Busola Motunrayo Olawale

Enterprise adoption of artificial intelligence (AI) creates sustainability, social and security risks that firms typically govern through separate ESG and cybersecurity structures. This study asks whether, and under what conditions, large U.S. firms disclose an integrated approach to governing AI across the three domains. It uses a configurational, text-as-data design: the latest Form 10-K of each S&P 500 registrant (N = 500; analysis sample, n = 488) was converted into dictionary-based measures of AI, AI-governance, environmental, social and cybersecurity disclosure, sentence-level co-mention of AI with each risk domain, and an eleven-element score of the Form 10-K Item 1C cybersecurity-governance disclosure. Latent Dirichlet allocation of 9,444 AI-context sentences recovered six themes. Latent profile analysis identified four overlapping disclosure profiles (sustainability-led and AI-light, social-led with weak cyber governance, AI-intensive integrators, and security-anchored moderates) that differ by sector and by year-on-year growth in AI-governance language. Fuzzy-set qualitative comparative analysis showed two equifinal paths to integrated AI-governance disclosure: social-risk framing combined with intensive AI use, which was retained in every robustness specification, and a narrower path combining environmental framing, AI use and large size. Item 1C cybersecurity governance did not enter either main path, but its role is uncertain because the score varies little across firms. AI was mentioned within Item 1C by 24.8% of firms, and 8-K Item 1.05 incident disclosures did not distinguish profiles. Because all measures come from automated coding of disclosure, the findings describe reported rather than verified governance, and the cross-sectional design does not support causal claims. The findings motivate a Unified ESG–Cybersecurity Framework for AI Governance that applies the four NIST AI RMF functions jointly across the three domains through one AI inventory, one risk taxonomy, one reporting line and one disclosure narrative.

Agentic AI systems plan and act through tools (code execution, database queries, file access, network requests) that enterprises now use for business analysis, and the governance of these systems depends on where a human is kept in the loop. This study asks where human approval gates should sit, how rapidly the software layer through which agents act is being secured, and how accountable its disclosure practice is. The study used every GitHub-reviewed security advisory affecting 365 LLM-application and agent packages on PyPI and npm (N = 2,172 advisories across 346 packages, through 4 October 2026) and mapped each exploit path, using its CWE class, to the control point at which a gate could intercept it. The design is event-historical and Bayesian rather than regression-based: Kaplan–Meier and package-stratified Cox models of vulnerability dwell time and post-disclosure remediation, Bayesian hierarchical models with package random effects for autonomous exploitability, severity and fix-at-disclosure, and a Bayesian decision analysis of gate configurations. Advisory volume rose sharply in 2026 (81.9% of records), led by MCP tool servers and agent frameworks. Agent-action surfaces (prompt injection, execution, queries, file access, network requests and data egress) account for 52.6% of advisories and are more severe but not more autonomously exploitable than platform flaws; the median dwell time of vulnerable code is 248 days and 79.2% of advisories are fixed at disclosure, but the package, rather than the surface, governs remediation speed and most late fixes never arrive. Five approval gates would intercept 59.6% (95% credible interval 55.3–64.0%) of the 499 high- or critical-severity paths that need neither user interaction nor privileges, with execution and egress gates contributing most and returns diminishing quickly. Residual risk, computed as 1 − e × coverage for an assumed, rather than observed, reviewer effectiveness of e, stays above 0.46 at e = 0.9 and cannot fall below 0.40 even with a perfect reviewer, so gates complement rather than replace platform controls. The findings ground a three-tier Human-in-the-Loop Governance Framework (security gates, accountability ledger, decision-control authority) for agentic business analysis.

