Kejiang Chen

Benlong Wu, Weiming Zhang, Kejiang Chen et al.

As large language models transition from bounded generative engines to agents with expansive execution privileges, AI going out of control precipitates a fundamental crisis in artificial intelligence security. Existing defense architectures heavily rely on empirical semantic guardrails and probabilistic large model adjudicators, mechanisms that fail to provide deterministic security lower bounds when facing complex semantic symbol decoupling attacks. To overcome this empirical semantic guardrail dilemma, this paper proposes a new security paradigm for agents based on the fundamental limitations of logical reasoning. Based on this paradigm, we further introduce an executable Proof-Constrained Action (ePCA) framework with a neural symbolic isolation architecture. This framework abandons semantic trust in natural language, forcing agents to losslessly formalize their intentions into first-order logical mathematical constraints before performing physical operations. Empirical evaluations of macroscopic and microscopic two-dimensional dynamic adversarial systems demonstrate that our formal verification mechanism achieves zero attack success rate and zero false positive rate across the evaluated scenarios, with extremely low computational latency. This research provides a conditional formal foundation under explicit system assumptions and an engineering paradigm for constructing the underlying defense foundation for future intelligent systems.

Xiaojian Yuan, Kejiang Chen, Jie Zhang et al.

Model inversion (MI) attacks have raised increasing concerns about privacy, which can reconstruct training data from public models. Indeed, MI attacks can be formalized as an optimization problem that seeks private data in a certain space. Recent MI attacks leverage a generative adversarial network (GAN) as an image prior to narrow the search space, and can successfully reconstruct even the high-dimensional data (e.g., face images). However, these generative MI attacks do not fully exploit the potential capabilities of the target model, still leading to a vague and coupled search space, i.e., different classes of images are coupled in the search space. Besides, the widely used cross-entropy loss in these attacks suffers from gradient vanishing. To address these problems, we propose Pseudo Label-Guided MI (PLG-MI) attack via conditional GAN (cGAN). At first, a top-n selection strategy is proposed to provide pseudo-labels for public data, and use pseudo-labels to guide the training of the cGAN. In this way, the search space is decoupled for different classes of images. Then a max-margin loss is introduced to improve the search process on the subspace of a target class. Extensive experiments demonstrate that our PLG-MI attack significantly improves the attack success rate and visual quality for various datasets and models, notably, 2~3 $\times$ better than state-of-the-art attacks under large distributional shifts. Our code is available at: https://github.com/LetheSec/PLG-MI-Attack.

Yao Zhu, Yuefeng Chen, Xiaodan Li et al.

Transferable adversarial attacks against Deep neural networks (DNNs) have received broad attention in recent years. An adversarial example can be crafted by a surrogate model and then attack the unknown target model successfully, which brings a severe threat to DNNs. The exact underlying reasons for the transferability are still not completely understood. Previous work mostly explores the causes from the model perspective, e.g., decision boundary, model architecture, and model capacity. adversarial attacks against Deep neural networks (DNNs) have received broad attention in recent years. An adversarial example can be crafted by a surrogate model and then attack the unknown target model successfully, which brings a severe threat to DNNs. The exact underlying reasons for the transferability are still not completely understood. Previous work mostly explores the causes from the model perspective. Here, we investigate the transferability from the data distribution perspective and hypothesize that pushing the image away from its original distribution can enhance the adversarial transferability. To be specific, moving the image out of its original distribution makes different models hardly classify the image correctly, which benefits the untargeted attack, and dragging the image into the target distribution misleads the models to classify the image as the target class, which benefits the targeted attack. Towards this end, we propose a novel method that crafts adversarial examples by manipulating the distribution of the image. We conduct comprehensive transferable attacks against multiple DNNs to demonstrate the effectiveness of the proposed method. Our method can significantly improve the transferability of the crafted attacks and achieves state-of-the-art performance in both untargeted and targeted scenarios, surpassing the previous best method by up to 40$\%$ in some cases.

Jiawei Zhao, Kejiang Chen, Xiaojian Yuan et al.

In recent years, the rapid development of large language models (LLMs) has achieved remarkable performance across various tasks. However, research indicates that LLMs are vulnerable to jailbreak attacks, where adversaries can induce the generation of harmful content through meticulously crafted prompts. This vulnerability poses significant challenges to the secure use and promotion of LLMs. Existing defense methods offer protection from different perspectives but often suffer from insufficient effectiveness or a significant impact on the model's capabilities. In this paper, we propose a plug-and-play and easy-to-deploy jailbreak defense framework, namely Prefix Guidance (PG), which guides the model to identify harmful prompts by directly setting the first few tokens of the model's output. This approach combines the model's inherent security capabilities with an external classifier to defend against jailbreak attacks. We demonstrate the effectiveness of PG across three models and five attack methods. Compared to baselines, our approach is generally more effective on average. Additionally, results on the Just-Eval benchmark further confirm PG's superiority to preserve the model's performance. our code is available at https://github.com/weiyezhimeng/Prefix-Guidance.

Chao Wang, Zijin Yang, Yaofei Wang et al.

Recent advancements in video generation technologies have been significant, resulting in their widespread application across multiple domains. However, concerns have been mounting over the potential misuse of generated content. Tracing the origin of generated videos has become crucial to mitigate potential misuse and identify responsible parties. Existing video attribution methods require additional operations or the training of source attribution models, which may degrade video quality or necessitate large amounts of training samples. To address these challenges, we define for the first time the "few-shot training-free generated video attribution" task and propose SWIFT, which is tightly integrated with the temporal characteristics of the video. By leveraging the "Pixel Frames(many) to Latent Frame(one)" temporal mapping within each video chunk, SWIFT applies a fixed-length sliding window to perform two distinct reconstructions: normal and corrupted. The variation in the losses between two reconstructions is then used as an attribution signal. We conducted an extensive evaluation of five state-of-the-art (SOTA) video generation models. Experimental results show that SWIFT achieves over 90% average attribution accuracy with merely 20 video samples across all models and even enables zero-shot attribution for HunyuanVideo, EasyAnimate, and Wan2.2. Our source code is available at https://github.com/wangchao0708/SWIFT.

Yaofei Wang, Rui Wang, Weilong Pang et al.

