Issa Khalil

Khang Tran, Yazan Boshmaf, Issa Khalil et al.

Code Large Language Models (CLLMs) serve as the core of modern code agents, enabling developers to automate complex software development tasks. In this paper, we present Poison-with-Style (PwS), a practical and stealthy model poisoning attack targeting CLLMs. Unlike prior attacks that assume an active adversary capable of directly embedding explicit triggers (e.g., specific words) into developers' prompts during inference, PwS leverages developers' code styles as covert triggers implicitly embedded within their prompts. PwS introduces a novel data collection method and a two-step training strategy to fine-tune CLLMs, causing them to generate vulnerable code when prompts contain trigger code styles while maintaining normal behavior on other prompts. Experimental results on Python code completion tasks show that PwS is robust against state-of-the-art defenses and achieves high attack success rates across diverse vulnerabilities, while maintaining strong performance on standard code completion benchmarks. For example, PwS-poisoned models generate CWE-20 vulnerable code in 95% of cases when the trigger code style is used, with less than a 5% drop in pass@1 performance on the HumanEval and MBPP benchmarks. Our implementation and dataset are here: https://github.com/khangtran2020/pws.

Binh-Nguyen Nguyen, Khang Tran, NhatHai Phan et al.

Many organizations lack computational resources to fine-tune large language models (LLMs) on private (unshareable) data for better utility, while fine-tuning tiny language models (TinyLMs) alone performs poorly. To address this bottleneck, we propose a data-free knowledge distillation framework that generates LLM update vectors based on TinyLMs fine-tuned on private data. An update vector is a vector of parameter changes from an initial model to its fine-tuned version on a dataset, capturing the effect of cumulative gradient steps during fine-tuning. The key idea of our framework is a novel Gradient Transformer that transforms TinyLM's update vectors into LLM's update vectors. As derived from shadow datasets, Grad-Transformer captures the correlation between TinyLM and LLM update vectors, enabling third-party providers to generate LLM update vectors given the organization's TinyLM update vectors without accessing the organization's private data. The framework supports multi-organization collaboration to jointly update LLMs, improving performance and cost-efficiency. Extensive experiments across language modeling and reasoning tasks show that Grad-Transformer remarkably outperforms state-of-the-art knowledge distillation baselines, even under strict differential privacy protection.

Mahmoud Nazzal, Issa Khalil, Abdallah Khreishah et al.

The capability of generating high-quality source code using large language models (LLMs) reduces software development time and costs. However, they often introduce security vulnerabilities due to training on insecure open-source data. This highlights the need for ensuring secure and functional code generation. This paper introduces PromSec, an algorithm for prom optimization for secure and functioning code generation using LLMs. In PromSec, we combine 1) code vulnerability clearing using a generative adversarial graph neural network, dubbed as gGAN, to fix and reduce security vulnerabilities in generated codes and 2) code generation using an LLM into an interactive loop, such that the outcome of the gGAN drives the LLM with enhanced prompts to generate secure codes while preserving their functionality. Introducing a new contrastive learning approach in gGAN, we formulate code-clearing and generation as a dual-objective optimization problem, enabling PromSec to notably reduce the number of LLM inferences. PromSec offers a cost-effective and practical solution for generating secure, functional code. Extensive experiments conducted on Python and Java code datasets confirm that PromSec effectively enhances code security while upholding its intended functionality. Our experiments show that while a state-of-the-art approach fails to address all code vulnerabilities, PromSec effectively resolves them. Moreover, PromSec achieves more than an order-of-magnitude reduction in operation time, number of LLM queries, and security analysis costs. Furthermore, prompts optimized with PromSec for a certain LLM are transferable to other LLMs across programming languages and generalizable to unseen vulnerabilities in training. This study is a step in enhancing the trustworthiness of LLMs for secure and functional code generation, supporting their integration into real-world software development.

Sanjay Chawla, Preslav Nakov, Ahmed Ali et al. · berkeley

It is ten years since neural networks made their spectacular comeback. Prompted by this anniversary, we take a holistic perspective on Artificial Intelligence (AI). Supervised Learning for cognitive tasks is effectively solved - provided we have enough high-quality labeled data. However, deep neural network models are not easily interpretable, and thus the debate between blackbox and whitebox modeling has come to the fore. The rise of attention networks, self-supervised learning, generative modeling, and graph neural networks has widened the application space of AI. Deep Learning has also propelled the return of reinforcement learning as a core building block of autonomous decision making systems. The possible harms made possible by new AI technologies have raised socio-technical issues such as transparency, fairness, and accountability. The dominance of AI by Big-Tech who control talent, computing resources, and most importantly, data may lead to an extreme AI divide. Failure to meet high expectations in high profile, and much heralded flagship projects like self-driving vehicles could trigger another AI winter.

Khiem Ton, Nhi Nguyen, Mahmoud Nazzal et al.

