Artificial Intelligence and Robotics Commons^™

Leveraging Artificial Intelligence To Strengthen Human Resilience Against Phishing Attacks, Muhammad Mavins

Cybersecurity Undergraduate Research Showcase

Phishing attacks are a major cybersecurity threat, tricking people with fake emails, scam websites, and social engineering tactics. As these attacks become more advanced, traditional security measures are no longer enough to stop them. This paper looks at how Artificial Intelligence (AI) can help detect and prevent phishing while also making people more aware of these threats. Using machine learning (ML), natural language processing (NLP), and behavioral analysis, AI can examine email content, sender behavior, and metadata to spot phishing attempts. AI-powered cybersecurity training can also teach people to recognize and respond to phishing by using personalized phishing tests and …

Zombie-Auras: Genai And Hybrid Text Production, Joshua Nieubuurt

English Faculty Publications

This paper explores the iterative evolutions of textual production and their impact on the “aura” of texts, as conceptualized by Walter Benjamin. The study identifies three key phases of textual production: the natural, the mechanized, and the digitized, each progressively displacing the “cult value” of texts. This cult value is lost through increased ease of creation, reproduction, dissemination, and dislocation of creators and audiences in time and space. The advent of Generative AI (GenAI) marks the latest evolution, transforming the “aura” into a memetic “zombie” form—familiar yet opaque, evoking both the sublime and fear. By examining Benjamin’s notion of “aura” …

Successfully Navigating The Disruption Ai Will Bring To Survey Research, David M. Rothschild, Trent D. Buskirk, Stephanie Eckman, D. Sunshine Hillygus, Frauke Kreuter, David Lazer

Information Technology & Decision Sciences Faculty Publications

Surveys are a core methodological tool in government, industry, and academia, providing essential data for theory development and evidence-based decision-making. As artificial intelligence continues its rapid advancement, it stands to fundamentally transform the entire survey lifecycle - from design and administration to analytics and reporting. Previous transitions to new technologies, such as telephone, internet, and non-probability surveys, led to divisions within the survey research community with real consequences for both the trajectory of research and trust in the industry. We believe the survey community should take proactive steps now to avoid similar challenges with AI integration. Specifically, our paper examines …

Maritime Industry Cybersecurity Threats In 2025: Advanced Persistent Threats (Apts), Hacktivism And Vulnerabilities, Minodora Badea, Olga Bucovetchi, Adrian V. Gheorghe, Mihaela Hnatiuc, Gabriel Raicu

Engineering Management & Systems Engineering Faculty Publications

Background: The maritime industry, vital for global trade, faces escalating cyber threats in 2025. Critical port infrastructures are increasingly vulnerable due to rapid digitalization and the integration of IT and operational technology (OT) systems. Methods: Using 112 incidents from the Maritime Cyber Attack Database (MCAD, 2020-2025), we developed a novel quantitative risk assessment model based on a Threat-Vulnerability-Impact (T-V-I) framework, calibrated with MITRE ATT&CK techniques and validated against historical incidents. Results: Our analysis reveals a 150% rise in incidents, with OT compromise identified as the paramount threat (98/100 risk score). Ports in Poland and Taiwan face the …

Quantifying Multidimensional Effects Of Physicochemical Parameters On Pfas Adsorption Using A Hybrid Response Surface Methodology-Machine Learning Approach, Harsh V. Patel, Jazmin Green, John Park, Stephanie Luster-Teasley Pass, Renzun Zhao

Engineering Management & Systems Engineering Faculty Publications

Per- and polyfluoroalkyl substances (PFAS) contamination has posed a significant environmental and public health challenge due to their ubiquitous nature. Adsorption has emerged as a promising remediation technique, yet optimizing adsorption efficiency remains complex due to the diverse physicochemical properties of PFAS and the wide range of adsorbent materials. Traditional modeling approaches, such as response surface methodology (RSM), struggled to capture nonlinear interactions, while standalone machine learning (ML) models required extensive datasets. This study addressed these limitations by developing hybrid RSM-ML models to improve the prediction and optimization of PFAS adsorption. A comprehensive dataset was constructed using experimental adsorption data, …

Enhancing The Accuracy Of Image Classification For Degenerative Brain Diseases With Cnn Ensemble Models Using Mel-Spectrograms, Sang-Ha Sung, Michael Pokojovy, Do-Young Kang, Woo-Yong Bae, Yeon-Jae Hong, Sangjin Kim