Generative linguistic steganography (GLS) enables covert communication by embedding secret messages into the natural language generation process. In practical deployment, however, GLS is vulnerable to tokenization ambiguity: the same surface text may be re-tokenized into a different token sequence at the receiver, breaking the shared decoding state between the communicating parties so that a single local mismatch can propagate into complete extraction failure. Existing solutions either remove ambiguous tokens -- distorting the generation distribution and compromising security -- or preserve the distribution at the cost of substantially reduced embedding capacity or prohibitive runtime overhead. To address this issue, we propose ReTokSync (Re-Tokenization Synchronization), a self-synchronizing disambiguation framework that monitors the receiver-view tokenization during generation and triggers a corrective reset only when ambiguity actually occurs. By confining the effect of tokenization ambiguity to sparse residual bit errors rather than global desynchronization, ReTokSync leaves ambiguity-free positions entirely untouched and remains compatible with the underlying steganographic algorithm. Experiments on both English and Chinese settings show that ReTokSync stays closest to the steganographic baseline in distributional security (zero KL divergence), text quality, embedding capacity, and runtime, while achieving extraction accuracy above 99.7\%. Building on this property, we further develop a two-channel covert communication mechanism in which ReTokSync serves as the primary channel and a reliable auxiliary channel corrects the remaining errors, achieving 100\% end-to-end recovery across all evaluated configurations.

Zexin Fan, Kejiang Chen, Kai Zeng et al.

Image steganography is a technique to conceal secret messages within digital images. Steganalysis, on the contrary, aims to detect the presence of secret messages within images. Recently, deep-learning-based steganalysis methods have achieved excellent detection performance. As a countermeasure, adversarial steganography has garnered considerable attention due to its ability to effectively deceive deep-learning-based steganalysis. However, steganalysts often employ unknown steganalytic models for detection. Therefore, the ability of adversarial steganography to deceive non-target steganalytic models, known as transferability, becomes especially important. Nevertheless, existing adversarial steganographic methods do not consider how to enhance transferability. To address this issue, we propose a novel adversarial steganographic scheme named Natias. Specifically, we first attribute the output of a steganalytic model to each neuron in the target middle layer to identify critical features. Next, we corrupt these critical features that may be adopted by diverse steganalytic models. Consequently, it can promote the transferability of adversarial steganography. Our proposed method can be seamlessly integrated with existing adversarial steganography frameworks. Thorough experimental analyses affirm that our proposed technique possesses improved transferability when contrasted with former approaches, and it attains heightened security in retraining scenarios.

Ki Sen Hung, Xi Yang, Chang Liu et al.

A central goal of LLM alignment is to balance helpfulness with harmlessness, yet these objectives conflict when the same knowledge serves both legitimate and malicious purposes. This tension is amplified by context-sensitive alignment: we observe that domain-specific contexts (e.g., chemistry) selectively relax defenses for domain-relevant harmful knowledge, while safety-research contexts (e.g., jailbreak studies) trigger broader relaxation spanning all harm categories. To systematically exploit this vulnerability, we propose Jargon, a framework combining safety-research contexts with multi-turn adversarial interactions that achieves attack success rates exceeding 93% across seven frontier models, including GPT-5.2, Claude-4.5, and Gemini-3, substantially outperforming existing methods. Activation space analysis reveals that Jargon queries occupy an intermediate region between benign and harmful inputs, a gray zone where refusal decisions become unreliable. To mitigate this vulnerability, we design a policy-guided safeguard that steers models toward helpful yet harmless responses, and internalize this capability through alignment fine-tuning, reducing attack success rates while preserving helpfulness.

Yang Yang, Yiyang Huang, Ming Shi et al.

Face privacy-preserving is one of the hotspots that arises dramatic interests of research. However, the existing face privacy-preserving methods aim at causing the missing of semantic information of face and cannot preserve the reusability of original facial information. To achieve the naturalness of the processed face and the recoverability of the original protected face, this paper proposes face privacy-preserving method based on Invertible "Mask" Network (IMN). In IMN, we introduce a Mask-net to generate "Mask" face firstly. Then, put the "Mask" face onto the protected face and generate the masked face, in which the masked face is indistinguishable from "Mask" face. Finally, "Mask" face can be put off from the masked face and obtain the recovered face to the authorized users, in which the recovered face is visually indistinguishable from the protected face. The experimental results show that the proposed method can not only effectively protect the privacy of the protected face, but also almost perfectly recover the protected face from the masked face.

Meng Tong, Kejiang Chen, Jie Zhang et al.

Large language models (LLMs), like ChatGPT, have greatly simplified text generation tasks. However, they have also raised concerns about privacy risks such as data leakage and unauthorized data collection. Existing solutions for privacy-preserving inference face practical challenges related to computation time and communication costs. In this paper, we propose InferDPT, the first practical framework for the privacy-preserving Inference of black-box LLMs, implementing Differential Privacy in Text generation. InferDPT comprises two key modules: the "perturbation module" utilizes the exponential mechanism to generate a perturbed prompt, facilitating privacy-preserving inference with black-box LLMs, and the "extraction module", inspired by knowledge distillation and retrieval-augmented generation, extracts coherent and consistent text from the perturbed generation result, ensuring successful text generation completion. To address privacy concerns related to previous exponential mechanisms' susceptibility to embedding revision attacks, we introduce RANTEXT, a novel differential privacy mechanism integrated into the perturbation module of InferDPT, which introduces the concept of "RANdom adjacency" for TEXT perturbation within the prompt. Experimental results across three datasets demonstrate that the text generation quality of InferDPT is comparable to that of non-private GPT-4, and RANTEXT surpasses existing state-of-the-art mechanisms, namely, SANTEXT+ and CUSTEXT+ in the trade-off between privacy and utility. Even with an privacy parameter epsilon value of 6.0, RANTEXT achieves an average privacy protection rate exceeding 90% against embedding revision attacks, which is 0.58 times higher than that of SANTEXT+ and 3.35 times higher than that of CUSTEXT+.

Zijin Yang, Yu Sun, Kejiang Chen et al.

