There are apps for everything, from online banking, social software to shopping. They have become one of the most important tools in our daily lives. This even implies that a mobile device has stored most of personal information, including photos, credit cards, and communications. If an intruder succeeds in hacking into the mobile device, all private properties must suffer from the leakage threats. Undoubtedly, a malware is the commonest tool used by an attacker to compromise a mobile phone. In particular, it is often disguised as a popular application through an obfuscated or packed form. That is the main reason why it is difficult to distinguish a malware from the legal ones. In this article, we have adopted machine learning technique to develop a static analysis mechanism for Android malware detection based on shallow feature and permission correlation (AMD). AMD first analyses the Application Programming Interfaces of the target to detect all possible and hidden privilege threats. It then filters this obfuscation information using permission correlation to eliminate noise and identify meaningful malicious indicators. The proposed approach leverages the correlation patterns between permissions and API calls to distinguish suspicious behaviours from legitimate ones. Thus, AMD can extract all the representative shallow features to achieve the high detection rate. Simulation results have shown that AMD can outperform related works under the datasets of CICAndMal2017 and CICMalDroid2020, which confirms the effectiveness of shallow features and permission correlation.

As Artificial Intelligence (AI) reshapes education, professional development (PD) must go beyond tool training to foster critical, meaningful integration. Initial PD should introduce AI’s uses and challenges, but also address the impact on teaching and learning. This paper explores and reflects upon Phase II of the FAITH project, a transatlantic design-based initiative developing an AI and Education (AI&ED) model for higher education. Effective AI pedagogy is grounded in socially constructed, hands-on experiences where educators design lessons, generate content, and critically assess AI outputs. Such approaches build confidence, competence, and prevent mechanical adoption. Leadership and policy must further support a dual PD strategy: immediate classroom applications alongside preparation for broader societal shifts. Early FAITH findings show introductory courses spark essential dialogue, but PD must remain dynamic, ethical, and intentional. Phase II combines theoretical exploration (e.g., sustainability, ethics) with context-relevant practice. Ultimately, AI&ED should be understood as a lifelong professional learning journey.

The rapid integration of generative AI (GenAI) into core organizational infrastructure requires a rethinking oftraditional governance models. This paper explores how organizations can effectively govern GenAI at scale whilemaintaining strategic, ethical, and societal alignment. By synthesizing key theoretical frameworks, including institutionaltheory and dynamic capabilities, we propose a conceptual framework organized around three interconnected domains:Model Stewardship, Operational Alignment, and Strategic Guardrails. We contend that scalable governance must evolvefrom static compliance to a dynamic, adaptable organizational capability. The paper concludes with the introduction of theGenAI Governance Maturity Model (GAI-GMM), a research-based tool for institutional alignment.

As generative artificial intelligence (AI) technologies—such as ChatGPT, DALL·E, and other large language and image models—become increasingly mainstream, they introduce new ethical, legal, and governance challenges that are particularly urgent in developing countries. Nigeria, Africa’s most populous nation and a regional technology hub, offers a compelling case study of how these technologies are being adopted in environments with minimal regulatory infrastructure and limited public awareness. This paper examines the ethical and societal implications of generative AI in Nigeria and interrogates the country's preparedness to manage these risks. Despite the creation of the National Centre for Artificial Intelligence and Robotics (NCAIR) in 2020 and the recent passage of legislation such as the Nigeria Data Protection Act (2023) and the Startup Act (2022), Nigeria lacks a unified national AI formal risk classification systems, or sector-specific ethical guidelines. These gaps are important given the widespread, unregulated use of generative AI tools in education, politics, and digital commerce. In higher education, students increasingly rely on generative AI for assignments and projects, raising concerns about academic integrity in a system already strained by infrastructural deficits. Meanwhile, in the political domain, deepfake videos and AI-generated misinformation have circulated in election periods, threatening democratic stability in a media world prone to disinformation and weak content regulation. The paper compares Nigeria’s regulatory trajectory with global trends, particularly the European Union’s Artificial Intelligence Act and similar initiatives in Kenya, South Africa, and Rwanda. It highlights how Nigeria’s reactive approach to AI governance contrasts sharply with more proactive global models. Sectoral analysis reveals risks including digital labour displacement, cultural misrepresentation through foreign-trained models, algorithmic bias, and the erosion of public trust. Ultimately, the study calls attention to Nigeria’s urgent need for a comprehensive, context-sensitive AI ethics and governance framework. Through an analysis grounded in local realities and informed by global comparisons, the paper contributes to broader conversations about equitable, responsible AI adoption in the Global South.