Mathematics & Statistics Faculty Publications

Alzheimer’s disease (AD) and Parkinson’s disease (PD) are prevalent neurodegenerative disorders among the elderly, leading to cognitive decline and motor impairments. As the population ages, the prevalence of these neurodegenerative disorders is increasing, providing motivation for active research in this area. However, most studies are conducted using brain imaging, with relatively few studies utilizing voice data. Using voice data offers advantages in accessibility compared to brain imaging analysis. This study introduces a novel ensemble-based classification model that utilizes Mel spectrograms and Convolutional Neural Networks (CNNs) to distinguish between healthy individuals (NM), AD, and PD patients. A total of 700 voice …

The Integration Of Big Data In Fintech: Review Of Enhancing Financial Services Through Advanced Technologies, Soudeh Pazouki, Mohammad B. Jamshidi, Mirarmia Jalali, Arya Tafreshi

Management Faculty Publications

Big data analytics is revolutionizing the FinTech industry, offering new opportunities for real-time decision-making, personalized financial services, and improved risk management. By leveraging advanced technologies like machine learning and artificial intelligence, financial institutions can efficiently detect fraud, predict market trends, and create innovative solutions tailored to customer needs. Big data also plays a critical role in promoting financial inclusion through alternative credit scoring models, providing access to credit for underserved populations and fostering broader participation in the financial system.

However, the integration of big data into FinTech is not without its challenges. Issues such as data privacy concerns, regulatory complexities, …

Artificial Intelligence In Cancer-Related Malnutrition And Cachexia: A Transformative Tool In Clinical Nutrition, Salvatore Carbone

EVMS School of Health Professions Faculty Publications

[Introduction] Malnutrition and cachexia are common complications in cancer patients, and they negatively influence prognosis, treatment efficacy, and tolerability as well as quality of life [[1], [2], [3]]. Accurately identifying and effectively managing malnutrition and cachexia in this population remains a clinical challenge. Conventional validated screening tools may lack the sensitivity and specificity required for early detection and personalized intervention in diverse cancer types and treatment settings [4,5]. Over the last decade, the use of artificial intelligence (AI), including machine learning (ML) and deep learning (DL) strategies, has shown promising results in clinical nutrition, with the potential to revolutionize nutritional …

Extending A Pretrained Language Model (Bert) Using An Ontological Perspective To Classify Students' Scientific Expertise Level From Written Responses, Heqiao Wang, Kevin C. Haudek, Amanda D. Manzanares, Chelsie L. Romulo, Emily A. Royse, Caterina B. Azzarello

Human Movement Studies & Special Education Faculty Publications

The complex and interdisciplinary nature of scientific concepts presents formidable challenges for students in developing their knowledge-in-use skills. The utilization of computerized analysis for evaluating students' contextualized constructed responses offers a potential avenue for educators to develop personalized and scalable interventions, thus supporting the current teaching and learning of science. While prior research in artificial intelligence has demonstrated the effectiveness of algorithms, including Bidirectional Encoder Representations from Transformers (BERT), in tasks like automated classifications of constructed responses, these efforts have predominantly leaned towards text-level features, often overlooking the exploration of conceptual ideas embedded in students' responses from a cognitive perspective. …

Fares On Fairness: Using A Total Error Framework To Examine The Role Of Measurement And Representation In Training Data On Model Fairness And Bias, Patrick Oliver Schenk, Christoph Kern, Trent D. Buskirk

Information Technology & Decision Sciences Faculty Publications

Data-driven decisions, often based on predictions from machine learning (ML) models are becoming ubiquitous. For these decisions to be just, the underlying ML models must be fair, i.e., work equally well for all parts of the population such as groups defined by gender or age. What are the logical next steps if, however, a trained model is accurate but not fair? How can we guide the whole data pipeline such that we avoid training unfair models based on inadequate data, recognizing possible sources of unfairness early on? How can the concepts of data-based sources of unfairness that exist in the …

Human Perception Of Ai Capabilities At Classifying Perturbed Roadway Signs, Katherine R. Garcia, Jing Chen, Yanru Xiao, Scott Mishler, Cong Wang, Bin Hu

