Research Papers

Modern face recognition systems leverage datasets containing images of hundreds of thousands of specific individuals' faces to train deep convolutional neural networks to learn an embedding space that maps an arbitrary individual's face to a vector representation of their identity. The performance of a face recognition system in face verification (1:1) and face identification (1:N) tasks is directly related to the ability of an embedding space to discriminate between identities. Recently, there has been significant public scrutiny into the source and privacy implications of large-scale face recognition training datasets such as MS-Celeb-1M and MegaFace, as many people are uncomfortable with their face being used to train dual-use technologies that can enable mass surveillance. However, the impact of an individual's inclusion in training data on a derived system's ability to recognize them has not previously been studied. In this work, we audit ArcFace, a state-of-the-art, open source face recognition system, in a large-scale face identification experiment with more than one million distractor images. We find a Rank-1 face identification accuracy of 79.71% for individuals present in the model's training data and an accuracy of 75.73% for those not present. This modest difference in accuracy demonstrates that face recognition systems using deep learning work better for individuals they are trained on, which has serious privacy implications when one considers all major open source face recognition training datasets do not obtain informed consent from individuals during their collection.

Establishing unique identities for both humans and end systems has been an active research problem in the security community, giving rise to innovative machine learning-based authentication techniques. Although such techniques offer an automated method to establish identity, they have not been vetted against sophisticated attacks that target their core machine learning technique. This paper demonstrates that mimicking the unique signatures generated by host fingerprinting and biometric authentication systems is possible. We expose the ineffectiveness of underlying machine learning classification models by constructing a blind attack based around the query synthesis framework and utilizing Explainable-AI (XAI) techniques. We launch an attack in under 130 queries on a state-of-the-art face authentication system, and under 100 queries on a host authentication system. We examine how these attacks can be defended against and explore their limitations. XAI provides an effective means for adversaries to infer decision boundaries and provides a new way forward in constructing attacks against systems using machine learning models for authentication.

Tree-based Dialogue Reinforced Policy Optimization for Red-Teaming Attacks

Ruohao Guo, Afshin Oroojlooy, Roshan Sridhar +3

Despite recent rapid progress in AI safety, current large language models remain vulnerable to adversarial attacks in multi-turn interaction settings, where attackers strategically adapt their prompts across conversation turns and pose a more critical yet realistic challenge. Existing approaches that discover safety vulnerabilities either rely on manual red-teaming with human experts or employ automated methods using pre-defined templates and human-curated attack data, with most focusing on single-turn attacks. However, these methods did not explore the vast space of possible multi-turn attacks, failing to consider novel attack trajectories that emerge from complex dialogue dynamics and strategic conversation planning. This gap is particularly critical given recent findings that LLMs exhibit significantly higher vulnerability to multi-turn attacks compared to single-turn attacks. We propose DialTree, an on-policy reinforcement learning framework integrated with tree search that autonomously discovers diverse multi-turn attack strategies by treating the dialogue as a sequential decision-making problem, enabling systematic exploration without manually curated data. Through extensive experiments, our approach not only achieves more than 44.2% higher ASR across 12 target models compared to previous state-of-the-art approaches, but also effectively uncovers new attack strategies by learning optimal dialogue policies that maximize attack success across multiple turns.

Benchmarking Large Language Models for Cryptanalysis and Side-Channel Vulnerabilities

Utsav Maskey, Chencheng Zhu, Usman Naseem

Recent advancements in large language models (LLMs) have transformed natural language understanding and generation, leading to extensive benchmarking across diverse tasks. However, cryptanalysis - a critical area for data security and its connection to LLMs' generalization abilities - remains underexplored in LLM evaluations. To address this gap, we evaluate the cryptanalytic potential of state-of-the-art LLMs on ciphertexts produced by a range of cryptographic algorithms. We introduce a benchmark dataset of diverse plaintexts, spanning multiple domains, lengths, writing styles, and topics, paired with their encrypted versions. Using zero-shot and few-shot settings along with chain-of-thought prompting, we assess LLMs' decryption success rate and discuss their comprehension abilities. Our findings reveal key insights into LLMs' strengths and limitations in side-channel scenarios and raise concerns about their susceptibility to under-generalization-related attacks. This research highlights the dual-use nature of LLMs in security contexts and contributes to the ongoing discussion on AI safety and security.

