Yunsi Fei

Ruyi Ding, Cheng Gongye, Siyue Wang et al.

Deep Neural Networks (DNN) are vulnerable to adversarial perturbations-small changes crafted deliberately on the input to mislead the model for wrong predictions. Adversarial attacks have disastrous consequences for deep learning-empowered critical applications. Existing defense and detection techniques both require extensive knowledge of the model, testing inputs, and even execution details. They are not viable for general deep learning implementations where the model internal is unknown, a common 'black-box' scenario for model users. Inspired by the fact that electromagnetic (EM) emanations of a model inference are dependent on both operations and data and may contain footprints of different input classes, we propose a framework, EMShepherd, to capture EM traces of model execution, perform processing on traces and exploit them for adversarial detection. Only benign samples and their EM traces are used to train the adversarial detector: a set of EM classifiers and class-specific unsupervised anomaly detectors. When the victim model system is under attack by an adversarial example, the model execution will be different from executions for the known classes, and the EM trace will be different. We demonstrate that our air-gapped EMShepherd can effectively detect different adversarial attacks on a commonly used FPGA deep learning accelerator for both Fashion MNIST and CIFAR-10 datasets. It achieves a 100% detection rate on most types of adversarial samples, which is comparable to the state-of-the-art 'white-box' software-based detectors.

Ruyi Ding, Shijin Duan, Xiaolin Xu et al.

Graph neural networks (GNNs) have brought superb performance to various applications utilizing graph structural data, such as social analysis and fraud detection. The graph links, e.g., social relationships and transaction history, are sensitive and valuable information, which raises privacy concerns when using GNNs. To exploit these vulnerabilities, we propose VertexSerum, a novel graph poisoning attack that increases the effectiveness of graph link stealing by amplifying the link connectivity leakage. To infer node adjacency more accurately, we propose an attention mechanism that can be embedded into the link detection network. Our experiments demonstrate that VertexSerum significantly outperforms the SOTA link inference attack, improving the AUC scores by an average of $9.8\%$ across four real-world datasets and three different GNN structures. Furthermore, our experiments reveal the effectiveness of VertexSerum in both black-box and online learning settings, further validating its applicability in real-world scenarios.

Ruyi Ding, Tianhong Xu, Xinyi Shen et al.

The transformer architecture has become a cornerstone of modern AI, fueling remarkable progress across applications in natural language processing, computer vision, and multimodal learning. As these models continue to scale explosively for performance, implementation efficiency remains a critical challenge. Mixture of Experts (MoE) architectures, selectively activating specialized subnetworks (experts), offer a unique balance between model accuracy and computational cost. However, the adaptive routing in MoE architectures, where input tokens are dynamically directed to specialized experts based on their semantic meaning inadvertently opens up a new attack surface for privacy breaches. These input-dependent activation patterns leave distinctive temporal and spatial traces in hardware execution, which adversaries could exploit to deduce sensitive user data. In this work, we propose MoEcho, discovering a side channel analysis based attack surface that compromises user privacy on MoE based systems. Specifically, in MoEcho, we introduce four novel architectural side channels on different computing platforms, including Cache Occupancy Channels and Pageout+Reload on CPUs, and Performance Counter and TLB Evict+Reload on GPUs, respectively. Exploiting these vulnerabilities, we propose four attacks that effectively breach user privacy in large language models (LLMs) and vision language models (VLMs) based on MoE architectures: Prompt Inference Attack, Response Reconstruction Attack, Visual Inference Attack, and Visual Reconstruction Attack. MoEcho is the first runtime architecture level security analysis of the popular MoE structure common in modern transformers, highlighting a serious security and privacy threat and calling for effective and timely safeguards when harnessing MoE based models for developing efficient large scale AI services.

Tong Zhou, Ruyi Ding, Gaowen Liu et al.

The rapid growth of digital and AI-generated images has amplified the need for secure and verifiable methods of image attribution. While digital watermarking offers more robust protection than metadata-based approaches--which can be easily stripped--current watermarking techniques remain vulnerable to forgery, creating risks of misattribution that can damage the reputations of AI model developers and the rights of digital artists. These vulnerabilities arise from two key issues: (1) content-agnostic watermarks, which, once learned or leaked, can be transferred across images to fake attribution, and (2) reliance on detector-based verification, which is unreliable since detectors can be tricked. We present MetaSeal, a novel framework for content-dependent watermarking with cryptographic security guarantees to safeguard image attribution. Our design provides (1) forgery resistance, preventing unauthorized replication and enforcing cryptographic verification; (2) robust, self-contained protection, embedding attribution directly into images while maintaining resilience against benign transformations; and (3) evidence of tampering, making malicious alterations visually detectable. Experiments demonstrate that MetaSeal effectively mitigates forgery attempts and applies to both natural and AI-generated images, establishing a new standard for secure image attribution.

Ruyi Ding, Tianhong Xu, Aidong Adam Ding et al.

Wide deployment of machine learning models on edge devices has rendered the model intellectual property (IP) and data privacy vulnerable. We propose GNNVault, the first secure Graph Neural Network (GNN) deployment strategy based on Trusted Execution Environment (TEE). GNNVault follows the design of 'partition-before-training' and includes a private GNN rectifier to complement with a public backbone model. This way, both critical GNN model parameters and the private graph used during inference are protected within secure TEE compartments. Real-world implementations with Intel SGX demonstrate that GNNVault safeguards GNN inference against state-of-the-art link stealing attacks with negligible accuracy degradation (<2%).