Digital watermarking is essential for securing generated images from diffusion models. Accurate watermark evaluation is critical for algorithm development, yet existing methods have significant limitations: they lack a unified framework for both residual and semantic watermarks, provide results without interpretability, neglect comprehensive security considerations, and often use inappropriate metrics for semantic watermarks. To address these gaps, we propose WMVLM, the first unified and interpretable evaluation framework for diffusion model image watermarking via vision-language models (VLMs). We redefine quality and security metrics for each watermark type: residual watermarks are evaluated by artifact strength and erasure resistance, while semantic watermarks are assessed through latent distribution shifts. Moreover, we introduce a three-stage training strategy to progressively enable the model to achieve classification, scoring, and interpretable text generation. Experiments show WMVLM outperforms state-of-the-art VLMs with strong generalization across datasets, diffusion models, and watermarking methods.

Xin Zhang, Zijin Yang, Kejiang Chen et al.

Latent-based watermarks, integrated into the generation process of latent diffusion models (LDMs), simplify detection and attribution of generated images. However, recent black-box forgery attacks, where an attacker needs at least one watermarked image and black-box access to the provider's model, can embed the provider's watermark into images not produced by the provider, posing outsized risk to provenance and trust. We propose SemBind, the first defense framework for latent-based watermarks that resists black-box forgery by binding latent signals to image semantics via a learned semantic masker. Trained with contrastive learning, the masker yields near-invariant codes for the same prompt and near-orthogonal codes across prompts; these codes are reshaped and permuted to modulate the target latent before any standard latent-based watermark. SemBind is generally compatible with existing latent-based watermarking schemes and keeps image quality essentially unchanged, while a simple mask-ratio parameter offers a tunable trade-off between anti-forgery strength and robustness. Across four mainstream latent-based watermark methods, our SemBind-enabled anti-forgery variants markedly reduce false acceptance under black-box forgery while providing a controllable robustness-security balance.

Jiajie Lu, Zhenkan Fu, Na Zhao et al.

The rapid progress of generative AI has led to the emergence of new generative models, while existing detection methods struggle to keep pace, resulting in significant degradation in the detection performance. This highlights the urgent need for continuously updating AI-generated image detectors to adapt to new generators. To overcome low efficiency and catastrophic forgetting in detector updates, we propose LiteUpdate, a lightweight framework for updating AI-generated image detectors. LiteUpdate employs a representative sample selection module that leverages image confidence and gradient-based discriminative features to precisely select boundary samples. This approach improves learning and detection accuracy on new distributions with limited generated images, significantly enhancing detector update efficiency. Additionally, LiteUpdate incorporates a model merging module that fuses weights from multiple fine-tuning trajectories, including pre-trained, representative, and random updates. This balances the adaptability to new generators and mitigates the catastrophic forgetting of prior knowledge. Experiments demonstrate that LiteUpdate substantially boosts detection performance in various detectors. Specifically, on AIDE, the average detection accuracy on Midjourney improved from 87.63% to 93.03%, a 6.16% relative increase.

Chuan Cui, Kejiang Chen, Zhihua Wei et al.

The rapid advancement of AI-generated image (AIGI) models presents new challenges for evaluating image quality, particularly across three aspects: perceptual quality, prompt correspondence, and authenticity. To address these challenges, we introduce M3-AGIQA, a comprehensive framework that leverages Multimodal Large Language Models (MLLMs) to enable more human-aligned, holistic evaluation of AI-generated images across both visual and textual domains. Besides, our framework features a structured multi-round evaluation process, generating and analyzing intermediate image descriptions to provide deeper insight into these three aspects. By aligning model outputs more closely with human judgment, M3-AGIQA delivers robust and interpretable quality scores. Extensive experiments on multiple benchmarks demonstrate that our method achieves state-of-the-art performance on tested datasets and aspects, and exhibits strong generalizability in most cross-dataset settings. Code is available at https://github.com/strawhatboy/M3-AGIQA.

Jindong Yang, Han Fang, Weiming Zhang et al.

Diffusion models have advanced rapidly in recent years, producing high-fidelity images while raising concerns about intellectual property protection and the misuse of generative AI. Image watermarking for diffusion models, particularly Noise-as-Watermark (NaW) methods, encode watermark as specific standard Gaussian noise vector for image generation, embedding the infomation seamlessly while maintaining image quality. For detection, the generation process is inverted to recover the initial noise vector containing the watermark before extraction. However, existing NaW methods struggle to balance watermark robustness with generation diversity. Some methods achieve strong robustness by heavily constraining initial noise sampling, which degrades user experience, while others preserve diversity but prove too fragile for real-world deployment. To address this issue, we propose T2SMark, a two-stage watermarking scheme based on Tail-Truncated Sampling (TTS). Unlike prior methods that simply map bits to positive or negative values, TTS enhances robustness by embedding bits exclusively in the reliable tail regions while randomly sampling the central zone to preserve the latent distribution. Our two-stage framework then ensures sampling diversity by integrating a randomly generated session key into both encryption pipelines. We evaluate T2SMark on diffusion models with both U-Net and DiT backbones. Extensive experiments show that it achieves an optimal balance between robustness and diversity. Our code is available at \href{https://github.com/0xD009/T2SMark}{https://github.com/0xD009/T2SMark}.

Xiaojian Yuan, Kejiang Chen, Wen Huang et al.

The popularity of Machine Learning as a Service (MLaaS) has led to increased concerns about Model Stealing Attacks (MSA), which aim to craft a clone model by querying MLaaS. Currently, most research on MSA assumes that MLaaS can provide soft labels and that the attacker has a proxy dataset with a similar distribution. However, this fails to encapsulate the more practical scenario where only hard labels are returned by MLaaS and the data distribution remains elusive. Furthermore, most existing work focuses solely on stealing the model accuracy, neglecting the model robustness, while robustness is essential in security-sensitive scenarios, e.g., face-scan payment. Notably, improving model robustness often necessitates the use of expensive techniques such as adversarial training, thereby further making stealing robustness a more lucrative prospect. In response to these identified gaps, we introduce a novel Data-Free Hard-Label Robustness Stealing (DFHL-RS) attack in this paper, which enables the stealing of both model accuracy and robustness by simply querying hard labels of the target model without the help of any natural data. Comprehensive experiments demonstrate the effectiveness of our method. The clone model achieves a clean accuracy of 77.86% and a robust accuracy of 39.51% against AutoAttack, which are only 4.71% and 8.40% lower than the target model on the CIFAR-10 dataset, significantly exceeding the baselines. Our code is available at: https://github.com/LetheSec/DFHL-RS-Attack.