To safeguard user data privacy, on-device inference has emerged as a prominent paradigm on mobile and Internet of Things (IoT) devices. This paradigm involves deploying a model provided by a third party on local devices to perform inference tasks. However, it exposes the private model to two primary security threats: model stealing (MS) and membership inference attacks (MIA). To mitigate these risks, existing wisdom deploys models within Trusted Execution Environments (TEEs), which is a secure isolated execution space. Nonetheless, the constrained secure memory capacity in TEEs makes it challenging to achieve full model security with low inference latency. This paper fills the gap with TensorShield, the first efficient on-device inference work that shields partial tensors of the model while still fully defending against MS and MIA. The key enabling techniques in TensorShield include: (i) a novel eXplainable AI (XAI) technique exploits the model's attention transition to assess critical tensors and shields them in TEE to achieve secure inference, and (ii) two meticulous designs with critical feature identification and latency-aware placement to accelerate inference while maintaining security. Extensive evaluations show that TensorShield delivers almost the same security protection as shielding the entire model inside TEE, while being up to 25.35$\times$ (avg. 5.85$\times$) faster than the state-of-the-art work, without accuracy loss.

Recent advances in diffusion models have enabled the creation of deceptively real images, posing significant security risks when misused. In this study, we empirically show that different timesteps of DDIM inversion reveal varying subtle distinctions between synthetic and real images that are extractable for detection, in the forms of such as Fourier power spectrum high-frequency discrepancies and inter-pixel variance distributions. Based on these observations, we propose a novel synthetic image detection method that directly utilizes features of intermediately noised images by training an ensemble on multiple noised timesteps, circumventing conventional reconstruction-based strategies. To enhance human comprehension, we introduce a metric-grounded explanation generation and refinement module to identify and explain AI-generated flaws. Additionally, we construct the GenHard and GenExplain benchmarks to provide detection samples of greater difficulty and high-quality rationales for fake images. Extensive experiments show that our method achieves state-of-the-art performance with 98.91% and 95.89% detection accuracy on regular and challenging samples respectively, and demonstrates generalizability and robustness. Our code and datasets are available at https://github.com/Shadowlized/ESIDE.

The evolution of wireless communication systems will be fundamentally impacted by an open radio access network (O-RAN), a new concept defining an intelligent architecture with enhanced flexibility, openness, and the ability to slice services more efficiently. For all its promises, and like any technological advancement, O-RAN is not without risks that need to be carefully assessed and properly addressed to accelerate its wide adoption in future mobile networks. In this paper, we present an in-depth security analysis of the O-RAN architecture, discussing the potential threats that may arise in the different O-RAN architecture layers and their impact on the Confidentiality, Integrity, and Availability (CIA) triad. We also promote the potential of zero trust, Moving Target Defense (MTD), blockchain, and large language models(LLM) technologies in fortifying O-RAN's security posture. Furthermore, we numerically demonstrate the effectiveness of MTD in empowering robust deep reinforcement learning methods for dynamic network slice admission control in the O-RAN architecture. Moreover, we examine the effect of explainable AI (XAI) based on LLMs in securing the system.

Chain-of-Jailbreak Attack for Image Generation Models via Editing Step by Step

Wenxuan Wang, Kuiyi Gao, Youliang Yuan +5

Text-based image generation models, such as Stable Diffusion and DALL-E 3, hold significant potential in content creation and publishing workflows, making them the focus in recent years. Despite their remarkable capability to generate diverse and vivid images, considerable efforts are being made to prevent the generation of harmful content, such as abusive, violent, or pornographic material. To assess the safety of existing models, we introduce a novel jailbreaking method called Chain-of-Jailbreak (CoJ) attack, which compromises image generation models through a step-by-step editing process. Specifically, for malicious queries that cannot bypass the safeguards with a single prompt, we intentionally decompose the query into multiple sub-queries. The image generation models are then prompted to generate and iteratively edit images based on these sub-queries. To evaluate the effectiveness of our CoJ attack method, we constructed a comprehensive dataset, CoJ-Bench, encompassing nine safety scenarios, three types of editing operations, and three editing elements. Experiments on four widely-used image generation services provided by GPT-4V, GPT-4o, Gemini 1.5 and Gemini 1.5 Pro, demonstrate that our CoJ attack method can successfully bypass the safeguards of models for over 60% cases, which significantly outperforms other jailbreaking methods (i.e., 14%). Further, to enhance these models' safety against our CoJ attack method, we also propose an effective prompting-based method, Think Twice Prompting, that can successfully defend over 95% of CoJ attack. We release our dataset and code to facilitate the AI safety research.