Transportation agencies are being asked to sustain ageing road, bridge and rail networks under tightening budgets, rising traffic loads and a changing climate, and artificial intelligence is increasingly presented as the mechanism through which asset management will become predictive rather than reactive. At the same time, the sensing, communication and computation layers that make such prediction possible convert physical assets into cyber-physical systems whose availability now depends on the integrity of data and models. This review critically examines the evidence supporting artificial intelligence-driven asset management frameworks for transportation infrastructure, and evaluates how far those frameworks address cyber-resilience rather than assuming it. Literature was identified through open scholarly indexes and citation searching, with bibliographic verification of every retained source, and was appraised for methodological adequacy, evidentiary strength and relevance rather than citation count alone. Three findings dominate the synthesis. First, the perception and prediction literature is technically mature but rests on condition data whose labelling reliability is itself contested, which places an unacknowledged ceiling on reported accuracy and undermines cross-study comparison. Second, sequential decision models, particularly those based on deep reinforcement learning, produce policies whose apparent superiority is conditional on simulated deterioration dynamics that have rarely been validated against observed network behaviour. Third, the cybersecurity evidence base concentrates on vehicles, communication channels and perception models, leaving the asset management pipeline itself, from field sensing through data platforms to maintenance programming, largely unexamined as an attack surface. Frameworks that claim to unify performance and resilience typically do so architecturally, without demonstrating that security controls preserve decision quality or that decision optimisation preserves security. The central unresolved question is therefore not whether artificial intelligence can improve asset management, but whether its benefits survive adversarial conditions, degraded data and institutional constraints. Research priorities include reliability-aware benchmarking, field validation of learned maintenance policies, transport-specific threat modelling of asset data pipelines, and joint evaluation of performance and resilience objectives.

Critical infrastructure operators increasingly deploy artificial intelligence (AI) to detect degradation, predict failures, optimise maintenance, identify cyber anomalies and support operational decisions. These capabilities create a plausible route to higher availability, reliability and quality, yet they also couple asset-performance decisions to data integrity, model uncertainty, software supply chains, human oversight and operational technology security. This critical narrative review examines how secure AI can contribute to asset performance without weakening safety or resilience. Literature spanning predictive maintenance and prognostics, industrial AI, digital twins, industrial cyber-physical security, adversarial machine learning, explainability, uncertainty, machine-learning operations and human factors was critically synthesised. The evidence is strongest for AI as a decision-support layer for condition monitoring, fault diagnosis and targeted maintenance where data provenance is controlled, failure modes are sufficiently represented and recommendations remain bounded by engineering constraints. Evidence for fully autonomous optimisation in high-consequence infrastructure is less mature because benchmark accuracy does not directly establish operational utility, transferability or safe behaviour under distribution shift and hostile manipulation. Digital twins can improve contextual diagnosis and testing, but their value depends on fidelity, synchronisation, governance and protection of the data-model-actuation pathway. Cybersecurity and safety therefore cannot be treated as independent assurance domains: poisoned training data, adversarial inputs, compromised updates or unavailable models can become physical reliability and quality risks. Explainability alone is also insufficient for assurance; uncertainty calibration, abstention, independent validation, auditability and meaningful human authority are required. The synthesis proposes a secure asset-performance loop in which AI recommendations are evaluated against safety, security, service and quality constraints, with outcomes fed back into monitored model and process improvement. The central implication is that asset performance should be maximised as a constrained socio-technical objective rather than as an unconstrained prediction or utilisation metric.

The increasing sophistication of intelligent cyber warfare, in which adversaries exploit artificial intelligence to automate reconnaissance, generate polymorphic malware, and conduct machine-speed attacks, has rendered conventional perimeter security and static Zero Trust implementations inadequate for autonomous and self-managing networks. This study addresses the absence of an integrated, adaptive architecture by designing and analytically evaluating an AI-driven adaptive Zero Trust framework that unifies behavioural analytics, federated intrusion detection, explainable trust scoring, and autonomous policy enforcement within a NIST-aligned model. Adopting a quantitative, experimental, and simulation-based design, the framework was evaluated using public benchmark datasets including CICIDS2017, UNSW NB15, and BoT IoT, with standardised preprocessing, balanced resampling, and stratified cross-validation. Eight classifiers were trained, among which gradient boosting achieved an accuracy and F1 score of approximately 0.9999 on the CICIDS2017 benchmark after leakage-prone identifier features were removed, while ensemble and convolutional models performed strongly. The dynamic trust engine, exercised on an illustrative cohort of ten simulated agents, enforced conservative session-level access decisions, and a simulated three-node federated learning configuration produced an aggregated F1 score of 0.9185, quantifying the privacy-performance trade-off under heterogeneous partitions. These findings support a coherent, identity-centric defence delivering continuous, explainable verification. The study contributes a conceptual architectural blueprint validated through simulation rather than an operationally deployed system, and sampled datasets and simulated agents constrain operational generalization, motivating future validation on live autonomous network testbeds.