Amel Nestor Docena, Thomas Wahl, Trevor Pearce et al.

Neural Networks are used today in numerous security- and safety-relevant domains and are, as such, a popular target of attacks that subvert their classification capabilities, by manipulating the network parameters. Prior work has introduced sensitive samples -- inputs highly sensitive to parameter changes -- to detect such manipulations, and proposed a gradient ascent-based approach to compute them. In this paper we offer an alternative, using symbolic constraint solvers. We model the network and a formal specification of a sensitive sample in the language of the solver and ask for a solution. This approach supports a rich class of queries, corresponding, for instance, to the presence of certain types of attacks. Unlike earlier techniques, our approach does not depend on convex search domains, or on the suitability of a starting point for the search. We address the performance limitations of constraint solvers by partitioning the search space for the solver, and exploring the partitions according to a balanced schedule that still retains completeness of the search. We demonstrate the impact of the use of solvers in terms of functionality and search efficiency, using a case study for the detection of Trojan attacks on Neural Networks.

Yukui Luo, Cheng Gongye, Yunsi Fei et al.

As Field-programmable gate arrays (FPGAs) are widely adopted in clouds to accelerate Deep Neural Networks (DNN), such virtualization environments have posed many new security issues. This work investigates the integrity of DNN FPGA accelerators in clouds. It proposes DeepStrike, a remotely-guided attack based on power glitching fault injections targeting DNN execution. We characterize the vulnerabilities of different DNN layers against fault injections on FPGAs and leverage time-to-digital converter (TDC) sensors to precisely control the timing of fault injections. Experimental results show that our proposed attack can successfully disrupt the FPGA DSP kernel and misclassify the target victim DNN application.

Elmira Karimi, Yunsi Fei, David Kaeli

GPUs are increasingly being used in security applications, especially for accelerating encryption/decryption. While GPUs are an attractive platform in terms of performance, the security of these devices raises a number of concerns. One vulnerability is the data-dependent timing information, which can be exploited by adversary to recover the encryption key. Memory system features are frequently exploited since they create detectable timing variations. In this paper, our attack model is a coalescing attack, which leverages a critical GPU microarchitectural feature -- the coalescing unit. As multiple concurrent GPU memory requests can refer to the same cache block, the coalescing unit collapses them into a single memory transaction. The access time of an encryption kernel is dependent on the number of transactions. Correlation between a guessed key value and the associated timing samples can be exploited to recover the secret key. In this paper, a series of hardware/software countermeasures are proposed to obfuscate the memory timing side channel, making the GPU more resilient without impacting performance. Our hardware-based approach attempts to randomize the width of the coalescing unit to lower the signal-to-noise ratio. We present a hierarchical Miss Status Holding Register (MSHR) design that can merge transactions across different warps. This feature boosts performance, while, at the same time, secures the execution. We also present a software-based approach to permute the organization of critical data structures, significantly changing the coalescing behavior and introducing a high degree of randomness. Equipped with our new protections, the effort to launch a successful attack is increased up to 1433X . 178X, while also improving encryption/decryption performance up to 7%.

Johann Knechtel, Elif Bilge Kavun, Francesco Regazzoni et al.

Modern electronic systems become evermore complex, yet remain modular, with integrated circuits (ICs) acting as versatile hardware components at their heart. Electronic design automation (EDA) for ICs has focused traditionally on power, performance, and area. However, given the rise of hardware-centric security threats, we believe that EDA must also adopt related notions like secure by design and secure composition of hardware. Despite various promising studies, we argue that some aspects still require more efforts, for example: effective means for compilation of assumptions and constraints for security schemes, all the way from the system level down to the "bare metal"; modeling, evaluation, and consideration of security-relevant metrics; or automated and holistic synthesis of various countermeasures, without inducing negative cross-effects. In this paper, we first introduce hardware security for the EDA community. Next we review prior (academic) art for EDA-driven security evaluation and implementation of countermeasures. We then discuss strategies and challenges for advancing research and development toward secure composition of circuits and systems.

Pu Zhao, Siyue Wang, Cheng Gongye et al.

Despite the great achievements of deep neural networks (DNNs), the vulnerability of state-of-the-art DNNs raises security concerns of DNNs in many application domains requiring high reliability.We propose the fault sneaking attack on DNNs, where the adversary aims to misclassify certain input images into any target labels by modifying the DNN parameters. We apply ADMM (alternating direction method of multipliers) for solving the optimization problem of the fault sneaking attack with two constraints: 1) the classification of the other images should be unchanged and 2) the parameter modifications should be minimized. Specifically, the first constraint requires us not only to inject designated faults (misclassifications), but also to hide the faults for stealthy or sneaking considerations by maintaining model accuracy. The second constraint requires us to minimize the parameter modifications (using L0 norm to measure the number of modifications and L2 norm to measure the magnitude of modifications). Comprehensive experimental evaluation demonstrates that the proposed framework can inject multiple sneaking faults without losing the overall test accuracy performance.