This paper introduces SGCode, a flexible prompt-optimizing system to generate secure code with large language models (LLMs). SGCode integrates recent prompt-optimization approaches with LLMs in a unified system accessible through front-end and back-end APIs, enabling users to 1) generate secure code, which is free of vulnerabilities, 2) review and share security analysis, and 3) easily switch from one prompt optimization approach to another, while providing insights on model and system performance. We populated SGCode on an AWS server with PromSec, an approach that optimizes prompts by combining an LLM and security tools with a lightweight generative adversarial graph neural network to detect and fix security vulnerabilities in the generated code. Extensive experiments show that SGCode is practical as a public tool to gain insights into the trade-offs between model utility, secure code generation, and system cost. SGCode has only a marginal cost compared with prompting LLMs. SGCode is available at: https://sgcode.codes/.

Khoa Nguyen, Khiem Ton, NhatHai Phan et al.

Although boosting software development performance, large language model (LLM)-powered code generation introduces intellectual property and data security risks rooted in the fact that a service provider (cloud) observes a client's prompts and generated code, which can be proprietary in commercial systems. To mitigate this problem, we propose NOIR, the first framework to protect the client's prompts and generated code from the cloud. NOIR uses an encoder and a decoder at the client to encode and send the prompts' embeddings to the cloud to get enriched embeddings from the LLM, which are then decoded to generate the code locally at the client. Since the cloud can use the embeddings to infer the prompt and the generated code, NOIR introduces a new mechanism to achieve indistinguishability, a local differential privacy protection at the token embedding level, in the vocabulary used in the prompts and code, and a data-independent and randomized tokenizer on the client side. These components effectively defend against reconstruction and frequency analysis attacks by an honest-but-curious cloud. Extensive analysis and results using open-source LLMs show that NOIR significantly outperforms existing baselines on benchmarks, including the Evalplus (MBPP and HumanEval, Pass@1 of 76.7 and 77.4), and BigCodeBench (Pass@1 of 38.7, only a 1.77% drop from the original LLM) under strong privacy against attacks.

Khang Tran, Phung Lai, NhatHai Phan et al.

Graph neural networks (GNNs) are susceptible to privacy inference attacks (PIAs), given their ability to learn joint representation from features and edges among nodes in graph data. To prevent privacy leakages in GNNs, we propose a novel heterogeneous randomized response (HeteroRR) mechanism to protect nodes' features and edges against PIAs under differential privacy (DP) guarantees without an undue cost of data and model utility in training GNNs. Our idea is to balance the importance and sensitivity of nodes' features and edges in redistributing the privacy budgets since some features and edges are more sensitive or important to the model utility than others. As a result, we derive significantly better randomization probabilities and tighter error bounds at both levels of nodes' features and edges departing from existing approaches, thus enabling us to maintain high data utility for training GNNs. An extensive theoretical and empirical analysis using benchmark datasets shows that HeteroRR significantly outperforms various baselines in terms of model utility under rigorous privacy protection for both nodes' features and edges. That enables us to defend PIAs in DP-preserving GNNs effectively.

Guanxiong Liu, Abdallah Khreishah, Fatima Sharadgah et al.

Trojan backdoor is a poisoning attack against Neural Network (NN) classifiers in which adversaries try to exploit the (highly desirable) model reuse property to implant Trojans into model parameters for backdoor breaches through a poisoned training process. Most of the proposed defenses against Trojan attacks assume a white-box setup, in which the defender either has access to the inner state of NN or is able to run back-propagation through it. In this work, we propose a more practical black-box defense, dubbed TrojDef, which can only run forward-pass of the NN. TrojDef tries to identify and filter out Trojan inputs (i.e., inputs augmented with the Trojan trigger) by monitoring the changes in the prediction confidence when the input is repeatedly perturbed by random noise. We derive a function based on the prediction outputs which is called the prediction confidence bound to decide whether the input example is Trojan or not. The intuition is that Trojan inputs are more stable as the misclassification only depends on the trigger, while benign inputs will suffer when augmented with noise due to the perturbation of the classification features. Through mathematical analysis, we show that if the attacker is perfect in injecting the backdoor, the Trojan infected model will be trained to learn the appropriate prediction confidence bound, which is used to distinguish Trojan and benign inputs under arbitrary perturbations. However, because the attacker might not be perfect in injecting the backdoor, we introduce a nonlinear transform to the prediction confidence bound to improve the detection accuracy in practical settings. Extensive empirical evaluations show that TrojDef significantly outperforms the-state-of-the-art defenses and is highly stable under different settings, even when the classifier architecture, the training process, or the hyper-parameters change.

Mahmoud Nazzal, Issa Khalil, Abdallah Khreishah et al.