The accelerating adoption of artificial intelligence across United States healthcare systems has intensified concerns over cyber resilience, data privacy, and regulatory compliance, yet existing governance frameworks remain fragmented and difficult to compare. This study addressed that gap by developing a governance-centred multi-criteria evaluation model (a structured governance-scoring instrument rather than a machine-learning system) that evaluates and ranks established frameworks on a single quantitative scale. Twelve governance constructs were derived from framework analysis and a control coverage mapping against adversary tactics, then weighted by combining the analytic hierarchy process with the entropy weight method to balance expert judgement and data-driven objectivity. The integrated weights produced composite readiness indices, dimensional resilience, privacy, and compliance scores, and a five-level maturity classification. Robustness was confirmed through bootstrap resampling, sensitivity perturbation, and consistency testing, returning a consistency ratio of 0.058 and rank stability of 0.958, while principal component analysis retained 86.3 percent of variance (interpreted descriptively given the small number of frameworks, n = 6). All six frameworks exceeded the readiness threshold, achieving a mean readiness of 0.782, with ISO/IEC 27001 ranked highest at 0.869 and the only Adaptive classification. The analysis isolated transparency and human oversight as the weakest constructs. The resulting instrument offers healthcare decision makers a reproducible, transparent, and internally consistent basis for strengthening accountability and prioritising critical governance improvements.

The rapid digitalisation of banking has increased exposure to financial cybercrime, while existing fraud-detection methods remain limited by single-modality data, class imbalance, concept drift, and poor explainability. This study developed and evaluated the Generative AI-Enhanced Multimodal Transformer Framework (GAMT-Fraud), an explainable artificial intelligence model that integrates transactional, behavioural, and network data for fraud detection. The framework combines a multimodal attention transformer, gradient-boosted learning, and variational autoencoder-based anomaly detection, while synthetic minority augmentation addresses data imbalance. Using a quantitative experimental design, the framework was trained and validated on the Institute of Electrical and Electronics Engineers Computational Intelligence Society (IEEE-CIS) and PaySim benchmark datasets through stratified data partitioning, five-fold cross-validation, and bootstrap significance testing. Performance was evaluated using precision, recall, F1-score, area under the ROC curve (AUC), Matthews correlation coefficient, and precision-recall area. Results showed that GAMT-Fraud consistently outperformed conventional machine-learning and deep-learning baselines across both datasets, achieving statistically significant improvements in fraud-detection performance. Shapley-value-based explainability further enhanced transparency and regulatory compliance by providing interpretable decision insights. The study demonstrates that integrating generative AI, sequential learning, and relational analysis within a unified framework can improve fraud-detection effectiveness. It contributes a scalable, explainable, and auditable fraud-intelligence architecture and provides a replicable foundation for future research in multimodal and adversarial financial fraud detection. The framework is presented as an experimental and auditable proof of concept rather than evidence of immediate real-world deployment.

