a vulnerability in pre-trained encoders, highlighting their susceptibility to downstream-agnostic adversarial examples (DAEs) meticulously crafted by attackers. The lingering question pertains to the feasibility of fortifying the robustness of downstream models against DAEs, particularly in scenarios where the pre-trained encoders are publicly accessible to the attackers. In this paper, we initially delve into existing defensive mechanisms against adversarial examples within the pre-training paradigm. Our findings reveal that the failure of current defenses stems from the domain shift between pre-training data and downstream tasks, as well as the sensitivity of encoder parameters. In response to these challenges, we propose Genetic Evolution-Nurtured Adversarial Fine-tuning (Gen-AF), a two-stage adversarial fine-tuning approach aimed at enhancing the robustness of downstream models. Our extensive experiments, conducted across ten self-supervised training methods and six datasets, demonstrate that Gen-AF attains high testing accuracy and robust testing accuracy against state-of-the-art DAEs.

Yechao Zhang, Yingzhe Xu, Junyu Shi et al.

Deep neural networks (DNNs) are susceptible to universal adversarial perturbations (UAPs). These perturbations are meticulously designed to fool the target model universally across all sample classes. Unlike instance-specific adversarial examples (AEs), generating UAPs is more complex because they must be generalized across a wide range of data samples and models. Our research reveals that existing universal attack methods, which optimize UAPs using DNNs with static model parameter snapshots, do not fully leverage the potential of DNNs to generate more effective UAPs. Rather than optimizing UAPs against static DNN models with a fixed training set, we suggest using dynamic model-data pairs to generate UAPs. In particular, we introduce a dynamic maximin optimization strategy, aiming to optimize the UAP across a variety of optimal model-data pairs. We term this approach DM-UAP. DM-UAP utilizes an iterative max-min-min optimization framework that refines the model-data pairs, coupled with a curriculum UAP learning algorithm to examine the combined space of model parameters and data thoroughly. Comprehensive experiments on the ImageNet dataset demonstrate that the proposed DM-UAP markedly enhances both cross-sample universality and cross-model transferability of UAPs. Using only 500 samples for UAP generation, DM-UAP outperforms the state-of-the-art approach with an average increase in fooling ratio of 12.108%.

Wei Wan, Yuxuan Ning, Zhicong Huang et al.

Federated Learning (FL) is a distributed paradigm aimed at protecting participant data privacy by exchanging model parameters to achieve high-quality model training. However, this distributed nature also makes FL highly vulnerable to backdoor attacks. Notably, the recently proposed state-of-the-art (SOTA) attack, 3DFed (SP2023), uses an indicator mechanism to determine whether the backdoor models have been accepted by the defender and adaptively optimizes backdoor models, rendering existing defenses ineffective. In this paper, we first reveal that the failure of existing defenses lies in the employment of empirical statistical measures that are loosely coupled with backdoor attacks. Motivated by this, we propose a Malignity-Aware backdooR defenSe (MARS) that leverages backdoor energy (BE) to indicate the malicious extent of each neuron. To amplify malignity, we further extract the most prominent BE values from each model to form a concentrated backdoor energy (CBE). Finally, a novel Wasserstein distance-based clustering method is introduced to effectively identify backdoor models. Extensive experiments demonstrate that MARS can defend against SOTA backdoor attacks and significantly outperforms existing defenses.

Ye Cheng, Minghui Xu, Yue Zhang et al.