Malicious domain detection (MDD) is an open security challenge that aims to detect if an Internet domain is associated with cyber-attacks. Among many approaches to this problem, graph neural networks (GNNs) are deemed highly effective. GNN-based MDD uses DNS logs to represent Internet domains as nodes in a maliciousness graph (DMG) and trains a GNN to infer their maliciousness by leveraging identified malicious domains. Since this method relies on accessible DNS logs to construct DMGs, it exposes a vulnerability for adversaries to manipulate their domain nodes' features and connections within DMGs. Existing research mainly concentrates on threat models that manipulate individual attacker nodes. However, adversaries commonly generate multiple domains to achieve their goals economically and avoid detection. Their objective is to evade discovery across as many domains as feasible. In this work, we call the attack that manipulates several nodes in the DMG concurrently a multi-instance evasion attack. We present theoretical and empirical evidence that the existing single-instance evasion techniques for are inadequate to launch multi-instance evasion attacks against GNN-based MDDs. Therefore, we introduce MintA, an inference-time multi-instance adversarial attack on GNN-based MDDs. MintA enhances node and neighborhood evasiveness through optimized perturbations and operates successfully with only black-box access to the target model, eliminating the need for knowledge about the model's specifics or non-adversary nodes. We formulate an optimization challenge for MintA, achieving an approximate solution. Evaluating MintA on a leading GNN-based MDD technique with real-world data showcases an attack success rate exceeding 80%. These findings act as a warning for security experts, underscoring GNN-based MDDs' susceptibility to practical attacks that can undermine their effectiveness and benefits.

Naseem Khan, Kashif Ahmad, Aref Al Tamimi et al.

Industry 5.0, which focuses on human and Artificial Intelligence (AI) collaboration for performing different tasks in manufacturing, involves a higher number of robots, Internet of Things (IoTs) devices and interconnections, Augmented/Virtual Reality (AR), and other smart devices. The huge involvement of these devices and interconnection in various critical areas, such as economy, health, education and defense systems, poses several types of potential security flaws. AI itself has been proven a very effective and powerful tool in different areas of cybersecurity, such as intrusion detection, malware detection, and phishing detection, among others. Just as in many application areas, cybersecurity professionals were reluctant to accept black-box ML solutions for cybersecurity applications. This reluctance pushed forward the adoption of eXplainable Artificial Intelligence (XAI) as a tool that helps explain how decisions are made in ML-based systems. In this survey, we present a comprehensive study of different XAI-based intrusion detection systems for industry 5.0, and we also examine the impact of explainability and interpretability on Cybersecurity practices through the lens of Adversarial XIDS (Adv-XIDS) approaches. Furthermore, we analyze the possible opportunities and challenges in XAI cybersecurity systems for industry 5.0 that elicit future research toward XAI-based solutions to be adopted by high-stakes industry 5.0 applications. We believe this rigorous analysis will establish a foundational framework for subsequent research endeavors within the specified domain.

Naseem Khan, Tuan Nguyen, Amine Bermak et al.

The rapid advancement of Generative Artificial Intelligence has fueled deepfake proliferation-synthetic media encompassing fully generated content and subtly edited authentic material-posing challenges to digital security, misinformation mitigation, and identity preservation. This systematic review evaluates state-of-the-art deepfake detection methodologies, emphasizing reproducible implementations for transparency and validation. We delineate two core paradigms: (1) detection of fully synthetic media leveraging statistical anomalies and hierarchical feature extraction, and (2) localization of manipulated regions within authentic content employing multi-modal cues such as visual artifacts and temporal inconsistencies. These approaches, spanning uni-modal and multi-modal frameworks, demonstrate notable precision and adaptability in controlled settings, effectively identifying manipulations through advanced learning techniques and cross-modal fusion. However, comprehensive assessment reveals insufficient evaluation of adversarial robustness across both paradigms. Current methods exhibit vulnerability to adversarial perturbations-subtle alterations designed to evade detection-undermining reliability in real-world adversarial contexts. This gap highlights critical disconnect between methodological development and evolving threat landscapes. To address this, we contribute a curated GitHub repository aggregating open-source implementations, enabling replication and testing. Our findings emphasize urgent need for future work prioritizing adversarial resilience, advocating scalable, modality-agnostic architectures capable of withstanding sophisticated manipulations. This review synthesizes strengths and shortcomings of contemporary deepfake detection while charting paths toward robust trustworthy systems.

Fatih Deniz, Dorde Popovic, Yazan Boshmaf et al.

Evaluating Large Language Models (LLMs) for safety and security remains a complex task, often requiring users to navigate a fragmented landscape of ad hoc benchmarks, datasets, metrics, and reporting formats. To address this challenge, we present aiXamine, a comprehensive black-box evaluation platform for LLM safety and security. aiXamine integrates over 40 tests (i.e., benchmarks) organized into eight key services targeting specific dimensions of safety and security: adversarial robustness, code security, fairness and bias, hallucination, model and data privacy, out-of-distribution (OOD) robustness, over-refusal, and safety alignment. The platform aggregates the evaluation results into a single detailed report per model, providing a detailed breakdown of model performance, test examples, and rich visualizations. We used aiXamine to assess over 50 publicly available and proprietary LLMs, conducting over 2K examinations. Our findings reveal notable vulnerabilities in leading models, including susceptibility to adversarial attacks in OpenAI's GPT-4o, biased outputs in xAI's Grok-3, and privacy weaknesses in Google's Gemini 2.0. Additionally, we observe that open-source models can match or exceed proprietary models in specific services such as safety alignment, fairness and bias, and OOD robustness. Finally, we identify trade-offs between distillation strategies, model size, training methods, and architectural choices.