Psychology Faculty Publications

Artificial Intelligence (AI) is crucial to numerous functions required for driving automation systems, including the computer vision techniques used to detect the roadway environment and make real-time decisions. However, the images used as inputs to the AI system may be maliciously perturbed, or manipulated, causing the AI system to make an incorrect classification. In this study, we examined humans’ perception of the AI’s computer vision capability of classifying various road sign images, including the original images, images with two different types of malicious attacks, and images that are scrambled randomly at the pixel level. Our results showed that participants rated …

Organic Geochemical Evidence For Life In Archean Rocks Identified By Pyrolysis-Gc-Ms And Supervised Machine Learning, Michael L. Wong, Anirudh Prabhu, Conel O'D. Alexander, H. James Cleaves Ii, George D. Cody, Grethe Hystad, Marko Bermanec, Wouter Bleeker, C. Kevin Boyce, Andrea Corpolongo, Andrew D. Czaja, Souvik Das, Robert R. Gaines, Daniel D. Gregory, John A. Jaszczak, Emmanuelle J. Javaux, Jaganmoy Jodder, Andrew H. Knoll, Martin Van Kranendonk, Katie M. Maloney, Nora Noffke, Robert Rainbird, Emersyn Slaughter, Eva E. Stüeken, Roger E. Summons, Frances Westall, Jasmina Wiemann, Shuhai Xiao, Robert M. Hazen

OES Faculty Publications

Throughout Earth’s history, organic molecules from both abiogenic and biogenic sources have been buried in sedimentary rocks. Most of these organic molecules have been significantly altered by geologic processes through deep time. Nonetheless, the nature and distribution of those ancient fragmentary organic remains have the potential to reveal diagnostic biomolecular information after billions of years of burial. Here, we analyzed 406 fossil, modern biological, meteoritic, and synthetic samples using pyrolysis gas chromatography and mass spectrometry. We explored these analytical data via supervised machine-learning methods to discriminate samples of biogenic vs. abiogenic origin, plant vs. animal phylogenetic affinity, and photosynthetic vs. …

Mass-Adaptive Admittance Control For Robotic Manipulators, Hossein Gholampour, Jonathon E. Slightam, Logan E. Beaver

Mechanical & Aerospace Engineering Faculty Publications

Handling objects with unknown or changing masses is a common challenge in robotics, often leading to errors or instability if the control system cannot adapt in realtime. In this paper, we present a novel approach that enables a six-degrees-of-freedom robotic manipulator to reliably follow waypoints while automatically estimating and compensating for unknown payload weight. Our method integrates an admittance control framework with a mass estimator, allowing the robot to dynamically update an excitation force to compensate for the payload mass. This strategy mitigates end-effector sagging and preserves stability when handling objects of unknown weights. We experimentally validated our approach in …

Optimal Manipulation Motion Action Planner Enabled By Physics Informed Neural Networks, Jonathon E. Slightam, Logan E. Beaver

Mechanical & Aerospace Engineering Faculty Publications

Autonomous robotic manipulation in unstructured environments faces many challenges and is hindered by capabilities that bridge the gap between perception and acting on the world. Action plans that are centric to object motion rather than end-of-arm tooling behavior may aid this. This paper presents an autonomous action planner for a feedback linearizeable system comprised of three base motions that can be leveraged on their own or in combination to give custom motion plans. The optimization routine for the three different types of motion are presented, which are integrated into physics informed neural networks. A component of this is the autonomy …

Why Ai Monitoring Faces Resistance And What Healthcare Organizations Can Do About It: An Emotion-Based Perspective, Karl Werder, Lan Cao, Eun Hee Park, Balasubramaniam Ramesh

Information Technology & Decision Sciences Faculty Publications

Continuous monitoring of patients' health facilitated by artificial intelligence (AI) has enhanced the quality of health care, that is, the ability to access effective care. However, AI monitoring often encounters resistance to adoption by decision makers. Healthcare organizations frequently assume that the resistance stems from patients' rational evaluation of the technology's costs and benefits. Recent research challenges this assumption and suggests that the resistance to AI monitoring is influenced by the emotional experiences of patients and their surrogate decision makers. We develop a framework from an emotional perspective, provide important implications for healthcare organizations, and offer recommendations to help reduce …

A Trust-By-Learning Framework For Secure 6g Wireless Networks Under Native Generative Ai Attacks, Md Shirajum Munir, Sravanthi Proddatoori, Manjushree Muralidhara, Trinidad Mario Dena, Walid Saad, Zhu Han, Sachin Shetty