Haozhe Chen, Weiming Zhang, Kunlin Liu et al.

As an effective method for intellectual property (IP) protection, model watermarking technology has been applied on a wide variety of deep neural networks (DNN), including speech classification models. However, how to design a black-box watermarking scheme for automatic speech recognition (ASR) models is still an unsolved problem, which is a significant demand for protecting remote ASR Application Programming Interface (API) deployed in cloud servers. Due to conditional independence assumption and label-detection-based evasion attack risk of ASR models, the black-box model watermarking scheme for speech classification models cannot apply to ASR models. In this paper, we propose the first black-box model watermarking framework for protecting the IP of ASR models. Specifically, we synthesize trigger audios by spreading the speech clips of model owners over the entire input audios and labeling the trigger audios with the stego texts, which hides the authorship information with linguistic steganography. Experiments on the state-of-the-art open-source ASR system DeepSpeech demonstrate the feasibility of the proposed watermarking scheme, which is robust against five kinds of attacks and has little impact on accuracy.

Zijin Yang, Kai Zeng, Kejiang Chen et al.

Ethical concerns surrounding copyright protection and inappropriate content generation pose challenges for the practical implementation of diffusion models. One effective solution involves watermarking the generated images. However, existing methods often compromise the model performance or require additional training, which is undesirable for operators and users. To address this issue, we propose Gaussian Shading, a diffusion model watermarking technique that is both performance-lossless and training-free, while serving the dual purpose of copyright protection and tracing of offending content. Our watermark embedding is free of model parameter modifications and thus is plug-and-play. We map the watermark to latent representations following a standard Gaussian distribution, which is indistinguishable from latent representations obtained from the non-watermarked diffusion model. Therefore we can achieve watermark embedding with lossless performance, for which we also provide theoretical proof. Furthermore, since the watermark is intricately linked with image semantics, it exhibits resilience to lossy processing and erasure attempts. The watermark can be extracted by Denoising Diffusion Implicit Models (DDIM) inversion and inverse sampling. We evaluate Gaussian Shading on multiple versions of Stable Diffusion, and the results demonstrate that Gaussian Shading not only is performance-lossless but also outperforms existing methods in terms of robustness.

Yudong Li, Zhongliang Yang, Kejiang Chen et al.

In this work, we propose LiveSecBench, a dynamic and continuously updated safety benchmark specifically for Chinese-language LLM application scenarios. LiveSecBench evaluates models across six critical dimensions (Legality, Ethics, Factuality, Privacy, Adversarial Robustness, and Reasoning Safety) rooted in the Chinese legal and social frameworks. This benchmark maintains relevance through a dynamic update schedule that incorporates new threat vectors, such as the planned inclusion of Text-to-Image Generation Safety and Agentic Safety in the next update. For now, LiveSecBench (v251030) has evaluated 18 LLMs, providing a landscape of AI safety in the context of Chinese language. The leaderboard is publicly accessible at https://livesecbench.intokentech.cn/.

Jiyan He, Weitao Feng, Yaosen Min et al. · microsoft-research

The expanding application of Artificial Intelligence (AI) in scientific fields presents unprecedented opportunities for discovery and innovation. However, this growth is not without risks. AI models in science, if misused, can amplify risks like creation of harmful substances, or circumvention of established regulations. In this study, we aim to raise awareness of the dangers of AI misuse in science, and call for responsible AI development and use in this domain. We first itemize the risks posed by AI in scientific contexts, then demonstrate the risks by highlighting real-world examples of misuse in chemical science. These instances underscore the need for effective risk management strategies. In response, we propose a system called SciGuard to control misuse risks for AI models in science. We also propose a red-teaming benchmark SciMT-Safety to assess the safety of different systems. Our proposed SciGuard shows the least harmful impact in the assessment without compromising performance in benign tests. Finally, we highlight the need for a multidisciplinary and collaborative effort to ensure the safe and ethical use of AI models in science. We hope that our study can spark productive discussions on using AI ethically in science among researchers, practitioners, policymakers, and the public, to maximize benefits and minimize the risks of misuse.

Benlong Wu, Guoqiang Chen, Kejiang Chen et al.

Penetration testing is essential to ensure Web security, which can detect and fix vulnerabilities in advance, and prevent data leakage and serious consequences. The powerful inference capabilities of large language models (LLMs) have made significant progress in various fields, and the development potential of LLM-based agents can revolutionize the cybersecurity penetration testing industry. In this work, we establish a comprehensive end-to-end penetration testing benchmark using a real-world penetration testing environment to explore the capabilities of LLM-based agents in this domain. Our results reveal that the agents are familiar with the framework of penetration testing tasks, but they still face limitations in generating accurate commands and executing complete processes. Accordingly, we summarize the current challenges, including the difficulty of maintaining the entire message history and the tendency for the agent to become stuck. Based on the above insights, we propose a Penetration testing State Machine (PSM) that utilizes the Finite State Machine (FSM) methodology to address these limitations. Then, we introduce AutoPT, an automated penetration testing agent based on the principle of PSM driven by LLMs, which utilizes the inherent inference ability of LLM and the constraint framework of state machines. Our evaluation results show that AutoPT outperforms the baseline framework ReAct on the GPT-4o mini model and improves the task completion rate from 22% to 41% on the benchmark target. Compared with the baseline framework and manual work, AutoPT also reduces time and economic costs further. Hence, our AutoPT has facilitated the development of automated penetration testing and significantly impacted both academia and industry.

Yuang Qi, Kejiang Chen, Kai Zeng et al.

Recent research in provably secure neural linguistic steganography has overlooked a crucial aspect: the sender must detokenize stegotexts to avoid raising suspicion from the eavesdropper. The segmentation ambiguity problem, which arises when using language models based on subwords, leads to occasional decoding failures in all neural language steganography implementations based on these models. Current solutions to this issue involve altering the probability distribution of candidate words, rendering them incompatible with provably secure steganography. We propose a novel secure disambiguation method named SyncPool, which effectively addresses the segmentation ambiguity problem. We group all tokens with prefix relationships in the candidate pool before the steganographic embedding algorithm runs to eliminate uncertainty among ambiguous tokens. To enable the receiver to synchronize the sampling process of the sender, a shared cryptographically-secure pseudorandom number generator (CSPRNG) is deployed to select a token from the ambiguity pool. SyncPool does not change the size of the candidate pool or the distribution of tokens and thus is applicable to provably secure language steganography methods. We provide theoretical proofs and experimentally demonstrate the applicability of our solution to various languages and models, showing its potential to significantly improve the reliability and security of neural linguistic steganography systems.

