Research Papers

Recent work has highlighted the importance of monitoring chain-of-thought reasoning for AI safety; however, current approaches that analyze textual reasoning steps can miss subtle harmful patterns and may be circumvented by models that hide unsafe reasoning. We present a sentence-level labeled dataset that enables activation-based monitoring of safety behaviors during LLM reasoning. Our dataset contains reasoning sequences with sentence-level annotations of safety behaviors such as expression of safety concerns or speculation on user intent, which we use to extract steering vectors for detecting and influencing these behaviors within model activations. The dataset fills a key gap in safety research: while existing datasets label reasoning holistically, effective application of steering vectors for safety monitoring could be improved by identifying precisely when specific behaviors occur within reasoning chains. We demonstrate the dataset's utility by extracting representations that both detect and steer safety behaviors in model activations, showcasing the potential of activation-level techniques for improving safety oversight on reasoning. Content Warning: This paper discusses AI safety in the context of harmful prompts and may contain references to potentially harmful content.

Mental health assessments are of central importance to individuals' well-being. Conventional assessment methodologies predominantly depend on clinical interviews and standardised self-report questionnaires. Nevertheless, the efficacy of these methodologies is frequently impeded by factors such as subjectivity, recall bias, and accessibility issues. Furthermore, concerns regarding bias and privacy may result in misreporting in data collected through self-reporting in mental health research. The present study examined the design opportunities and challenges inherent in the development of a mental health assessment tool based on natural language interaction with large language models (LLMs). An interactive prototype system was developed using conversational AI for non-invasive mental health assessment, and was evaluated through semi-structured interviews with 11 mental health professionals (six counsellors and five psychiatrists). The analysis identified key design considerations for future development, highlighting how AI-driven adaptive questioning could potentially enhance the reliability of self-reported data while identifying critical challenges, including privacy protection, algorithmic bias, and cross-cultural applicability. This study provides an empirical foundation for mental health technology innovation by demonstrating the potential and limitations of natural language interaction in mental health assessment.

Employers increasingly expect graduates to utilize large language models (LLMs) in the workplace, yet the competencies needed for computing roles across Africa remain unclear given varying national contexts. This study examined how six LLMs, namely ChatGPT 4, DeepSeek, Gemini, Claude 3.5, Llama 3, and Mistral AI, describe entry-level computing career expectations across ten African countries. Using the Computing Curricula 2020 framework and drawing on Digital Colonialism Theory and Ubuntu Philosophy, content analysis of 60 LLM responses to standardized prompts reveals consistent coverage of technical competencies such as cloud computing and programming, but notable differences in non-technical competencies, particularly ethics and responsible AI use. Models vary considerably in recognizing country-specific factors, including local technology ecosystems, language requirements, and national policies averaging only 35.4% contextual awareness overall. Open-source models demonstrated stronger contextual awareness and better balance between technical and professional skills, with Llama (4.47/5) and DeepSeek (4.25/5) outperforming proprietary alternatives ChatGPT-4 (3.90/5) and Claude (3.46/5). However, Mistral's poor contextual performance (0.00/4) despite being open-source indicates that development philosophy alone does not guarantee contextual responsiveness. This first comprehensive comparison of LLM career guidance for African computing students uncovers entrenched infrastructure assumptions and Western-centric biases that create gaps between technical recommendations and local realities. The findings challenge assumptions about AI tool quality in resource-constrained settings and underscore the need for decolonial approaches to AI in education, emphasizing contextual relevance and hybrid human-AI guidance models.

Debugging is a core application of explainable reinforcement learning (XRL) algorithms; however, limited comparative evaluations have been conducted to understand their relative performance. We propose a novel evaluation methodology to test whether users can identify an agent's goal from an explanation of its decision-making. Utilising the Atari's Ms. Pacman environment and four XRL algorithms, we find that only one achieved greater than random accuracy for the tested goals and that users were generally overconfident in their selections. Further, we find that users' self-reported ease of identification and understanding for every explanation did not correlate with their accuracy.