This research developed and evaluated hybrid privacy-enhancing technologies (PETs) for secure sharing of land use and land cover (LULC) data across collaborative research institutions and open data ecosystems. The work was motivated by privacy risks, including re-identification via linkage attacks, inference of sensitive attributes, and data-sovereignty concerns in geospatial environmental datasets. A comprehensive literature review established the theoretical foundations of federated learning (FL), differential privacy (DP), secure multi-party computation (SMPC), homomorphic encryption (HE), and blockchain. The methodology employed a quantitative, simulation-based design. Because access to real institutional archives was beyond the scope of this study, synthetic Sentinel-2-like patches were generated programmatically and augmented with EuroSAT and NLCD-derived class schemas. These were partitioned into non-IID institutional silos to emulate heterogeneous collaborations. Hybrid PET pipelines integrated FL with DP via Opacus, SMPC aggregation, HE verification, and blockchain-based audit logging of model updates. Evaluation used accuracy, F1-score, the (ε, δ) budget tracked by a Rényi accountant, communication cost, and statistical tests. Following the standard DP convention, a smaller ε corresponds to more added noise and stronger privacy, while a larger ε relaxes privacy in exchange for utility. Results revealed substantial privacy-utility trade-offs and confirmed the expected inverse relationship between privacy level and utility. The unusually high centralized accuracy and the near-zero vanilla-FL accuracy are interpreted as artifacts of the small synthetic corpus and the lightweight model, rather than realistic estimates. The study addresses key gaps by providing practical blueprints for secure LULC collaboration. Conclusions affirm the potential of PETs, while highlighting the architectural refinements required for deployment at scale. Recommendations include scaling to real archives, mitigating non-IID heterogeneity, adopting post-quantum-ready cryptography, and embedding indigenous data-sovereignty safeguards into PET design.

Increasing climate variability, particularly recurrent droughts and floods, poses significant threats to soil health and agricultural productivity, especially in climate-sensitive regions. Climate-resilient soil management practices have emerged as important strategies for enhancing soil stability, water retention, and agricultural resilience under extreme environmental conditions. This study investigates the effects of drought and flooding on soil systems and agricultural productivity, while evaluating the effectiveness and adoption of climate-resilient soil management practices within an integrated analytical framework. The study focuses on climate-sensitive agricultural regions in Nigeria, Ethiopia, and Tanzania between 2010 and 2024 using open-access datasets obtained from NASA, FAOSTAT, World Bank climate databases, and LSMS-ISA surveys. A combination of fixed effects panel regression, Difference-in-Differences estimation, binary logistic regression, and Structural Equation Modeling was employed to examine environmental, behavioral, and institutional interactions influencing agricultural resilience. The results show that drought and flooding significantly reduce crop yield, with estimated effects of -0.401 and -0.284 respectively, while soil moisture exerts a strong positive influence (0.537), highlighting its critical role in mediating climate impacts. The adoption of soil management practices increases crop productivity by 34.2%, confirming their causal effectiveness. However, adoption is strongly influenced by institutional factors, as access to extension services increases the likelihood of adoption by over 146%, outperforming both farmer knowledge and policy support. Furthermore, system-level analysis reveals that both soil resilience (0.514) and adoption (0.486) significantly contribute to yield outcomes, demonstrating the interdependence of environmental, behavioral, and institutional systems. These findings suggest that enhancing agricultural resilience requires coordinated strategies that strengthen extension services, improve farmer capacity, and align policy incentives with effective implementation mechanisms.

This research developed a simulation-based framework for quantifying the carbon footprint and operational efficiency of machine-speed cybersecurity defenses in autonomous industrial control systems (ICS) and smart manufacturing environments. Traditional high-energy intrusion detection systems (IDS) relying on intensive machine learning models contribute significantly to carbon emissions, creating an urgent need for green cybersecurity solutions that balance robust threat detection with sustainability. Using publicly available datasets mirroring UNSW-NB15 and CIC-IDS2017, the study modeled and compared lightweight models (Random Forest and Logistic Regression) against an intensive Long Short-Term Memory network. Carbon emissions were tracked with CodeCarbon, while the proposed eco-efficiency index (EEI) integrated F1-score, carbon footprint, and latency to evaluate trade-offs. Results demonstrated that the ultra-light Logistic Regression model achieved up to 91% carbon reduction and ultra-low latency (<3 ms), yielding the highest EEI while satisfying machine-speed requirements for real-time ICS protection. The GreenShield-ICS framework offers a reproducible approach for low-carbon defenses in Industry 4.0, with implications for guiding policy standardization and accelerating industry adoption toward net-zero, energy-efficient cybersecurity practices.