Xiaofei Kong, Lei Li, Zhaoyun Chen et al.

Low-Rank Adaptation (LoRA) enables efficient fine-tuning of pre-trained language models through low-rank matrix approximation, achieving effectiveness in many scenarios. However, its representation capacity is constrained in complex tasks or high-rank dependency settings, potentially limiting model adaptability. To overcome the expressive bottleneck in classical low-rank approximation for fine-tuning large language models (LLMs), we propose Quantum Tensor Hybrid Adaptation (QTHA), a parameter-efficient fine-tuning method that integrates a quantum neural network (QNN) with a tensor network. QTHA explores quantum tensor hybrid fine-tuning within low-rank spaces by decomposing pre-trained weights into quantum neural network and tensor network representations, leveraging quantum state superposition to overcome classical rank limitations. Experiments demonstrate that QTHA achieves performance comparable to or surpassing LoRA in parameter-efficient fine-tuning. Compared to LoRA, QTHA reduces trainable parameters by 76% while reducing training loss by up to 17% and improving test set performance by up to 17% within the same training steps. This research not only enables lightweight adaptation of quantum resources to the billion-parameter models but also validates the feasibility of quantum hardware optimization driven by LLM tasks. It establishes the first engineering-ready foundation for future quantum-enhanced Artificial General Intelligence (AGI) systems.

Zijin Yang, Xin Zhang, Kejiang Chen et al.

Ethical concerns surrounding copyright protection and inappropriate content generation pose challenges for the practical implementation of diffusion models. One effective solution involves watermarking the generated images. Existing methods primarily focus on ensuring that watermark embedding does not degrade the model performance. However, they often overlook critical challenges in real-world deployment scenarios, such as the complexity of watermark key management, user-defined generation parameters, and the difficulty of verification by arbitrary third parties. To address this issue, we propose Gaussian Shading++, a diffusion model watermarking method tailored for real-world deployment. We propose a double-channel design that leverages pseudorandom error-correcting codes to encode the random seed required for watermark pseudorandomization, achieving performance-lossless watermarking under a fixed watermark key and overcoming key management challenges. Additionally, we model the distortions introduced during generation and inversion as an additive white Gaussian noise channel and employ a novel soft decision decoding strategy during extraction, ensuring strong robustness even when generation parameters vary. To enable third-party verification, we incorporate public key signatures, which provide a certain level of resistance against forgery attacks even when model inversion capabilities are fully disclosed. Extensive experiments demonstrate that Gaussian Shading++ not only maintains performance losslessness but also outperforms existing methods in terms of robustness, making it a more practical solution for real-world deployment.

Jun Jiang, Weiming Zhang, Nenghai Yu et al.

Linguistic steganography enables covert communication through embedding secret messages into innocuous texts; however, current methods face critical limitations in payload capacity and security. Traditional modification-based methods introduce detectable anomalies, while retrieval-based strategies suffer from low embedding capacity. Modern generative steganography leverages language models to generate natural stego text but struggles with limited entropy in token predictions, further constraining capacity. To address these issues, we propose an entropy-driven framework called RTMStega that integrates rank-based adaptive coding and context-aware decompression with normalized entropy. By mapping secret messages to token probability ranks and dynamically adjusting sampling via context-aware entropy-based adjustments, RTMStega achieves a balance between payload capacity and imperceptibility. Experiments across diverse datasets and models demonstrate that RTMStega triples the payload capacity of mainstream generative steganography, reduces processing time by over 50%, and maintains high text quality, offering a trustworthy solution for secure and efficient covert communication.

Wei Fan, Kejiang Chen, Chang Liu et al.

The rapid advancement of speech generation models has heightened privacy and security concerns related to voice cloning (VC). Recent studies have investigated disrupting unauthorized voice cloning by introducing adversarial perturbations. However, determined attackers can mitigate these protective perturbations and successfully execute VC. In this study, we conduct the first systematic evaluation of these protective perturbations against VC under realistic threat models that include perturbation purification. Our findings reveal that while existing purification methods can neutralize a considerable portion of the protective perturbations, they still lead to distortions in the feature space of VC models, which degrades the performance of VC. From this perspective, we propose a novel two-stage purification method: (1) Purify the perturbed speech; (2) Refine it using phoneme guidance to align it with the clean speech distribution. Experimental results demonstrate that our method outperforms state-of-the-art purification methods in disrupting VC defenses. Our study reveals the limitations of adversarial perturbation-based VC defenses and underscores the urgent need for more robust solutions to mitigate the security and privacy risks posed by VC. The code and audio samples are available at https://de-antifake.github.io.

Xiaoyun Zhang, Xiaojian Yuan, Di Huang et al.

Reasoning ability has become a defining capability of Large Language Models (LLMs), with Reinforcement Learning with Verifiable Rewards (RLVR) emerging as a key paradigm to enhance it. However, RLVR training often suffers from policy entropy collapse, where the policy becomes overly deterministic, hindering exploration and limiting reasoning performance. While entropy regularization is a common remedy, its effectiveness is highly sensitive to the fixed coefficient, making it unstable across tasks and models. In this work, we revisit entropy regularization in RLVR and argue that its potential has been largely underestimated. Our analysis shows that (i) tasks of varying difficulty demand distinct exploration intensities, and (ii) balanced exploration may require the policy entropy to be maintained within a moderate range below its initial level. Therefore, we propose Adaptive Entropy Regularization (AER)--a framework that dynamically balances exploration and exploitation via three components: difficulty-aware coefficient allocation, initial-anchored target entropy, and dynamic global coefficient adjustment. Experiments on multiple mathematical reasoning benchmarks show that AER consistently outperforms baselines, improving both reasoning accuracy and exploration capability.

Meng Tong, Yuntao Du, Kejiang Chen et al.