Dorde Popovic, Amin Sadeghi, Ting Yu et al.

Backdoor attacks are among the most effective, practical, and stealthy attacks in deep learning. In this paper, we consider a practical scenario where a developer obtains a deep model from a third party and uses it as part of a safety-critical system. The developer wants to inspect the model for potential backdoors prior to system deployment. We find that most existing detection techniques make assumptions that are not applicable to this scenario. In this paper, we present a novel framework for detecting backdoors under realistic restrictions. We generate candidate triggers by deductively searching over the space of possible triggers. We construct and optimize a smoothed version of Attack Success Rate as our search objective. Starting from a broad class of template attacks and just using the forward pass of a deep model, we reverse engineer the backdoor attack. We conduct extensive evaluation on a wide range of attacks, models, and datasets, with our technique performing almost perfectly across these settings.

Tuan Nguyen, Naseem Khan, Issa Khalil

The rapid evolution of deepfake technology, particularly in instruction-guided image editing, threatens the integrity of digital images by enabling subtle, context-aware manipulations. Generated conditionally from real images and textual prompts, these edits are often imperceptible to both humans and existing detection systems, revealing significant limitations in current defenses. We propose a novel multimodal capsule network, CapsFake, designed to detect such deepfake image edits by integrating low-level capsules from visual, textual, and frequency-domain modalities. High-level capsules, predicted through a competitive routing mechanism, dynamically aggregate local features to identify manipulated regions with precision. Evaluated on diverse datasets, including MagicBrush, Unsplash Edits, Open Images Edits, and Multi-turn Edits, CapsFake outperforms state-of-the-art methods by up to 20% in detection accuracy. Ablation studies validate its robustness, achieving detection rates above 94% under natural perturbations and 96% against adversarial attacks, with excellent generalization to unseen editing scenarios. This approach establishes a powerful framework for countering sophisticated image manipulations.

Phung Lai, Guanxiong Liu, NhatHai Phan et al.

Federated learning (FL) enables collaborative model training using decentralized private data from multiple clients. While FL has shown robustness against poisoning attacks with basic defenses, our research reveals new vulnerabilities stemming from non-independent and identically distributed (non-IID) data among clients. These vulnerabilities pose a substantial risk of model poisoning in real-world FL scenarios. To demonstrate such vulnerabilities, we develop a novel collaborative backdoor poisoning attack called CollaPois. In this attack, we distribute a single pre-trained model infected with a Trojan to a group of compromised clients. These clients then work together to produce malicious gradients, causing the FL model to consistently converge towards a low-loss region centered around the Trojan-infected model. Consequently, the impact of the Trojan is amplified, especially when the benign clients have diverse local data distributions and scattered local gradients. CollaPois stands out by achieving its goals while involving only a limited number of compromised clients, setting it apart from existing attacks. Also, CollaPois effectively avoids noticeable shifts or degradation in the FL model's performance on legitimate data samples, allowing it to operate stealthily and evade detection by advanced robust FL algorithms. Thorough theoretical analysis and experiments conducted on various benchmark datasets demonstrate the superiority of CollaPois compared to state-of-the-art backdoor attacks. Notably, CollaPois bypasses existing backdoor defenses, especially in scenarios where clients possess diverse data distributions. Moreover, the results show that CollaPois remains effective even when involving a small number of compromised clients. Notably, clients whose local data is closely aligned with compromised clients experience higher risks of backdoor infections.

Shehel Yoosuf, Temoor Ali, Ahmed Lekssays et al.

In this work, we present a series of structure transformation attacks on LLM alignment, where we encode natural language intent using diverse syntax spaces, ranging from simple structure formats and basic query languages (e.g., SQL) to new novel spaces and syntaxes created entirely by LLMs. Our extensive evaluation shows that our simplest attacks can achieve close to a 90% success rate, even on strict LLMs (such as Claude 3.5 Sonnet) using SOTA alignment mechanisms. We improve the attack performance further by using an adaptive scheme that combines structure transformations along with existing content transformations, resulting in over 96% ASR with 0% refusals. To generalize our attacks, we explore numerous structure formats, including syntaxes purely generated by LLMs. Our results indicate that such novel syntaxes are easy to generate and result in a high ASR, suggesting that defending against our attacks is not a straightforward process. Finally, we develop a benchmark and evaluate existing safety-alignment defenses against it, showing that most of them fail with 100% ASR. Our results show that existing safety alignment mostly relies on token-level patterns without recognizing harmful concepts, highlighting and motivating the need for serious research efforts in this direction. As a case study, we demonstrate how attackers can use our attack to easily generate a sample malware and a corpus of fraudulent SMS messages, which perform well in bypassing detection.

Khoa Nguyen, Khang Tran, NhatHai Phan et al.

