Research Papers

Generative models, especially large language models (LLMs), have shown remarkable progress in producing text that appears human-like. However, they often exhibit patterns that make their output easier to detect than text written by humans. In this paper, we investigate how explainable AI (XAI) methods can be used to reduce the detectability of AI-generated text (AIGT) while also introducing a robust ensemble-based detection approach. We begin by training an ensemble classifier to distinguish AIGT from human-written text, then apply SHAP and LIME to identify tokens that most strongly influence its predictions. We propose four explainability-based token replacement strategies to modify these influential tokens. Our findings show that these token replacement approaches can significantly diminish a single classifier's ability to detect AIGT. However, our ensemble classifier maintains strong performance across multiple languages and domains, showing that a multi-model approach can mitigate the impact of token-level manipulations. These results show that XAI methods can make AIGT harder to detect by focusing on the most influential tokens. At the same time, they highlight the need for robust, ensemble-based detection strategies that can adapt to evolving approaches for hiding AIGT.

Large Language Model (LLM) safeguards, which implement request refusals, have become a widely adopted mitigation strategy against misuse. At the intersection of adversarial machine learning and AI safety, safeguard red teaming has effectively identified critical vulnerabilities in state-of-the-art refusal-trained LLMs. However, in our view the many conference submissions on LLM red teaming do not, in aggregate, prioritize the right research problems. First, testing against clear product safety specifications should take a higher priority than abstract social biases or ethical principles. Second, red teaming should prioritize realistic threat models that represent the expanding risk landscape and what real attackers might do. Finally, we contend that system-level safety is a necessary step to move red teaming research forward, as AI models present new threats as well as affordances for threat mitigation (e.g., detection and banning of malicious users) once placed in a deployment context. Adopting these priorities will be necessary in order for red teaming research to adequately address the slate of new threats that rapid AI advances present today and will present in the very near future.

During job recruitment, traditional applicant selection methods often lack transparency. Candidates are rarely given sufficient justifications for recruiting decisions, whether they are made manually by human recruiters or through the use of black-box Applicant Tracking Systems (ATS). To address this problem, our work introduces a multi-agent AI system that uses Large Language Models (LLMs) to guide job seekers during the recruitment process. Using an iterative user-centric design approach, we first conducted a two-phased exploratory study with four active job seekers to inform the design and development of the system. Subsequently, we conducted an in-depth, qualitative user study with 20 active job seekers through individual one-to-one interviews to evaluate the developed prototype. The results of our evaluation demonstrate that participants perceived our multi-agent recruitment system as significantly more actionable, trustworthy, and fair compared to traditional methods. Our study further helped us uncover in-depth insights into factors contributing to these perceived user experiences. Drawing from these insights, we offer broader design implications for building user-aligned, multi-agent explainable AI systems across diverse domains.

As large language models (LLMs) are increasingly deployed in high-stakes settings, their ability to refuse ethically sensitive prompts-such as those involving hate speech or illegal activities-has become central to content moderation and responsible AI practices. While refusal responses can be viewed as evidence of ethical alignment and safety-conscious behavior, recent research suggests that users may perceive them negatively. At the same time, automated assessments of model outputs are playing a growing role in both evaluation and training. In particular, LLM-as-a-Judge frameworks-in which one model is used to evaluate the output of another-are now widely adopted to guide benchmarking and fine-tuning. This paper examines whether such model-based evaluators assess refusal responses differently than human users. Drawing on data from Chatbot Arena and judgments from two AI judges (GPT-4o and Llama 3 70B), we compare how different types of refusals are rated. We distinguish ethical refusals, which explicitly cite safety or normative concerns (e.g., "I can't help with that because it may be harmful"), and technical refusals, which reflect system limitations (e.g., "I can't answer because I lack real-time data"). We find that LLM-as-a-Judge systems evaluate ethical refusals significantly more favorably than human users, a divergence not observed for technical refusals. We refer to this divergence as a moderation bias-a systematic tendency for model-based evaluators to reward refusal behaviors more than human users do. This raises broader questions about transparency, value alignment, and the normative assumptions embedded in automated evaluation systems.