Membership inference attacks (MIAs) are widely used to assess the privacy risks associated with machine learning models. However, when these attacks are applied to pre-trained large language models (LLMs), they encounter significant challenges, including mislabeled samples, distribution shifts, and discrepancies in model size between experimental and real-world settings. To address these limitations, we introduce tokenizers as a new attack vector for membership inference. Specifically, a tokenizer converts raw text into tokens for LLMs. Unlike full models, tokenizers can be efficiently trained from scratch, thereby avoiding the aforementioned challenges. In addition, the tokenizer's training data is typically representative of the data used to pre-train LLMs. Despite these advantages, the potential of tokenizers as an attack vector remains unexplored. To this end, we present the first study on membership leakage through tokenizers and explore five attack methods to infer dataset membership. Extensive experiments on millions of Internet samples reveal the vulnerabilities in the tokenizers of state-of-the-art LLMs. To mitigate this emerging risk, we further propose an adaptive defense. Our findings highlight tokenizers as an overlooked yet critical privacy threat, underscoring the urgent need for privacy-preserving mechanisms specifically designed for them.

Wei Fan, Kejiang Chen, Xiangkun Wang et al.

Data hiding is essential for secure communication across digital media, and recent advances in Deep Neural Networks (DNNs) provide enhanced methods for embedding secret information effectively. However, previous audio hiding methods often result in unsatisfactory quality when recovering secret audio, due to their inherent limitations in the modeling of time-frequency relationships. In this paper, we explore these limitations and introduce a new DNN-based approach. We use a flow-based invertible neural network to establish a direct link between stego audio, cover audio, and secret audio, enhancing the reversibility of embedding and extracting messages. To address common issues from time-frequency transformations that degrade secret audio quality during recovery, we implement a time-frequency loss on the time-domain signal. This approach not only retains the benefits of time-frequency constraints but also enhances the reversibility of message recovery, which is vital for practical applications. We also add an encryption technique to protect the hidden data from unauthorized access. Experimental results on the VCTK and LibriSpeech datasets demonstrate that our method outperforms previous approaches in terms of subjective and objective metrics and exhibits robustness to various types of noise, suggesting its utility in targeted secure communication scenarios.

Xingkai Peng, Jun Jiang, Meng Tong et al.

Text-to-image (T2I) models have been widely applied in generating high-fidelity images across various domains. However, these models may also be abused to produce Not-Safe-for-Work (NSFW) content via jailbreak attacks. Existing jailbreak methods primarily manipulate the textual prompt, leaving potential vulnerabilities in image-based inputs largely unexplored. Moreover, text-based methods face challenges in bypassing the model's safety filters. In response to these limitations, we propose the Multimodal Prompt Decoupling Attack (MPDA), which utilizes image modality to separate the harmful semantic components of the original unsafe prompt. MPDA follows three core steps: firstly, a large language model (LLM) decouples unsafe prompts into pseudo-safe prompts and harmful prompts. The former are seemingly harmless sub-prompts that can bypass filters, while the latter are sub-prompts with unsafe semantics that trigger filters. Subsequently, the LLM rewrites the harmful prompts into natural adversarial prompts to bypass safety filters, which guide the T2I model to modify the base image into an NSFW output. Finally, to ensure semantic consistency between the generated NSFW images and the original unsafe prompts, the visual language model generates image captions, providing a new pathway to guide the LLM in iterative rewriting and refining the generated content.

Chao Wang, Kejiang Chen, Zijin Yang et al.

The rapid advancement of image-generation technologies has made it possible for anyone to create photorealistic images using generative models, raising significant security concerns. To mitigate malicious use, tracing the origin of such images is essential. Reconstruction-based attribution methods offer a promising solution, but they often suffer from reduced accuracy and high computational costs when applied to state-of-the-art (SOTA) models. To address these challenges, we propose AEDR (AutoEncoder Double-Reconstruction), a novel training-free attribution method designed for generative models with continuous autoencoders. Unlike existing reconstruction-based approaches that rely on the value of a single reconstruction loss, AEDR performs two consecutive reconstructions using the model's autoencoder, and adopts the ratio of these two reconstruction losses as the attribution signal. This signal is further calibrated using the image homogeneity metric to improve accuracy, which inherently cancels out absolute biases caused by image complexity, with autoencoder-based reconstruction ensuring superior computational efficiency. Experiments on eight top latent diffusion models show that AEDR achieves 25.5% higher attribution accuracy than existing reconstruction-based methods, while requiring only 1% of the computational time.

Shuai Li, Jie Zhang, Yuang Qi et al.

Large-scale image retrieval using deep hashing has become increasingly popular due to the exponential growth of image data and the remarkable feature extraction capabilities of deep neural networks (DNNs). However, deep hashing methods are vulnerable to malicious attacks, including adversarial and backdoor attacks. It is worth noting that these attacks typically involve altering the query images, which is not a practical concern in real-world scenarios. In this paper, we point out that even clean query images can be dangerous, inducing malicious target retrieval results, like undesired or illegal images. To the best of our knowledge, we are the first to study data \textbf{p}oisoning \textbf{a}ttacks against \textbf{d}eep \textbf{hash}ing \textbf{(\textit{PADHASH})}. Specifically, we first train a surrogate model to simulate the behavior of the target deep hashing model. Then, a strict gradient matching strategy is proposed to generate the poisoned images. Extensive experiments on different models, datasets, hash methods, and hash code lengths demonstrate the effectiveness and generality of our attack method.

Yuang Qi, Kejiang Chen, Na Zhao et al.

The rapid development of image generation models has facilitated the widespread dissemination of generated images on social networks, creating favorable conditions for provably secure image steganography. However, existing methods face issues such as low quality of generated images and lack of semantic control in the generation process. To leverage provably secure steganography with more effective and high-performance image generation models, and to ensure that stego images can accurately extract secret messages even after being uploaded to social networks and subjected to lossy processing such as JPEG compression, we propose a high-quality, provably secure, and robust image steganography method based on state-of-the-art autoregressive (AR) image generation models using Vector-Quantized (VQ) tokenizers. Additionally, we employ a cross-modal error-correction framework that generates stego text from stego images to aid in restoring lossy images, ultimately enabling the extraction of secret messages embedded within the images. Extensive experiments have demonstrated that the proposed method provides advantages in stego quality, embedding capacity, and robustness, while ensuring provable undetectability.

Xiao Yu, Yuang Qi, Kejiang Chen et al.