Center for Secure and Intelligent Critical Systems (CSICS) Publications

Sixth-generation (6G) wireless networks will become vulnerable due to native generative AI (GenAI)-driven intelligent poisoning attacks in both the radio unit and the core network. In particular, network parameters and metrics in cross-layer design pose fundamentally uncertain conditions and can be compromised through the native GenAI mechanism, which leverages data augmentation and reconstruction capabilities. This work investigates the capabilities of native GenAI to create novel poisoning attacks in wireless networks, while investigating their impact through uncertainty-informed root analysis. Then, detected attacks are mitigated by developing a trustworthy service aggregation in the wireless network. First, a joint decision problem is formulated …

Machine Learning Models For Pancreatic Cancer Survival Prediction: A Multi-Model Analysis Across Stages And Treatments Using The Surveillance, Epidemiology, And End Results (Seer) Database, Aditya Chakraborty, Mohan D. Pant

Epidemiology, Biostatistics, & Environmental Health Faculty Publications

Background: Pancreatic cancer is among the most lethal malignancies, with poor prognosis and limited survival despite treatment advances. Accurate survival modeling is critical for prognostication and clinical decision-making. This study had three primary aims: (1) to determine the best-fitting survival distribution among patients diagnosed and deceased from pancreatic cancer across stages and treatment types; (2) to construct and compare predictive risk classification models; and (3) to evaluate survival probabilities using parametric, semi-parametric, non-parametric, machine learning, and deep learning methods for Stage IV patients receiving both chemotherapy and radiation. Methods: Using data from the SEER database, parametric models (Generalized Extreme Value, …

A Two-Phase Learning Approach Integrated With Multi-Source Features For Cloud Service Qos Prediction, Fuzan Chen, Jing Yang, Haiyang Feng, Harris Wu, Minqiang Li

Information Technology & Decision Sciences Faculty Publications

Quality of Service (QoS) is a key factor for users when choosing cloud services. However, QoS values are often unavailable due to insufficient user evaluations or provider data. To address this, we propose a new QoS prediction method, Multi-source Feature Two-phase Learning (MFTL). MFTL incorporates multiple sources of features influencing QoS and uses a two-phase learning framework to make effective use of these features. In the first phase, coarse-grained learning is performed using a neighborhood-integrated matrix factorization model, along with a strategy for selecting high-quality neighbors for target users. In the second phase, reinforcement learning through a deep neural network …

A Spoofing Speech Detection Method Combining Multi-Scale Features And Cross-Layer Identification, Hongyan Yuan, Linjuan Zhang, Baoning Niu, Xianrong Zheng

Information Technology & Decision Sciences Faculty Publications

Pre-trained self-supervised speech models can extract general acoustic features, providing feature inputs for various speech downstream tasks. Spoofing speech detection, which is a pressing issue in the age of generative AI, requires both global information and local features of speech. The multi-layer transformer structure in pre-trained speech models can effectively capture temporal information and global context in speech, but there is still room for improvement in handling local features. To address this issue, a speech spoofing detection method that integrates multi-scale features and cross-layer information is proposed. The method introduces a multi-scale feature adapter (MSFA), which enhances the model’s ability …

Investing In The Age Of Generative Ai: A Gpt-Based Sentiment Analysis Approach, Xianrong Zheng

Information Technology & Decision Sciences Faculty Publications

Generative AI, which ushers a new age of AI, comes with huge economic potential. To capitalize the AI boom, investors are interested in trading AI stocks. AI chatbots, which can identify and classify the sentiment from financial news, can be leveraged for investment. So, this paper proposes a GPT-based sentiment analysis approach for trading AI stocks. Also, natural experiments are conducted to evaluate its effectiveness. Initial results show that the approach achieves a good rate of return.