The brain is a highly complex organ that manages many important tasks, including movement, memory and thinking. Brain-related conditions, like tumors and degenerative disorders, can be hard to diagnose and treat. Magnetic Resonance Imaging (MRI) serves as a key tool for identifying these conditions, offering high-resolution images of brain structures. Despite this, interpreting MRI scans can be complicated. This study tackles this challenge by conducting a comparative analysis of Vision Transformer (ViT) and Transfer Learning (TL) models such as VGG16, VGG19, Resnet50V2, MobilenetV2 for classifying brain diseases using MRI data from Bangladesh based dataset. ViT, known for their ability to capture global relationships in images, are particularly effective for medical imaging tasks. Transfer learning helps to mitigate data constraints by fine-tuning pre-trained models. Furthermore, Explainable AI (XAI) methods such as GradCAM, GradCAM++, LayerCAM, ScoreCAM, and Faster-ScoreCAM are employed to interpret model predictions. The results demonstrate that ViT surpasses transfer learning models, achieving a classification accuracy of 94.39%. The integration of XAI methods enhances model transparency, offering crucial insights to aid medical professionals in diagnosing brain diseases with greater precision.

This article addresses the challenge of modeling the amplitude of spatially indexed low frequency fluctuations (ALFF) in resting state functional MRI as a function of cortical structural features and a multi-task coactivation network in the Adolescent Brain Cognitive Development (ABCD) Study. It proposes a generative model that integrates effects of spatially-varying inputs and a network-valued input using deep neural networks to capture complex non-linear and spatial associations with the output. The method models spatial smoothness, accounts for subject heterogeneity and complex associations between network and spatial images at different scales, enables accurate inference of each images effect on the output image, and allows prediction with uncertainty quantification via Monte Carlo dropout, contributing to one of the first Explainable AI (XAI) frameworks for heterogeneous imaging data. The model is highly scalable to high-resolution data without the heavy pre-processing or summarization often required by Bayesian methods. Empirical results demonstrate its strong performance compared to existing statistical and deep learning methods. We applied the XAI model to the ABCD data which revealed associations between cortical features and ALFF throughout the entire brain. Our model performed comparably to existing methods in predictive accuracy but provided superior uncertainty quantification and faster computation, demonstrating its effectiveness for large-scale neuroimaging analysis. Open-source software in Python for XAI is available.

Cooperative game theory methods, notably Shapley values, have significantly enhanced machine learning (ML) interpretability. However, existing explainable AI (XAI) frameworks mainly attribute average model predictions, overlooking predictive uncertainty. This work addresses that gap by proposing a novel, model-agnostic uncertainty attribution (UA) method grounded in conformal prediction (CP). By defining cooperative games where CP interval properties-such as width and bounds-serve as value functions, we systematically attribute predictive uncertainty to input features. Extending beyond the traditional Shapley values, we use the richer class of Harsanyi allocations, and in particular the proportional Shapley values, which distribute attribution proportionally to feature importance. We propose a Monte Carlo approximation method with robust statistical guarantees to address computational feasibility, significantly improving runtime efficiency. Our comprehensive experiments on synthetic benchmarks and real-world datasets demonstrate the practical utility and interpretative depth of our approach. By combining cooperative game theory and conformal prediction, we offer a rigorous, flexible toolkit for understanding and communicating predictive uncertainty in high-stakes ML applications.

As Large Language Models (LLMs) become increasingly integrated into real-world decision-making systems, understanding their behavioural vulnerabilities remains a critical challenge for AI safety and alignment. While existing evaluation metrics focus primarily on reasoning accuracy or factual correctness, they often overlook whether LLMs are robust to adversarial manipulation or capable of using adaptive strategy in dynamic environments. This paper introduces an adversarial evaluation framework designed to systematically stress-test the decision-making processes of LLMs under interactive and adversarial conditions. Drawing on methodologies from cognitive psychology and game theory, our framework probes how models respond in two canonical tasks: the two-armed bandit task and the Multi-Round Trust Task. These tasks capture key aspects of exploration-exploitation trade-offs, social cooperation, and strategic flexibility. We apply this framework to several state-of-the-art LLMs, including GPT-3.5, GPT-4, Gemini-1.5, and DeepSeek-V3, revealing model-specific susceptibilities to manipulation and rigidity in strategy adaptation. Our findings highlight distinct behavioral patterns across models and emphasize the importance of adaptability and fairness recognition for trustworthy AI deployment. Rather than offering a performance benchmark, this work proposes a methodology for diagnosing decision-making weaknesses in LLM-based agents, providing actionable insights for alignment and safety research.