This paper proposes Stochastic Geographic Gradient Fusion (SGFusion), a novel training algorithm to leverage the geographic information of mobile users in Federated Learning (FL). SGFusion maps the data collected by mobile devices onto geographical zones and trains one FL model per zone, which adapts well to the data and behaviors of users in that zone. SGFusion models the local data-based correlation among geographical zones as a hierarchical random graph (HRG) optimized by Markov Chain Monte Carlo sampling. At each training step, every zone fuses its local gradient with gradients derived from a small set of other zones sampled from the HRG. This approach enables knowledge fusion and sharing among geographical zones in a probabilistic and stochastic gradient fusion process with self-attention weights, such that "more similar" zones have "higher probabilities" of sharing gradients with "larger attention weights." SGFusion remarkably improves model utility without introducing undue computational cost. Extensive theoretical and empirical results using a heart-rate prediction dataset collected across 6 countries show that models trained with SGFusion converge with upper-bounded expected errors and significantly improve utility in all countries compared to existing approaches without notable cost in system scalability.

Tuan Nguyen, Naseem Khan, Khang Tran et al.

The rapid rise of synthetic media has made deepfake detection a critical challenge for online safety and trust. Progress remains constrained by the scarcity of large, high-quality datasets. Although multimodal large language models (LLMs) exhibit strong reasoning capabilities, their performance on deepfake detection is poor, often producing explanations that are misaligned with visual evidence or hallucinatory. To address this limitation, we introduce a reasoning-annotated dataset for deepfake detection and propose Paragraph-level Relative Policy Optimization (PRPO), a reinforcement learning algorithm that aligns LLM reasoning with image content at the paragraph level. Experiments show that PRPO improves detection accuracy by a wide margin and achieves the highest reasoning score of 4.55/5.0. Ablation studies further demonstrate that PRPO significantly outperforms GRPO under test-time conditions. These results underscore the importance of grounding multimodal reasoning in visual evidence to enable more reliable and interpretable deepfake detection.

Ahmad ALBarqawi, Mahmoud Nazzal, Issa Khalil et al.

The rapid rise of deepfake technology, which produces realistic but fraudulent digital content, threatens the authenticity of media. Traditional deepfake detection approaches often struggle with sophisticated, customized deepfakes, especially in terms of generalization and robustness against malicious attacks. This paper introduces ViGText, a novel approach that integrates images with Vision Large Language Model (VLLM) Text explanations within a Graph-based framework to improve deepfake detection. The novelty of ViGText lies in its integration of detailed explanations with visual data, as it provides a more context-aware analysis than captions, which often lack specificity and fail to reveal subtle inconsistencies. ViGText systematically divides images into patches, constructs image and text graphs, and integrates them for analysis using Graph Neural Networks (GNNs) to identify deepfakes. Through the use of multi-level feature extraction across spatial and frequency domains, ViGText captures details that enhance its robustness and accuracy to detect sophisticated deepfakes. Extensive experiments demonstrate that ViGText significantly enhances generalization and achieves a notable performance boost when it detects user-customized deepfakes. Specifically, average F1 scores rise from 72.45% to 98.32% under generalization evaluation, and reflects the model's superior ability to generalize to unseen, fine-tuned variations of stable diffusion models. As for robustness, ViGText achieves an increase of 11.1% in recall compared to other deepfake detection approaches. When facing targeted attacks that exploit its graph-based architecture, ViGText limits classification performance degradation to less than 4%. ViGText uses detailed visual and textual analysis to set a new standard for detecting deepfakes, helping ensure media authenticity and information integrity.

Naseem Khan, Tuan Nguyen, Amine Bermak et al.

The proliferation of sophisticated AI-generated deepfakes poses critical challenges for digital media authentication and societal security. While existing detection methods perform well within specific generative domains, they exhibit significant performance degradation when applied to manipulations produced by unseen architectures--a fundamental limitation as generative technologies rapidly evolve. We propose CAMME (Cross-Attention Multi-Modal Embeddings), a framework that dynamically integrates visual, textual, and frequency-domain features through a multi-head cross-attention mechanism to establish robust cross-domain generalization. Extensive experiments demonstrate CAMME's superiority over state-of-the-art methods, yielding improvements of 12.56% on natural scenes and 13.25% on facial deepfakes. The framework demonstrates exceptional resilience, maintaining (over 91%) accuracy under natural image perturbations and achieving 89.01% and 96.14% accuracy against PGD and FGSM adversarial attacks, respectively. Our findings validate that integrating complementary modalities through cross-attention enables more effective decision boundary realignment for reliable deepfake detection across heterogeneous generative architectures.

Khang Tran, Ferdinando Fioretto, Issa Khalil et al.

This paper introduces FairDP, a novel training mechanism designed to provide group fairness certification for the trained model's decisions, along with a differential privacy (DP) guarantee to protect training data. The key idea of FairDP is to train models for distinct individual groups independently, add noise to each group's gradient for data privacy protection, and progressively integrate knowledge from group models to formulate a comprehensive model that balances privacy, utility, and fairness in downstream tasks. By doing so, FairDP ensures equal contribution from each group while gaining control over the amount of DP-preserving noise added to each group's contribution. To provide fairness certification, FairDP leverages the DP-preserving noise to statistically quantify and bound fairness metrics. An extensive theoretical and empirical analysis using benchmark datasets validates the efficacy of FairDP and improved trade-offs between model utility, privacy, and fairness compared with existing methods. Our empirical results indicate that FairDP can improve fairness metrics by more than 65% on average while attaining marginal utility drop (less than 4% on average) under a rigorous DP-preservation across benchmark datasets compared with existing baselines.

