Research Papers

Large Language Models (LLMs) are increasingly integrated into academic research pipelines; however, the Terms of Service governing their use remain under-examined. We present a comparative analysis of the Terms of Service of five major LLM providers (Anthropic, DeepSeek, Google, OpenAI, and xAI) collected in November 2025. Our analysis reveals substantial variation in the stringency and specificity of usage restrictions for general users and researchers. We identify specific complexities for researchers in security research, computational social sciences, and psychological studies. We identify `regulatory gray areas' where Terms of Service create uncertainty for legitimate use. We contribute a publicly available resource comparing terms across platforms (OSF) and discuss implications for general users and researchers navigating this evolving landscape.

Neural networks are growing more capable on their own, but we do not understand their neural mechanisms. Understanding these mechanisms' decision-making processes, or mechanistic interpretability, enables (1) accountability and control in high-stakes domains, (2) the study of digital brains and the emergence of cognition, and (3) discovery of new knowledge when AI systems outperform humans. This paper traces how attention head intervention emerged as a key method for causal interpretability of transformers. The evolution from visualization to intervention represents a paradigm shift from observing correlations to causally validating mechanistic hypotheses through direct intervention. Head intervention studies revealed robust empirical findings while also highlighting limitations that complicate interpretation. Recent work demonstrates that mechanistic understanding now enables targeted control of model behaviour, successfully suppressing toxic outputs and manipulating semantic content through selective attention head intervention, validating the practical utility of interpretability research for AI safety.

This policy report draws on country studies from China, South Korea, Singapore, and the United Kingdom to identify effective tools and key barriers to interoperability in AI safety governance. It offers practical recommendations to support a globally informed yet locally grounded governance ecosystem. Interoperability is a central goal of AI governance, vital for reducing risks, fostering innovation, enhancing competitiveness, promoting standardization, and building public trust. However, structural gaps such as fragmented regulations and lack of global coordination, and conceptual gaps, including limited Global South engagement, continue to hinder progress. Focusing on three high-stakes domains - autonomous vehicles, education, and cross-border data flows - the report compares ethical, legal, and technical frameworks across the four countries. It identifies areas of convergence, divergence, and potential alignment, offering policy recommendations that support the development of interoperability mechanisms aligned with the Global Digital Compact and relevant UN resolutions. The analysis covers seven components: objectives, regulators, ethics, binding measures, targeted frameworks, technical standards, and key risks.

Generative AI models reproduce the human biases in their training data and further amplify them through mechanisms such as mode collapse. The loss of diversity produces homogenization, which not only harms the minoritized but impoverishes everyone. We argue homogenization should be a central concern in AI safety. To meaningfully characterize homogenization in Large Language Models (LLMs), we introduce a framework that allows stakeholders to encode their context and value system. We illustrate our approach with an experiment that surfaces gender bias in an LLM (Claude 3.5 Haiku) on an open-ended story prompt. Building from queer theory, we formalize homogenization in terms of normativity. Borrowing language from feminist theory, we introduce the concept of xeno-reproduction as a class of tasks for mitigating homogenization by promoting diversity. Our work opens a collaborative line of research that seeks to understand and advance diversity in AI.

Heart failure (HF) affects 11.8% of adults aged 65 and older, reducing quality of life and longevity. Preventing HF can reduce morbidity and mortality. We hypothesized that artificial intelligence (AI) applied to 24-hour single-lead electrocardiogram (ECG) data could predict the risk of HF within five years. To research this, the Technion-Leumit Holter ECG (TLHE) dataset, including 69,663 recordings from 47,729 patients, collected over 20 years was used. Our deep learning model, DeepHHF, trained on 24-hour ECG recordings, achieved an area under the receiver operating characteristic curve of 0.80 that outperformed a model using 30-second segments and a clinical score. High-risk individuals identified by DeepHHF had a two-fold chance of hospitalization or death incidents. Explainability analysis showed DeepHHF focused on arrhythmias and heart abnormalities. This study highlights the feasibility of deep learning to model 24-hour continuous ECG data, capturing paroxysmal events essential for reliable risk prediction. Artificial intelligence applied to single-lead Holter ECG is non-invasive, inexpensive, and widely accessible, making it a promising tool for HF risk prediction.