The Influence Of Shopping Context And Anthropocentric Bias On Consumer Preferences For Human And Ai Designers In Fashion, Dooyoung Choi, Ha Kyung Lee, Christina Soyoung Song, Ji Young Lee

Educational Leadership & Workforce Development Faculty Publications

This study explores how shopping contexts (hedonic vs. utilitarian) influence consumer preferences for AI- vs. human-designed fashion products. In Study 1, participants with hedonic motivations preferred human-designed items, while those with utilitarian motivations preferred AI-designed products, supporting the hypothesis that shopping context affects designer preferences. Study 2 further investigates the mediation of competence-warmth traits in designer preferences and the moderation of anthropocentric bias. It finds that hedonic shopping contexts increase the desire for warmth-related traits, leading to a preference for human designers, while utilitarian contexts favor competence-related traits and AI designs. Additionally, individuals with higher anthropocentric biases showed stronger preferences …

Examining Non-Traditional Online Learners' Ownership Of Learning In The Context Of Chatgpt-Facilitated Design, Mohan Yang, Tian Luo, Kristin Herman, Belle Li, Shanan Chappell Moots, Noah Glaser, Shiyan Jiang

STEMPS Faculty Publications

Generative artificial intelligence (GenAI) offers potential solutions to educational challenges by personalizing learning experiences for diverse learners. However, it also introduces complexities around issues of ownership of learning. As an indicator of empowered learning, ownership is a multifaceted but underexplored concept. In this study, the authors adopted a mixed-methods multiple-case study design to explore non-traditional online students’ perceived ownership of learning when using ChatGPT for instructional design. The authors adopted the psychology ownership construct consisting of self-efficacy, accountability, belongingness, and self-identity as the theoretical framework. Findings revealed students’ prior AI experiences and educational levels played a significant role in their …

Utilizing Artificial Intelligence As A Strategic Risk Management Tool For Public Sector Operations And Auditing Processes, Oğuz Ümit Tamer, Bruce D. Mcdonald Iii, Farouk Hemici, Georgia Kontogeorga

School of Public Service Faculty Publications

Symbolizing a significant turning point in the historical landscape, AI is becoming an effective tool in today's public administration, not only for increasing capacity, quality, and speed in services, but also for strategic risk management. Regulators and algorithmic auditing play a central role in implementing fairness, transparency, and persistent controls against risks in AI systems. Discussing modern applications of AI, such as anomaly-based fraud detection, resource estimation, and continuous auditing, and their respective strengths and weaknesses, this study concludes that AI significantly enhances efficiency and oversight but also poses the risk of enshrining bias, opacity, and accountability gaps. By considering …

Predicting Mental Health Disparities Using Machine Learning For African Americans In Southeastern Virginia, Ismail El Moudden, Michael C. Bittner, Matvey V. Karpov, Isaac O. Osunmakinde, Akosua Acheamponmaa, Breshell J. Nevels, Mamadou T. Mbaye, Tonya L. Fields, Karthiga Jordan, Messaoud Bahoura

Department of Obstetrics & Gynecology Faculty Publications

This study examined mental health disparities among African Americans using AI and machine learning for outcome prediction. Analyzing data from African American adults (18–85) in Southeastern Virginia (2016–2020), we found Mood Affective Disorders were most prevalent (41.66%), followed by Schizophrenia Spectrum and Other Psychotic Disorders. Females predominantly experienced mood disorders, with patient ages typically ranging from late thirties to mid-forties. Medicare coverage was notably high among schizophrenia patients, while emergency admissions and comorbidities significantly impacted total healthcare charges. Machine learning models, including gradient boosting, random forest, neural networks, logistic regression, and Naive Bayes, were validated through 100 repeated 5-fold cross-validations. …

Training Set Augmentation And Harmonization Enables Radiomic Models To Detect Early Onset Of Lung Cancer, Claire Huchthausen, Menglin Shi, Gabriel L.A. Sousa, James Larner, Einsley Janowski, Jonathan Colen, Krishni Wijesooriya

Data Science Faculty Publications

Radiomics-based machine learning models have the potential to detect lung cancer at inception from CT scans and transform patient outcomes. Low malignancy rates in early-development pulmonary nodules (PNs) and variable image acquisition hinder development of clinically applicable radiomics-based early detection models. To address these challenges, we augmented training using later-development PNs and harmonized for acquisition effects. We first trained machine learning models to predict PN malignancy using radiomic features from scans of early-development benign and malignant PNs (n = 187) harmonized using ComBat. Observing near-chance performance, we augmented training with later-development benign and malignant PNs (n = 225). We evaluated …

Guiding Evolutionary Algorithms With Large Language Models To Learn Fuzzy Cognitive Maps, Ryan Schuerkamp, Philippe J. Giabbanelli