Large language models (LLMs) can amplify misinformation, undermining societal goals such as the UN SDGs. We study three documented drivers of misinformation (valence framing, information overload, and oversimplification) often shaped by default beliefs. Building on evidence that LLMs encode such defaults (e.g., "joy is positive", "math is complex") and can act as "bags of heuristics", we ask whether belief-driven heuristics behind misinformation-related behaviour can be recovered from black-box LLM behaviour as explicit rules. A key obstacle is that global rule-extraction methods in explainable AI (XAI) are built for numerical input-output data, not text. We address this by eliciting global LLM beliefs and mapping them to numerical scores via statistically validated abstractions, enabling off-the-shelf global XAI to detect belief-driven heuristics. For ground truth, we inject nonlinear behavioural triggers of increasing complexity (univariate, conjunctive, non-convex) into GPT-family and Llama models via system instructions. We find that RuleFit often misses non-univariate triggers, while global SHAP better ranks conjunctive trigger features but yields no symbolic rules. To bridge this gap, we propose RuleSHAP, a rule-extraction algorithm that couples global SHAP aggregates with rule induction to better capture non-univariate triggers, improving MRR@1 over RuleFit by +82% on average. Our results suggest a practical pathway for surfacing behavioural triggers in LLMs.

Noise Injection Systemically Degrades Large Language Model Safety Guardrails

Prithviraj Singh Shahani, Kaveh Eskandari Miandoab, Matthias Scheutz

Safety guardrails in large language models (LLMs) are a critical component in preventing harmful outputs. Yet, their resilience under perturbation remains poorly understood. In this paper, we investigate the robustness of safety fine-tuning in LLMs by systematically injecting Gaussian noise into model activations. We show across multiple open-weight models that (1) Gaussian noise raises harmful-output rates (p < 0.001) by up to 27%, (2) that deeper safety fine-tuning affords no extra protection, and (3) that chain-of-thought reasoning remains largely intact. The findings reveal critical vulnerabilities in current safety alignment techniques and highlight the potential of reasoning-based and reinforcement learning approaches as promising direction for developing more robust AI safety systems. These results have important implications for real-world deployment of LLMs in safety-critical applications as these results imply that widely-deployed safety tuning methods can fail even without adversarial prompts.

Integrating heterogeneous biomedical data including imaging, omics, and clinical records supports accurate diagnosis and personalised care. Graph-based models fuse such non-Euclidean data by capturing spatial and relational structure, yet clinical uptake requires regulator-ready interpretability. We present the first technical survey of interpretable graph based models for multimodal biomedical data, covering 26 studies published between Jan 2019 and Sep 2024. Most target disease classification, notably cancer and rely on static graphs from simple similarity measures, while graph-native explainers are rare; post-hoc methods adapted from non-graph domains such as gradient saliency, and SHAP predominate. We group existing approaches into four interpretability families, outline trends such as graph-in-graph hierarchies, knowledge-graph edges, and dynamic topology learning, and perform a practical benchmark. Using an Alzheimer disease cohort, we compare Sensitivity Analysis, Gradient Saliency, SHAP and Graph Masking. SHAP and Sensitivity Analysis recover the broadest set of known AD pathways and Gene-Ontology terms, whereas Gradient Saliency and Graph Masking surface complementary metabolic and transport signatures. Permutation tests show all four beat random gene sets, but with distinct trade-offs: SHAP and Graph Masking offer deeper biology at higher compute cost, while Gradient Saliency and Sensitivity Analysis are quicker though coarser. We also provide a step-by-step flowchart covering graph construction, explainer choice and resource budgeting to help researchers balance transparency and performance. This review synthesises the state of interpretable graph learning for multimodal medicine, benchmarks leading techniques, and charts future directions, from advanced XAI tools to under-studied diseases, serving as a concise reference for method developers and translational scientists.

Large Language Models (LLMs) have demonstrated human-like capabilities in language comprehension and generation, becoming active participants in social and cognitive domains. This study investigates whether LLMs exhibit personality-like traits and how these traits compare with human personality, focusing on the applicability of conventional personality assessment tools. A behavior-based approach was used across three empirical studies. Study 1 examined test-retest stability and found that LLMs show higher variability and are more input-sensitive than humans, lacking long-term stability. Based on this, we propose the Distributed Personality Framework, conceptualizing LLM traits as dynamic and input-driven. Study 2 analyzed cross-variant consistency in personality measures and found LLMs' responses were highly sensitive to item wording, showing low internal consistency compared to humans. Study 3 explored personality retention during role-playing, showing LLM traits are shaped by prompt and parameter settings. These findings suggest that LLMs express fluid, externally dependent personality patterns, offering insights for constructing LLM-specific personality frameworks and advancing human-AI interaction. This work contributes to responsible AI development and extends the boundaries of personality psychology in the age of intelligent systems.