A major concern amongst AI safety practitioners is the possibility of loss of control, whereby humans lose the ability to exert control over increasingly advanced AI systems. The range of concerns is wide, spanning current day risks to future existential risks, and a range of loss of control pathways from rapid AI self-exfiltration scenarios to more gradual disempowerment scenarios. In this work we set out to firstly, provide a more structured framework for discussing and characterizing loss of control and secondly, to use this framework to assist those responsible for the safe operation of AI-containing socio-technical systems to identify causal factors leading to loss of control. We explore how these two needs can be better met by making use of a methodology developed within the safety-critical systems community known as STAMP and its associated hazard analysis technique of STPA. We select the STAMP methodology primarily because it is based around a world-view that socio-technical systems can be functionally modeled as control structures, and that safety issues arise when there is a loss of control in these structures.

Plant diseases pose a significant threat to global food security, necessitating accurate and interpretable disease detection methods. This study introduces an interpretable attention-guided Convolutional Neural Network (CNN), CBAM-VGG16, for plant leaf disease detection. By integrating Convolution Block Attention Module (CBAM) at each convolutional stage, the model enhances feature extraction and disease localization. Trained on five diverse plant disease datasets, our approach outperforms recent techniques, achieving high accuracy (up to 98.87%) and demonstrating robust generalization. Here, we show the effectiveness of our method through comprehensive evaluation and interpretability analysis using CBAM attention maps, Grad-CAM, Grad-CAM++, and Layer-wise Relevance Propagation (LRP). This study advances the application of explainable AI in agricultural diagnostics, offering a transparent and reliable system for smart farming. The code of our proposed work is available at https://github.com/BS0111/PlantAttentionCBAM.

Artificial intelligence systems are increasingly deployed in domains that shape human behaviour, institutional decision-making, and societal outcomes. Existing responsible AI and governance efforts provide important normative principles but often lack enforceable engineering mechanisms that operate throughout the system lifecycle. This paper introduces the Social Responsibility Stack (SRS), a six-layer architectural framework that embeds societal values into AI systems as explicit constraints, safeguards, behavioural interfaces, auditing mechanisms, and governance processes. SRS models responsibility as a closed-loop supervisory control problem over socio-technical systems, integrating design-time safeguards with runtime monitoring and institutional oversight. We develop a unified constraint-based formulation, introduce safety-envelope and feedback interpretations, and show how fairness, autonomy, cognitive burden, and explanation quality can be continuously monitored and enforced. Case studies in clinical decision support, cooperative autonomous vehicles, and public-sector systems illustrate how SRS translates normative objectives into actionable engineering and operational controls. The framework bridges ethics, control theory, and AI governance, providing a practical foundation for accountable, adaptive, and auditable socio-technical AI systems.

This study applies a Geospatial Explainable AI (GeoXAI) framework to analyze the spatially heterogeneous and nonlinear determinants of traffic crash density in Florida. By combining a high-performing machine learning model with GeoShapley, the framework provides interpretable, tract-level insights into how roadway characteristics and socioeconomic factors contribute to crash risk. Specifically, results show that variables such as road density, intersection density, neighborhood compactness, and educational attainment exhibit complex nonlinear relationships with crashes. Extremely dense urban areas, such as Miami, show sharply elevated crash risk due to intensified pedestrian activities and roadway complexity. The GeoShapley approach also captures strong spatial heterogeneity in the influence of these factors. Major metropolitan areas including Miami, Orlando, Tampa, and Jacksonville display significantly higher intrinsic crash contributions, while rural tracts generally have lower baseline risk. Each factor exhibits pronounced spatial variation across the state. Based on these findings, the study proposes targeted, geography-sensitive policy recommendations, including traffic calming in compact neighborhoods, adaptive intersection design, speed management on high-volume corridors such as I-95 in Miami, and equity-focused safety interventions in disadvantaged rural areas of central and northern Florida. Moreover, this paper compares the results obtained from GeoShapley framework against other established methods (e.g., SHAP and MGWR), demonstrating its powerful ability to explain nonlinearity and spatial heterogeneity simultaneously.