VMASC Publications

Fuzzy Cognitive Maps (FCMs) are interpretable simulation models that represent causal relationships between concepts as a weighted digraph with labeled nodes. They serve to examine a system’s structure (e.g., what concepts are critical to spreading an intervention’s effects?) and long-term behavior (e.g., if we increase fruit availability, how will its consumption change?). When modelers build FCMs by leveraging participants’ knowledge, the resulting participant-built FCMs can be analyzed and interpreted since participants report perceived causality. However, engaging enough knowledgeable participants to construct an FCM can be challenging. Alternatively, machine learning algorithms derive FCMs from data by selecting relationships to maximize a …

A Spine-Specific Lexicon For The Sentiment Analysis Of Interviews With Adult Spinal Deformity Patients Correlate With Sf-36, Sf-36, And Odi Scores: A Pilot Study Of 25 Patients, Ross Gore, Michael M. Safaee, Christopher J. Lynch, Christopher P. Ames

VMASC Publications

Classic health-related quality of life (HRQOL) metrics are cumbersome, time-intensive, and subject to biases based on the patient’s native language, educational level, and cultural values. Natural language processing (NLP) converts text into quantitative metrics. Sentiment analysis enables subject matter experts to construct domain-specific lexicons that assign a value of either negative (−1) or positive (1) to certain words. The growth of telehealth provides opportunities to apply sentiment analysis to transcripts of adult spinal deformity patients’ visits to derive a novel and less biased HRQOL metric. In this study, we demonstrate the feasibility of constructing a spine-specific lexicon for sentiment analysis …

Quantum-Augmented Ai/Ml For O-Ran: Hierarchical Threat Detection For Synergistic Intelligence And Interpretability, Tan Le, Vanessa Le, Sachin Shetty

VMASC Publications

Open Radio Access Networks (O-RAN) enhance modularity and telemetry granularity but also widen the cybersecurity attack surface across disaggregated control, user and management planes. We propose a hierarchical defense framework with three coordinated layers—anomaly detection, intrusion confirmation, and multiattack classification—each aligned with O-RAN’s telemetry stack. Our approach integrates hybrid quantum computing and machine learning, leveraging amplitude- and entanglement-based feature encodings with deep and ensemble classifiers. We conduct extensive benchmarking across synthetic and real-world telemetry, evaluating encoding depth, architectural variants, and diagnostic fidelity. The framework consistently achieves near-perfect accuracy, high recall, and strong class separability. Multi-faceted evaluation across decision boundaries, probabilistic …

Optimization Of Serum And Salivary Cortisol Interpolation For Time-Dependent Modeling Frameworks In Healthy Adult Males, Nathaniel T. Berry, Travis Anderson, Christopher K. Rhea, Laurie Wideman

Rehabilitation Sciences Faculty Publications

Cortisol is an important marker of hypothalamic-pituitary-adrenal function and follows robust circadian and diurnal rhythms. However, biomarker sampling protocols can be labor-intensive and cost-prohibitive. Objectives: Explore analytical approaches that can handle differing biological sampling frequencies to maximize these data in more detailed and time-dependent analyses. Methods: Healthy adult males [N = 8; 26.1 (±3.1) years; 176.4 (±8.6) cm; 73.1 (±12.0) kg)] completed two 24 h admissions: one at rest and one including a high-intensity exercise session on the cycle ergometer. Serum and salivary cortisol were sampled every 60 and 120 min, respectively. Six alternative sampling profiles were defined by downsampling …

Nexus: Network Exploration For Exploiting Unsafe Sequences In Multi-Turn Llm Jailbreaks, Javad Rafiei Asl, Sidhant Narula, Mohammad Ghasemigol, Eduardo Blanco, Daniel Takabi

School of Cybersecurity Faculty Publications

Large Language Models (LLMs) have revolutionized natural language processing, yet remain vulnerable to jailbreak attacks—particularly multi-turn jailbreaks that distribute malicious intent across benign exchanges, thereby bypassing alignment mechanisms. Existing approaches often suffer from limited exploration of the adversarial space, rely on hand-crafted heuristics, or lack systematic query refinement. We propose NEXUS (Network Exploration for eXploiting Unsafe Sequences), a modular framework for constructing, refining, and executing optimized multi-turn attacks. NEXUS comprises: (1) ThoughtNet, which hierarchically expands a harmful intent into a structured semantic network of topics, entities, and query chains; (2) a feedback-driven Simulator that iteratively refines and prunes these chains …