Large language models (LLMs) have the potential to generate texts that pose risks of misuse, such as plagiarism, planting fake reviews on e-commerce platforms, or creating inflammatory false tweets. Consequently, detecting whether a text is generated by LLMs has become increasingly important. Existing high-quality detection methods usually require access to the interior of the model to extract the intrinsic characteristics. However, since we do not have access to the interior of the black-box model, we must resort to surrogate models, which impacts detection quality. In order to achieve high-quality detection of black-box models, we would like to extract deep intrinsic characteristics of the black-box model generated texts. We view the generation process as a coupled process of prompt and intrinsic characteristics of the generative model. Based on this insight, we propose to decouple prompt and intrinsic characteristics (DPIC) for LLM-generated text detection method. Specifically, given a candidate text, DPIC employs an auxiliary LLM to reconstruct the prompt corresponding to the candidate text, then uses the prompt to regenerate text by the auxiliary LLM, which makes the candidate text and the regenerated text align with their prompts, respectively. Then, the similarity between the candidate text and the regenerated text is used as a detection feature, thus eliminating the prompt in the detection process, which allows the detector to focus on the intrinsic characteristics of the generative model. Compared to the baselines, DPIC has achieved an average improvement of 6.76\% and 2.91\% in detecting texts from different domains generated by GPT4 and Claude3, respectively.

Xi Yang, Kejiang Chen, Weiming Zhang et al.

LLMs now exhibit human-like skills in various fields, leading to worries about misuse. Thus, detecting generated text is crucial. However, passive detection methods are stuck in domain specificity and limited adversarial robustness. To achieve reliable detection, a watermark-based method was proposed for white-box LLMs, allowing them to embed watermarks during text generation. The method involves randomly dividing the model vocabulary to obtain a special list and adjusting the probability distribution to promote the selection of words in the list. A detection algorithm aware of the list can identify the watermarked text. However, this method is not applicable in many real-world scenarios where only black-box language models are available. For instance, third-parties that develop API-based vertical applications cannot watermark text themselves because API providers only supply generated text and withhold probability distributions to shield their commercial interests. To allow third-parties to autonomously inject watermarks into generated text, we develop a watermarking framework for black-box language model usage scenarios. Specifically, we first define a binary encoding function to compute a random binary encoding corresponding to a word. The encodings computed for non-watermarked text conform to a Bernoulli distribution, wherein the probability of a word representing bit-1 being approximately 0.5. To inject a watermark, we alter the distribution by selectively replacing words representing bit-0 with context-based synonyms that represent bit-1. A statistical test is then used to identify the watermark. Experiments demonstrate the effectiveness of our method on both Chinese and English datasets. Furthermore, results under re-translation, polishing, word deletion, and synonym substitution attacks reveal that it is arduous to remove the watermark without compromising the original semantics.

Kejiang Chen, Xianhan Zeng, Qichao Ying et al.

Deep learning has achieved enormous success in various industrial applications. Companies do not want their valuable data to be stolen by malicious employees to train pirated models. Nor do they wish the data analyzed by the competitors after using them online. We propose a novel solution for dataset protection in this scenario by robustly and reversibly transform the images into adversarial images. We develop a reversible adversarial example generator (RAEG) that introduces slight changes to the images to fool traditional classification models. Even though malicious attacks train pirated models based on the defensed versions of the protected images, RAEG can significantly weaken the functionality of these models. Meanwhile, the reversibility of RAEG ensures the performance of authorized models. Extensive experiments demonstrate that RAEG can better protect the data with slight distortion against adversarial defense than previous methods.

Xi Yang, Jie Zhang, Kejiang Chen et al.

Text content created by humans or language models is often stolen or misused by adversaries. Tracing text provenance can help claim the ownership of text content or identify the malicious users who distribute misleading content like machine-generated fake news. There have been some attempts to achieve this, mainly based on watermarking techniques. Specifically, traditional text watermarking methods embed watermarks by slightly altering text format like line spacing and font, which, however, are fragile to cross-media transmissions like OCR. Considering this, natural language watermarking methods represent watermarks by replacing words in original sentences with synonyms from handcrafted lexical resources (e.g., WordNet), but they do not consider the substitution's impact on the overall sentence's meaning. Recently, a transformer-based network was proposed to embed watermarks by modifying the unobtrusive words (e.g., function words), which also impair the sentence's logical and semantic coherence. Besides, one well-trained network fails on other different types of text content. To address the limitations mentioned above, we propose a natural language watermarking scheme based on context-aware lexical substitution (LS). Specifically, we employ BERT to suggest LS candidates by inferring the semantic relatedness between the candidates and the original sentence. Based on this, a selection strategy in terms of synchronicity and substitutability is further designed to test whether a word is exactly suitable for carrying the watermark signal. Extensive experiments demonstrate that, under both objective and subjective metrics, our watermarking scheme can well preserve the semantic integrity of original sentences and has a better transferability than existing methods. Besides, the proposed LS approach outperforms the state-of-the-art approach on the Stanford Word Substitution Benchmark.

Kai Zeng, Kejiang Chen, Yaofei Wang et al.

Nowadays, people are sharing their pictures on online social networks (OSNs), so OSN is a good platform for Steganography. But OSNs usually perform JPEG compression on the uploaded image, which will invalidate most of the existing steganography algorithms. Recently, some works try to design robust steganography which can resist JPEG compression, such as Dither Modulation-based robust Adaptive Steganography (DMAS) and Generalized dither Modulation-based robust Adaptive Steganography (GMAS). They relieve the problem that the receivers cannot extract the message correctly when the quality factor of channel JPEG compression is larger than that of cover images. However, they only can realize limited resistance to detection and compression due to robust domain selection. To overcome this problem, we meticulously explore three lossy operations in the JPEG recompression and discover that the key problem is spatial overflow. Then two preprocessing methods Overall Scaling (OS) and Specific Truncation (ST) are presented to remove overflow before message embedding as well as generate a reference image. The reference image is employed as the guidance to build asymmetric distortion for removing overflow during embedding. Experimental results show that the proposed methods significantly surpass GMAS in terms of security and achieve comparable robustness.

Hang Zhou, Weiming Zhang, Kejiang Chen et al.