Access control in the Internet of Things (IoT) is becoming increasingly complex, as policies must account for dynamic and contextual factors such as time, location, user behavior, and environmental conditions. However, existing platforms either offer only coarse-grained controls or rely on rigid rule matching, making them ill-suited for semantically rich or ambiguous access scenarios. Moreover, the policy authoring process remains fragmented: domain experts describe requirements in natural language, but developers must manually translate them into code, introducing semantic gaps and potential misconfiguration. In this work, we present LACE, the Language-based Access Control Engine, a hybrid framework that leverages large language models (LLMs) to bridge the gap between human intent and machine-enforceable logic. LACE combines prompt-guided policy generation, retrieval-augmented reasoning, and formal validation to support expressive, interpretable, and verifiable access control. It enables users to specify policies in natural language, automatically translates them into structured rules, validates semantic correctness, and makes access decisions using a hybrid LLM-rule-based engine. We evaluate LACE in smart home environments through extensive experiments. LACE achieves 100% correctness in verified policy generation and up to 88% decision accuracy with 0.79 F1-score using DeepSeek-V3, outperforming baselines such as GPT-3.5 and Gemini. The system also demonstrates strong scalability under increasing policy volume and request concurrency. Our results highlight LACE's potential to enable secure, flexible, and user-friendly access control across real-world IoT platforms.

Yechao Zhang, Yuxuan Zhou, Tianyu Li et al.

Transfer learning from pre-trained encoders has become essential in modern machine learning, enabling efficient model adaptation across diverse tasks. However, this combination of pre-training and downstream adaptation creates an expanded attack surface, exposing models to sophisticated backdoor embeddings at both the encoder and dataset levels--an area often overlooked in prior research. Additionally, the limited computational resources typically available to users of pre-trained encoders constrain the effectiveness of generic backdoor defenses compared to end-to-end training from scratch. In this work, we investigate how to mitigate potential backdoor risks in resource-constrained transfer learning scenarios. Specifically, we conduct an exhaustive analysis of existing defense strategies, revealing that many follow a reactive workflow based on assumptions that do not scale to unknown threats, novel attack types, or different training paradigms. In response, we introduce a proactive mindset focused on identifying clean elements and propose the Trusted Core (T-Core) Bootstrapping framework, which emphasizes the importance of pinpointing trustworthy data and neurons to enhance model security. Our empirical evaluations demonstrate the effectiveness and superiority of T-Core, specifically assessing 5 encoder poisoning attacks, 7 dataset poisoning attacks, and 14 baseline defenses across five benchmark datasets, addressing four scenarios of 3 potential backdoor threats.

Yuhang Huang, Boyang Ma, Biwei Yan et al.

The Model Context Protocol (MCP) is an open and standardized interface that enables large language models (LLMs) to interact with external tools and services, and is increasingly adopted by AI agents. However, the security of MCP-based systems remains largely unexplored.In this work, we conduct a large-scale security analysis of MCP servers integrated within MCP clients. We show that treating MCP servers as trusted entities without authenticating the caller identity is fundamentally insecure. Since MCP servers often cannot distinguish who is invoking a request, a single authorization decision may implicitly grant access to multiple, potentially untrusted callers.Our empirical study reveals that most MCP servers rely on persistent authorization states, allowing tool invocations after an initial authorization without re-authentication, regardless of the caller. In addition, many MCP servers fail to enforce authentication at the per-tool level, enabling unauthorized access to sensitive operations.These findings demonstrate that one-time authorization and server-level trust significantly expand the attack surface of MCP-based systems, highlighting the need for explicit caller authentication and fine-grained authorization mechanisms.

Minghui Li, Hao Zhang, Yechao Zhang et al.

Direct Prompt Injection (DPI) attacks pose a critical security threat to Large Language Models (LLMs) due to their low barrier of execution and high potential damage. To address the impracticality of existing white-box/gray-box methods and the poor transferability of black-box methods, we propose an activations-guided prompt injection attack framework. We first construct an Energy-based Model (EBM) using activations from a surrogate model to evaluate the quality of adversarial prompts. Guided by the trained EBM, we employ the token-level Markov Chain Monte Carlo (MCMC) sampling to adaptively optimize adversarial prompts, thereby enabling gradient-free black-box attacks. Experimental results demonstrate our superior cross-model transferability, achieving 49.6% attack success rate (ASR) across five mainstream LLMs and 34.6% improvement over human-crafted prompts, and maintaining 36.6% ASR on unseen task scenarios. Interpretability analysis reveals a correlation between activations and attack effectiveness, highlighting the critical role of semantic patterns in transferable vulnerability exploitation.