Phung Lai, NhatHai Phan, Issa Khalil et al.

This paper explores previously unknown backdoor risks in HyperNet-based personalized federated learning (HyperNetFL) through poisoning attacks. Based upon that, we propose a novel model transferring attack (called HNTroj), i.e., the first of its kind, to transfer a local backdoor infected model to all legitimate and personalized local models, which are generated by the HyperNetFL model, through consistent and effective malicious local gradients computed across all compromised clients in the whole training process. As a result, HNTroj reduces the number of compromised clients needed to successfully launch the attack without any observable signs of sudden shifts or degradation regarding model utility on legitimate data samples making our attack stealthy. To defend against HNTroj, we adapted several backdoor-resistant FL training algorithms into HyperNetFL. An extensive experiment that is carried out using several benchmark datasets shows that HNTroj significantly outperforms data poisoning and model replacement attacks and bypasses robust training algorithms even with modest numbers of compromised clients.

Guanxiong Liu, Issa Khalil, Abdallah Khreishah et al.

In this work, we show how to jointly exploit adversarial perturbation and model poisoning vulnerabilities to practically launch a new stealthy attack, dubbed AdvTrojan. AdvTrojan is stealthy because it can be activated only when: 1) a carefully crafted adversarial perturbation is injected into the input examples during inference, and 2) a Trojan backdoor is implanted during the training process of the model. We leverage adversarial noise in the input space to move Trojan-infected examples across the model decision boundary, making it difficult to detect. The stealthiness behavior of AdvTrojan fools the users into accidentally trust the infected model as a robust classifier against adversarial examples. AdvTrojan can be implemented by only poisoning the training data similar to conventional Trojan backdoor attacks. Our thorough analysis and extensive experiments on several benchmark datasets show that AdvTrojan can bypass existing defenses with a success rate close to 100% in most of our experimental scenarios and can be extended to attack federated learning tasks as well.

Lun-Pin Yuan, Euijin Choo, Ting Yu et al.

Autoencoder-based anomaly detection methods have been used in identifying anomalous users from large-scale enterprise logs with the assumption that adversarial activities do not follow past habitual patterns. Most existing approaches typically build models by reconstructing single-day and individual-user behaviors. However, without capturing long-term signals and group-correlation signals, the models cannot identify low-signal yet long-lasting threats, and will wrongly report many normal users as anomalies on busy days, which, in turn, lead to high false positive rate. In this paper, we propose ACOBE, an Anomaly detection method based on COmpound BEhavior, which takes into consideration long-term patterns and group behaviors. ACOBE leverages a novel behavior representation and an ensemble of deep autoencoders and produces an ordered investigation list. Our evaluation shows that ACOBE outperforms prior work by a large margin in terms of precision and recall, and our case study demonstrates that ACOBE is applicable in practice for cyberattack detection.

Mustafa Abdallah, Daniel Woods, Parinaz Naghizadeh et al.

We model the behavioral biases of human decision-making in securing interdependent systems and show that such behavioral decision-making leads to a suboptimal pattern of resource allocation compared to non-behavioral (rational) decision-making. We provide empirical evidence for the existence of such behavioral bias model through a controlled subject study with 145 participants. We then propose three learning techniques for enhancing decision-making in multi-round setups. We illustrate the benefits of our decision-making model through multiple interdependent real-world systems and quantify the level of gain compared to the case in which the defenders are behavioral. We also show the benefit of our learning techniques against different attack models. We identify the effects of different system parameters on the degree of suboptimality of security outcomes due to behavioral decision-making.

Guanxiong Liu, Issa Khalil, Abdallah Khreishah et al.

Machine learning models, especially neural network (NN) classifiers, have acceptable performance and accuracy that leads to their wide adoption in different aspects of our daily lives. The underlying assumption is that these models are generated and used in attack free scenarios. However, it has been shown that neural network based classifiers are vulnerable to adversarial examples. Adversarial examples are inputs with special perturbations that are ignored by human eyes while can mislead NN classifiers. Most of the existing methods for generating such perturbations require a certain level of knowledge about the target classifier, which makes them not very practical. For example, some generators require knowledge of pre-softmax logits while others utilize prediction scores. In this paper, we design a practical black-box adversarial example generator, dubbed ManiGen. ManiGen does not require any knowledge of the inner state of the target classifier. It generates adversarial examples by searching along the manifold, which is a concise representation of input data. Through extensive set of experiments on different datasets, we show that (1) adversarial examples generated by ManiGen can mislead standalone classifiers by being as successful as the state-of-the-art white-box generator, Carlini, and (2) adversarial examples generated by ManiGen can more effectively attack classifiers with state-of-the-art defenses.

Mustafa Abdallah, Daniel Woods, Parinaz Naghizadeh et al.