As design thinking education grows in secondary and tertiary contexts, educators face the challenge of evaluating creative artefacts that combine visual and textual elements. Traditional rubric-based assessment is laborious, time-consuming, and inconsistent due to reliance on Teaching Assistants (TA) in large, multi-section cohorts. This paper presents an exploratory study investigating the reliability and perceived accuracy of AI-assisted assessment compared to TA-assisted assessment in evaluating student posters in design thinking education. Two activities were conducted with 33 Ministry of Education (MOE) Singapore school teachers to (1) compare AI-generated scores with TA grading across three key dimensions: empathy and user understanding, identification of pain points and opportunities, and visual communication, and (2) examine teacher preferences for AI-assigned, TA-assigned, and hybrid scores. Results showed low statistical agreement between instructor and AI scores for empathy and pain points, with slightly higher alignment for visual communication. Teachers preferred TA-assigned scores in six of ten samples. Qualitative feedback highlighted the potential of AI for formative feedback, consistency, and student self-reflection, but raised concerns about its limitations in capturing contextual nuance and creative insight. The study underscores the need for hybrid assessment models that integrate computational efficiency with human insights. This research contributes to the evolving conversation on responsible AI adoption in creative disciplines, emphasizing the balance between automation and human judgment for scalable and pedagogically sound assessment.

Climate change demands effective legislative action to mitigate its impacts. This study explores the application of machine learning (ML) to understand the progression of climate policy from announcement to adoption, focusing on policies within the European Green Deal. We present a dataset of 165 policies, incorporating text and metadata. We aim to predict a policy's progression status, and compare text representation methods, including TF-IDF, BERT, and ClimateBERT. Metadata features are included to evaluate the impact on predictive performance. On text features alone, ClimateBERT outperforms other approaches (RMSE = 0.17, R^2 = 0.29), while BERT achieves superior performance with the addition of metadata features (RMSE = 0.16, R^2 = 0.38). Using methods from explainable AI highlights the influence of factors such as policy wording and metadata including political party and country representation. These findings underscore the potential of ML tools in supporting climate policy analysis and decision-making.

AutoRed: A Free-form Adversarial Prompt Generation Framework for Automated Red Teaming

Muxi Diao, Yutao Mou, Keqing He +6

The safety of Large Language Models (LLMs) is crucial for the development of trustworthy AI applications. Existing red teaming methods often rely on seed instructions, which limits the semantic diversity of the synthesized adversarial prompts. We propose AutoRed, a free-form adversarial prompt generation framework that removes the need for seed instructions. AutoRed operates in two stages: (1) persona-guided adversarial instruction generation, and (2) a reflection loop to iteratively refine low-quality prompts. To improve efficiency, we introduce a verifier to assess prompt harmfulness without querying the target models. Using AutoRed, we build two red teaming datasets -- AutoRed-Medium and AutoRed-Hard -- and evaluate eight state-of-the-art LLMs. AutoRed achieves higher attack success rates and better generalization than existing baselines. Our results highlight the limitations of seed-based approaches and demonstrate the potential of free-form red teaming for LLM safety evaluation. We will open source our datasets in the near future.

Fourier Analysis on the Boolean Hypercube via Hoeffding Functional Decomposition

Baptiste Ferrere, Nicolas Bousquet, Fabrice Gamboa +2

Fourier analysis on the Boolean hypercube is fundamentally defined as the orthogonal decomposition of the space of pseudo-Boolean functions with respect to the uniform probability measure. In this work, we propose an ANOVA-based generalization of the Fourier decomposition on the Boolean hypercube endowed with any arbitrary probability measure. We provide an \emph{explicit} decomposition basis which generalizes the Walsh-Hadamard (or parity functions) basis under any \emph{arbitrary} probability measure on the Boolean hypercube. We formulate the computation of the entire functional decomposition as a least squares problem and also provide a method to address the classical \emph{curse of dimensionality} challenge. We provide a comprehensive generalization of Fourier analysis on the Boolean hypercube, enabling the handling of non-uniform configuration spaces inherent to real-world machine learning tasks, \textit{e.g.} when dealing with \emph{one-hot encoded} features. Finally, we demonstrate its practical impact in the field of explainable AI, by conducting comparative studies with feature attribution methods such as SHAP or TreeHFD.