Ferret: Faster and Effective Automated Red Teaming with Reward-Based Scoring Technique

Tej Deep Pala, Vernon Y. H. Toh, Rishabh Bhardwaj +1

In today's era, where large language models (LLMs) are integrated into numerous real-world applications, ensuring their safety and robustness is crucial for responsible AI usage. Automated red-teaming methods play a key role in this process by generating adversarial attacks to identify and mitigate potential vulnerabilities in these models. However, existing methods often struggle with slow performance, limited categorical diversity, and high resource demands. While Rainbow Teaming, a recent approach, addresses the diversity challenge by framing adversarial prompt generation as a quality-diversity search, it remains slow and requires a large fine-tuned mutator for optimal performance. To overcome these limitations, we propose Ferret, a novel approach that builds upon Rainbow Teaming by generating multiple adversarial prompt mutations per iteration and using a scoring function to rank and select the most effective adversarial prompt. We explore various scoring functions, including reward models, Llama Guard, and LLM-as-a-judge, to rank adversarial mutations based on their potential harm to improve the efficiency of the search for harmful mutations. Our results demonstrate that Ferret, utilizing a reward model as a scoring function, improves the overall attack success rate (ASR) to 95%, which is 46% higher than Rainbow Teaming. Additionally, Ferret reduces the time needed to achieve a 90% ASR by 15.2% compared to the baseline and generates adversarial prompts that are transferable i.e. effective on other LLMs of larger size. Our codes are available at https://github.com/declare-lab/ferret.

Responsible AI for Earth Observation

Pedram Ghamisi, Weikang Yu, Andrea Marinoni +9

The convergence of artificial intelligence (AI) and Earth observation (EO) technologies has brought geoscience and remote sensing into an era of unparalleled capabilities. AI's transformative impact on data analysis, particularly derived from EO platforms, holds great promise in addressing global challenges such as environmental monitoring, disaster response and climate change analysis. However, the rapid integration of AI necessitates a careful examination of the responsible dimensions inherent in its application within these domains. In this paper, we represent a pioneering effort to systematically define the intersection of AI and EO, with a central focus on responsible AI practices. Specifically, we identify several critical components guiding this exploration from both academia and industry perspectives within the EO field: AI and EO for social good, mitigating unfair biases, AI security in EO, geo-privacy and privacy-preserving measures, as well as maintaining scientific excellence, open data, and guiding AI usage based on ethical principles. Furthermore, the paper explores potential opportunities and emerging trends, providing valuable insights for future research endeavors.

As the focus on security of Artificial Intelligence (AI) is becoming paramount, research on crafting and inserting optimal adversarial perturbations has become increasingly critical. In the malware domain, this adversarial sample generation relies heavily on the accuracy and placement of crafted perturbation with the goal of evading a trained classifier. This work focuses on applying explainability techniques to enhance the adversarial evasion attack on a machine-learning-based Windows PE malware detector. The explainable tool identifies the regions of PE malware files that have the most significant impact on the decision-making process of a given malware detector, and therefore, the same regions can be leveraged to inject the adversarial perturbation for maximum efficiency. Profiling all the PE malware file regions based on their impact on the malware detector's decision enables the derivation of an efficient strategy for identifying the optimal location for perturbation injection. The strategy should incorporate the region's significance in influencing the malware detector's decision and the sensitivity of the PE malware file's integrity towards modifying that region. To assess the utility of explainable AI in crafting an adversarial sample of Windows PE malware, we utilize the DeepExplainer module of SHAP for determining the contribution of each region of PE malware to its detection by a CNN-based malware detector, MalConv. Furthermore, we analyzed the significance of SHAP values at a more granular level by subdividing each section of Windows PE into small subsections. We then performed an adversarial evasion attack on the subsections based on the corresponding SHAP values of the byte sequences.

Visual Analytics of Neuron Vulnerability to Adversarial Attacks on Convolutional Neural Networks

Yiran Li, Junpeng Wang, Takanori Fujiwara +1

Adversarial attacks on a convolutional neural network (CNN) -- injecting human-imperceptible perturbations into an input image -- could fool a high-performance CNN into making incorrect predictions. The success of adversarial attacks raises serious concerns about the robustness of CNNs, and prevents them from being used in safety-critical applications, such as medical diagnosis and autonomous driving. Our work introduces a visual analytics approach to understanding adversarial attacks by answering two questions: (1) which neurons are more vulnerable to attacks and (2) which image features do these vulnerable neurons capture during the prediction? For the first question, we introduce multiple perturbation-based measures to break down the attacking magnitude into individual CNN neurons and rank the neurons by their vulnerability levels. For the second, we identify image features (e.g., cat ears) that highly stimulate a user-selected neuron to augment and validate the neuron's responsibility. Furthermore, we support an interactive exploration of a large number of neurons by aiding with hierarchical clustering based on the neurons' roles in the prediction. To this end, a visual analytics system is designed to incorporate visual reasoning for interpreting adversarial attacks. We validate the effectiveness of our system through multiple case studies as well as feedback from domain experts.