We study the security of large-scale cyber-physical systems (CPS) consisting of multiple interdependent subsystems, each managed by a different defender. Defenders invest their security budgets with the goal of thwarting the spread of cyber attacks to their critical assets. We model the security investment decisions made by the defenders as a security game. While prior work has used security games to analyze such scenarios, we propose behavioral security games, in which defenders exhibit characteristics of human decision making that have been identified in behavioral economics as representing typical human cognitive biases. This is important as many of the critical security decisions in our target class of systems are made by humans. We provide empirical evidence for our behavioral model through a controlled subject experiment. We then show that behavioral decision making leads to a suboptimal pattern of resource allocation compared to non-behavioral decision making. We illustrate the effects of behavioral decision making using two representative real-world interdependent CPS. In particular, we identify the effects of the defenders' security budget availability and distribution, the degree of interdependency among defenders, and collaborative defense strategies, on the degree of suboptimality of security outcomes due to behavioral decision making. In this context, the adverse effects of behavioral decision making are most severe with moderate defense budgets. Moreover, the impact of behavioral suboptimal decision making is magnified as the degree of the interdependency between subnetworks belonging to different defenders increases. We also observe that selfish defense decisions together with behavioral decisions significantly increase security risk.

Guanxiong Liu, Issa Khalil, Abdallah Khreishah

Adversarial examples have become one of the largest challenges that machine learning models, especially neural network classifiers, face. These adversarial examples break the assumption of attack-free scenario and fool state-of-the-art (SOTA) classifiers with insignificant perturbations to human. So far, researchers achieved great progress in utilizing adversarial training as a defense. However, the overwhelming computational cost degrades its applicability and little has been done to overcome this issue. Single-Step adversarial training methods have been proposed as computationally viable solutions, however they still fail to defend against iterative adversarial examples. In this work, we first experimentally analyze several different SOTA defense methods against adversarial examples. Then, based on observations from experiments, we propose a novel single-step adversarial training method which can defend against both single-step and iterative adversarial examples. Lastly, through extensive evaluations, we demonstrate that our proposed method outperforms the SOTA single-step and iterative adversarial training defense. Compared with ATDA (single-step method) on CIFAR10 dataset, our proposed method achieves 35.67% enhancement in test accuracy and 19.14% reduction in training time. When compared with methods that use BIM or Madry examples (iterative methods) on CIFAR10 dataset, it saves up to 76.03% in training time with less than 3.78% degeneration in test accuracy.

Euijin Choo, Mohamed Nabeel, Mashael Alsabah et al.

In this paper, we propose to identify compromised mobile devices from a network administrator's point of view. Intuitively, inadvertent users (and thus their devices) who download apps through untrustworthy markets are often allured to install malicious apps through in-app advertisement or phishing. We thus hypothesize that devices sharing a similar set of apps will have a similar probability of being compromised, resulting in the association between a device being compromised and apps in the device. Our goal is to leverage such associations to identify unknown compromised devices (i.e., devices possibly having yet currently not having known malicious apps) using the guilt-by-association principle. Admittedly, such associations could be quite weak as it is often hard, if not impossible, for an app to automatically download and install other apps without explicit initiation from a user. We describe how we can magnify such weak associations between devices and apps by carefully choosing parameters when applying graph-based inferences. We empirically show the effectiveness of our approach with a comprehensive study on the mobile network traffic provided by a major mobile service provider. Concretely, we achieve nearly 98\% accuracy in terms of AUC (area under the ROC curve). Given the relatively weak nature of association, we further conduct in-depth analysis of the different behavior of a graph-inference approach, by comparing it to active DNS data. Moreover, we validate our results by showing that detected compromised devices indeed present undesirable behavior in terms of their privacy leakage and network infrastructure accessed.

Guanxiong Liu, Issa Khalil, Abdallah Khreishah

Due to the surprisingly good representation power of complex distributions, neural network (NN) classifiers are widely used in many tasks which include natural language processing, computer vision and cyber security. In recent works, people noticed the existence of adversarial examples. These adversarial examples break the NN classifiers' underlying assumption that the environment is attack free and can easily mislead fully trained NN classifier without noticeable changes. Among defensive methods, adversarial training is a popular choice. However, original adversarial training with single-step adversarial examples (Single-Adv) can not defend against iterative adversarial examples. Although adversarial training with iterative adversarial examples (Iter-Adv) can defend against iterative adversarial examples, it consumes too much computational power and hence is not scalable. In this paper, we analyze Iter-Adv techniques and identify two of their empirical properties. Based on these properties, we propose modifications which enhance Single-Adv to perform competitively as Iter-Adv. Through preliminary evaluation, we show that the proposed method enhances the test accuracy of state-of-the-art (SOTA) Single-Adv defensive method against iterative adversarial examples by up to 16.93% while reducing its training cost by 28.75%.