Understanding how prices evolve over time often requires peeling back the layers of market noise to identify clear, structural behavior. Many of the tools commonly used for this purpose technical indicators, chart heuristics, or even sophisticated predictive models leave important questions unanswered. Technical indicators depend on platform-specific rules, and predictive systems typically offer little in terms of explanation. In settings that demand transparency or auditability, this poses a significant challenge. We introduce the Stock Pattern Assistant (SPA), a deterministic framework designed to extract monotonic price runs, attach relevant public events through a symmetric correlation window, and generate explanations that are factual, historical, and guardrailed. SPA relies only on daily OHLCV data and a normalized event stream, making the pipeline straight-forward to audit and easy to reproduce. To illustrate SPA's behavior in practice, we evaluate it across four equities-AAPL, NVDA, SCHW, and PGR-chosen to span a range of volatility regimes and sector characteristics. Although the evaluation period is modest, the results demonstrate how SPA consistently produces stable structural decompositions and contextual narratives. Ablation experiments further show how deterministic segmentation, event alignment, and constrained explanation each contribute to interpretability. SPA is not a forecasting system, nor is it intended to produce trading signals. Its value lies in offering a transparent, reproducible view of historical price structure that can complement analyst workflows, risk reviews, and broader explainable-AI pipelines.

Cisco Integrated AI Security and Safety Framework Report

Amy Chang, Tiffany Saade, Sanket Mendapara +2

Artificial intelligence (AI) systems are being readily and rapidly adopted, increasingly permeating critical domains: from consumer platforms and enterprise software to networked systems with embedded agents. While this has unlocked potential for human productivity gains, the attack surface has expanded accordingly: threats now span content safety failures (e.g., harmful or deceptive outputs), model and data integrity compromise (e.g., poisoning, supply-chain tampering), runtime manipulations (e.g., prompt injection, tool and agent misuse), and ecosystem risks (e.g., orchestration abuse, multi-agent collusion). Existing frameworks such as MITRE ATLAS, National Institute of Standards and Technology (NIST) AI 100-2 Adversarial Machine Learning (AML) taxonomy, and OWASP Top 10s for Large Language Models (LLMs) and Agentic AI Applications provide valuable viewpoints, but each covers only slices of this multi-dimensional space. This paper presents Cisco's Integrated AI Security and Safety Framework ("AI Security Framework"), a unified, lifecycle-aware taxonomy and operationalization framework that can be used to classify, integrate, and operationalize the full range of AI risks. It integrates AI security and AI safety across modalities, agents, pipelines, and the broader ecosystem. The AI Security Framework is designed to be practical for threat identification, red-teaming, risk prioritization, and it is comprehensive in scope and can be extensible to emerging deployments in multimodal contexts, humanoids, wearables, and sensory infrastructures. We analyze gaps in prevailing frameworks, discuss design principles for our framework, and demonstrate how the taxonomy provides structure for understanding how modern AI systems fail, how adversaries exploit these failures, and how organizations can build defenses across the AI lifecycle that evolve alongside capability advancements.

Fairness-Regularized Online Optimization with Switching Costs

Pengfei Li, Yuelin Han, Adam Wierman +1

Fairness and action smoothness are two crucial considerations in many online optimization problems, but they have yet to be addressed simultaneously. In this paper, we study a new and challenging setting of fairness-regularized smoothed online convex optimization with switching costs. First, to highlight the fundamental challenges introduced by the long-term fairness regularizer evaluated based on the entire sequence of actions, we prove that even without switching costs, no online algorithms can possibly achieve a sublinear regret or finite competitive ratio compared to the offline optimal algorithm as the problem episode length $T$ increases. Then, we propose FairOBD (Fairness-regularized Online Balanced Descent), which reconciles the tension between minimizing the hitting cost, switching cost, and fairness cost. Concretely, FairOBD decomposes the long-term fairness cost into a sequence of online costs by introducing an auxiliary variable and then leverages the auxiliary variable to regularize the online actions for fair outcomes. Based on a new approach to account for switching costs, we prove that FairOBD offers a worst-case asymptotic competitive ratio against a novel benchmark -- the optimal offline algorithm with parameterized constraints -- by considering $T\to\infty$. Finally, we run trace-driven experiments of dynamic computing resource provisioning for socially responsible AI inference to empirically evaluate FairOBD, showing that FairOBD can effectively reduce the total fairness-regularized cost and better promote fair outcomes compared to existing baseline solutions.