Three-dimensional (3-D) meshes are commonly used to represent virtual surfaces and volumes. Over the past decade, 3-D meshes have emerged in industrial, medical, and entertainment applications, being of large practical significance for 3-D mesh steganography and steganalysis. In this article, we provide a systematic survey of the literature on 3-D mesh steganography and steganalysis. Compared with an earlier survey [1], we propose a new taxonomy of steganographic algorithms with four categories: 1) two-state domain, 2) LSB domain, 3) permutation domain, and 4) transform domain. Regarding steganalysis algorithms, we divide them into two categories: 1) universal steganalysis and 2) specific steganalysis. For each category, the history of technical developments and the current technological level are introduced and discussed. Finally, we highlight some promising future research directions and challenges in improving the performance of 3-D mesh steganography and steganalysis.

Chaoning Zhang, Chenguo Lin, Philipp Benz et al.

Data hiding is the art of concealing messages with limited perceptual changes. Recently, deep learning has enriched it from various perspectives with significant progress. In this work, we conduct a brief yet comprehensive review of existing literature for deep learning based data hiding (deep hiding) by first classifying it according to three essential properties (i.e., capacity, security and robustness), and outline three commonly used architectures. Based on this, we summarize specific strategies for different applications of data hiding, including basic hiding, steganography, watermarking and light field messaging. Finally, further insight into deep hiding is provided by incorporating the perspective of adversarial attack.

Kejiang Chen, Yuefeng Chen, Hang Zhou et al.

Deep neural networks have been proved that they are vulnerable to adversarial examples, which are generated by adding human-imperceptible perturbations to images. To defend these adversarial examples, various detection based methods have been proposed. However, most of them perform poorly on detecting adversarial examples with extremely slight perturbations. By exploring these adversarial examples, we find that there exists compliance between perturbations and prediction confidence, which guides us to detect few-perturbation attacks from the aspect of prediction confidence. To detect both few-perturbation attacks and large-perturbation attacks, we propose a method beyond image space by a two-stream architecture, in which the image stream focuses on the pixel artifacts and the gradient stream copes with the confidence artifacts. The experimental results show that the proposed method outperforms the existing methods under oblivious attacks and is verified effective to defend omniscient attacks as well.

Hang Zhou, Dongdong Chen, Jing Liao et al.

Deep neural networks have made tremendous progress in 3D point-cloud recognition. Recent works have shown that these 3D recognition networks are also vulnerable to adversarial samples produced from various attack methods, including optimization-based 3D Carlini-Wagner attack, gradient-based iterative fast gradient method, and skeleton-detach based point-dropping. However, after a careful analysis, these methods are either extremely slow because of the optimization/iterative scheme, or not flexible to support targeted attack of a specific category. To overcome these shortcomings, this paper proposes a novel label guided adversarial network (LG-GAN) for real-time flexible targeted point cloud attack. To the best of our knowledge, this is the first generation based 3D point cloud attack method. By feeding the original point clouds and target attack label into LG-GAN, it can learn how to deform the point clouds to mislead the recognition network into the specific label only with a single forward pass. In detail, LGGAN first leverages one multi-branch adversarial network to extract hierarchical features of the input point clouds, then incorporates the specified label information into multiple intermediate features using the label encoder. Finally, the encoded features will be fed into the coordinate reconstruction decoder to generate the target adversarial sample. By evaluating different point-cloud recognition models (e.g., PointNet, PointNet++ and DGCNN), we demonstrate that the proposed LG-GAN can support flexible targeted attack on the fly while guaranteeing good attack performance and higher efficiency simultaneously.

Kejiang Chen, Hang Zhou, Yuefeng Chen et al.

Recent work has demonstrated that neural networks are vulnerable to adversarial examples. To escape from the predicament, many works try to harden the model in various ways, in which adversarial training is an effective way which learns robust feature representation so as to resist adversarial attacks. Meanwhile, the self-supervised learning aims to learn robust and semantic embedding from data itself. With these views, we introduce self-supervised learning to against adversarial examples in this paper. Specifically, the self-supervised representation coupled with k-Nearest Neighbour is proposed for classification. To further strengthen the defense ability, self-supervised adversarial training is proposed, which maximizes the mutual information between the representations of original examples and the corresponding adversarial examples. Experimental results show that the self-supervised representation outperforms its supervised version in respect of robustness and self-supervised adversarial training can further improve the defense ability efficiently.

Hang Zhou, Kejiang Chen, Weiming Zhang et al.

Neural networks are vulnerable to adversarial examples, which poses a threat to their application in security sensitive systems. We propose a Denoiser and UPsampler Network (DUP-Net) structure as defenses for 3D adversarial point cloud classification, where the two modules reconstruct surface smoothness by dropping or adding points. In this paper, statistical outlier removal (SOR) and a data-driven upsampling network are considered as denoiser and upsampler respectively. Compared with baseline defenses, DUP-Net has three advantages. First, with DUP-Net as a defense, the target model is more robust to white-box adversarial attacks. Second, the statistical outlier removal provides added robustness since it is a non-differentiable denoising operation. Third, the upsampler network can be trained on a small dataset and defends well against adversarial attacks generated from other point cloud datasets. We conduct various experiments to validate that DUP-Net is very effective as defense in practice. Our best defense eliminates 83.8% of C&W and l_2 loss based attack (point shifting), 50.0% of C&W and Hausdorff distance loss based attack (point adding) and 9.0% of saliency map based attack (point dropping) under 200 dropped points on PointNet.

Kejiang Chen, Hang Zhou, Hanqing Zhao et al.

Steganography is the art and science of hiding secret messages in public communication so that the presence of the secret messages cannot be detected. There are two distribution-preserving steganographic frameworks, one is sampling-based and the other is compression-based. The former requires a perfect sampler which yields data following the same distribution, and the latter needs explicit distribution of generative objects. However, these two conditions are too strict even unrealistic in the traditional data environment, e.g. the distribution of natural images is hard to seize. Fortunately, generative models bring new vitality to distribution-preserving steganography, which can serve as the perfect sampler or provide the explicit distribution of generative media. Take text-to-speech generation task as an example, we propose distribution-preserving steganography based on WaveGlow and WaveNet, which corresponds to the former two categories. Steganalysis experiments and theoretical analysis are conducted to demonstrate that the proposed methods can preserve the distribution.