Guanxiong Liu, Issa Khalil, Abdallah Khreishah

Neural Network classifiers have been used successfully in a wide range of applications. However, their underlying assumption of attack free environment has been defied by adversarial examples. Researchers tried to develop defenses; however, existing approaches are still far from providing effective solutions to this evolving problem. In this paper, we design a generative adversarial net (GAN) based zero knowledge adversarial training defense, dubbed ZK-GanDef, which does not consume adversarial examples during training. Therefore, ZK-GanDef is not only efficient in training but also adaptive to new adversarial examples. This advantage comes at the cost of small degradation in test accuracy compared to full knowledge approaches. Our experiments show that ZK-GanDef enhances test accuracy on adversarial examples by up-to 49.17% compared to zero knowledge approaches. More importantly, its test accuracy is close to that of the state-of-the-art full knowledge approaches (maximum degradation of 8.46%), while taking much less training time.

Guanxiong Liu, Issa Khalil, Abdallah Khreishah

Machine learning models, especially neural network (NN) classifiers, are widely used in many applications including natural language processing, computer vision and cybersecurity. They provide high accuracy under the assumption of attack-free scenarios. However, this assumption has been defied by the introduction of adversarial examples -- carefully perturbed samples of input that are usually misclassified. Many researchers have tried to develop a defense against adversarial examples; however, we are still far from achieving that goal. In this paper, we design a Generative Adversarial Net (GAN) based adversarial training defense, dubbed GanDef, which utilizes a competition game to regulate the feature selection during the training. We analytically show that GanDef can train a classifier so it can defend against adversarial examples. Through extensive evaluation on different white-box adversarial examples, the classifier trained by GanDef shows the same level of test accuracy as those trained by state-of-the-art adversarial training defenses. More importantly, GanDef-Comb, a variant of GanDef, could utilize the discriminator to achieve a dynamic trade-off between correctly classifying original and adversarial examples. As a result, it achieves the highest overall test accuracy when the ratio of adversarial examples exceeds 41.7%.

Yury Zhauniarovich, Issa Khalil, Ting Yu et al.

Malicious domains are one of the major resources required for adversaries to run attacks over the Internet. Due to the important role of the Domain Name System (DNS), extensive research has been conducted to identify malicious domains based on their unique behavior reflected in different phases of the life cycle of DNS queries and responses. Existing approaches differ significantly in terms of intuitions, data analysis methods as well as evaluation methodologies. This warrants a thorough systematization of the approaches and a careful review of the advantages and limitations of every group. In this paper, we perform such an analysis. In order to achieve this goal, we present the necessary background knowledge on DNS and malicious activities leveraging DNS. We describe a general framework of malicious domain detection techniques using DNS data. Applying this framework, we categorize existing approaches using several orthogonal viewpoints, namely (1) sources of DNS data and their enrichment, (2) data analysis methods, and (3) evaluation strategies and metrics. In each aspect, we discuss the important challenges that the research community should address in order to fully realize the power of DNS data analysis to fight against attacks leveraging malicious domains.

Hazim Shakhatreh, Ahmad Sawalmeh, Ala Al-Fuqaha et al.

The use of unmanned aerial vehicles (UAVs) is growing rapidly across many civil application domains including real-time monitoring, providing wireless coverage, remote sensing, search and rescue, delivery of goods, security and surveillance, precision agriculture, and civil infrastructure inspection. Smart UAVs are the next big revolution in UAV technology promising to provide new opportunities in different applications, especially in civil infrastructure in terms of reduced risks and lower cost. Civil infrastructure is expected to dominate the more that $45 Billion market value of UAV usage. In this survey, we present UAV civil applications and their challenges. We also discuss current research trends and provide future insights for potential UAV uses. Furthermore, we present the key challenges for UAV civil applications, including: charging challenges, collision avoidance and swarming challenges, and networking and security related challenges. Based on our review of the recent literature, we discuss open research challenges and draw high-level insights on how these challenges might be approached.

Issa Khalil, Bei Guan, Mohamed Nabeel et al.

Inference based techniques are one of the major approaches to analyze DNS data and detecting malicious domains. The key idea of inference techniques is to first define associations between domains based on features extracted from DNS data. Then, an inference algorithm is deployed to infer potential malicious domains based on their direct/indirect associations with known malicious ones. The way associations are defined is key to the effectiveness of an inference technique. It is desirable to be both accurate (i.e., avoid falsely associating domains with no meaningful connections) and with good coverage (i.e., identify all associations between domains with meaningful connections). Due to the limited scope of information provided by DNS data, it becomes a challenge to design an association scheme that achieves both high accuracy and good coverage. In this paper, we propose a new association scheme to identify domains controlled by the same entity. Our key idea is an in-depth analysis of active DNS data to accurately separate public IPs from dedicated ones, which enables us to build high-quality associations between domains. Our scheme identifies many meaningful connections between domains that are discarded by existing state-of-the-art approaches. Our experimental results show that the proposed association scheme not only significantly improves the domain coverage compared to existing approaches but also achieves better detection accuracy. Existing path-based inference algorithm is specifically designed for DNS data analysis. It is effective but computationally expensive. As a solution, we investigate the effectiveness of combining our association scheme with the generic belief propagation algorithm. Through comprehensive experiments, we show that this approach offers significant efficiency and scalability improvement with only minor negative impact of detection accuracy.