This study explores the integration of contextual explanations into AI-powered loan decision systems to enhance trust and usability. While traditional AI systems rely heavily on algorithmic transparency and technical accuracy, they often fail to account for broader social and economic contexts. Through a qualitative study, I investigated user interactions with AI explanations and identified key gaps, in- cluding the inability of current systems to provide context. My findings underscore the limitations of purely technical transparency and the critical need for contex- tual explanations that bridge the gap between algorithmic outputs and real-world decision-making. By aligning explanations with user needs and broader societal factors, the system aims to foster trust, improve decision-making, and advance the design of human-centered AI systems

Recent advances in image generation models have led to models that produce synthetic images that are increasingly difficult for standard AI detectors to identify, even though they often remain distinguishable by humans. To identify this discrepancy, we introduce \textbf{Mirage}, a curated dataset comprising a diverse range of AI-generated images exhibiting visible artifacts, where current state-of-the-art detection methods largely fail. Furthermore, we investigate whether Large Vision-Language Models (LVLMs), which are increasingly employed as substitutes for human judgment in various tasks, can be leveraged for explainable AI image detection. Our experiments on both Mirage and existing benchmark datasets demonstrate that while LVLMs are highly effective at detecting AI-generated images with visible artifacts, their performance declines when confronted with images lacking such cues.

IndiCASA: A Dataset and Bias Evaluation Framework in LLMs Using Contrastive Embedding Similarity in the Indian Context

Santhosh G S, Akshay Govind S, Gokul S Krishnan +2

Large Language Models (LLMs) have gained significant traction across critical domains owing to their impressive contextual understanding and generative capabilities. However, their increasing deployment in high stakes applications necessitates rigorous evaluation of embedded biases, particularly in culturally diverse contexts like India where existing embedding-based bias assessment methods often fall short in capturing nuanced stereotypes. We propose an evaluation framework based on a encoder trained using contrastive learning that captures fine-grained bias through embedding similarity. We also introduce a novel dataset - IndiCASA (IndiBias-based Contextually Aligned Stereotypes and Anti-stereotypes) comprising 2,575 human-validated sentences spanning five demographic axes: caste, gender, religion, disability, and socioeconomic status. Our evaluation of multiple open-weight LLMs reveals that all models exhibit some degree of stereotypical bias, with disability related biases being notably persistent, and religion bias generally lower likely due to global debiasing efforts demonstrating the need for fairer model development.

Industrial question-answering (QA) systems require higher safety and reliability than general-purpose dialogue models, as errors in high-risk scenarios such as equipment fault diagnosis can have severe consequences. Although multi-agent large language models enhance reasoning depth, they suffer from uncontrolled iterations and unverifiable outputs, and conventional distillation methods struggle to transfer collaborative reasoning capabilities to lightweight, deployable student models. To address these challenges, we propose Knowledge Graph-guided Multi-Agent System Distillation (KG-MASD). Our approach formulates distillation as a Markov Decision Process and incorporates a knowledge graph as a verifiable structured prior to enrich state representation and ensure convergence. By integrating collaborative reasoning with knowledge grounding, KG-MASD generates high-confidence instruction-tuning data and jointly distills reasoning depth and verifiability into compact student models suitable for edge deployment. Experiments on an industrial QA dataset show that KG-MASD improves accuracy by 2.4 per cent to 20.1 per cent over baselines and significantly enhances reliability, enabling trustworthy AI deployment in safety-critical industrial scenarios. Code and data are available at https://github.com/erwinmsmith/KG-MAD/.