An organisation's security is fundamentally reliant on its people. Regardless of the sophistication of its cybersecurity infrastructure, the absence of comprehensive training and awareness can lead to vulnerabilities. Traditional phishing awareness training typically involves sending simulated phishing emails to employees, allowing organisations to monitor actions such as link clicks, email reporting, and responses. While this method offers valuable insights into employee behaviour, it often struggles to engage users effectively. This conventional approach may not create a dynamic learning environment conducive to better retention of vital security practices. Furthermore, users generally do not receive immediate feedback regarding their interactions with phishing links, leaving organisations more susceptible to social engineering attacks. This research seeks to address the issue by developing an interactive gamified phishing simulator that employs reinforcement learning (RL). The methodology for this study consists of two key components. First, a literature review was conducted to assess existing phishing awareness techniques and explore how RL can be applied effectively. This review examined the integration of RL within cybersecurity education and explored the impact of gamification on user behaviour. For the RL agent, a dataset comprising both phishing and legitimate emails was compiled. The agent was then trained to discern phishing emails from legitimate ones based on various email features. Then the agent presents users with email challenges and delivers real-time feedback on their selections. The simulator incorporates a reward and badge system that promotes active participation and ongoing learning. This approach aims to overcome the limitations traditionally associated with static phishing training by fostering continuous learning, ultimately reducing user susceptibility. The effectiveness of the proposed simulator was evaluated based on its classification accuracy of phishing and legitimate emails.

Books—both fiction and nonfiction—are among the first consumer goods that artificial intelligence (AI) can fully generate at a quality indistinguishable from that created by humans. While prior research in domains such as art and music documents a preference bias against AI-created works, the transferability of these findings to typical consumption contexts—and whether such biases persist amid rapid AI diffusion—remains unclear. This study examines the effect of authorship disclosure on consumer preferences for AI- versus human-authored books, while also exploring psychological and economic mechanisms that may explain these differences. It further investigates heterogeneity of this effect by factors such as familiarity with AI to anticipate future states of AI adoption. For empirical evaluation of the effects, we conducted an online experiment that systematically manipulated authorship of books from multiple literary genres, using a representative sample of approximately 1,500 U.S. adults. Results showed a consistent preference bias against AI-generated books across genres. Among the potential explanatory mechanisms, perceived author effort, emotional attachment, and perceived proximity to the author emerged as the most influential. Heterogeneity analysis among early adopters and AI welcomers revealed that these effects are weakened but remained persistent. The findings indicate that consumers’ reluctance towards AI-created books arises not only from assessments of product utility but also from psychological factors. Although technological advances may alleviate consumer quality concerns, persistent social and emotional attachments are likely to result in separate market segments for AI- versus human-created works.

Collaborative Problem Solving (CPS) is a vital 21st-century skill that demands nuanced coordination of cognitive, social, and regulatory processes among learners. In face-to-face classrooms, peer dialogue offers rich data for studying CPS, but manual annotation of such unstructured, oral interaction is labour-intensive and difficult to scale. This study investigates whether open-source Large Language Models (LLMs), including Llama and Qwen, can perform inductive qualitative coding on classroom peer dialogues using zero-shot prompting alone—without fine-tuning or training data. We collected over 210,000 words of dialogue transcripts from 38 student dyads across two CPS tasks at university classrooms: an engineering design activity and a GenAI-supported peer assessment of lesson plans. Through a multi-phase process, we iteratively developed three zero-shot prompting strategies (self-prompting, chain-of-thought prompting, and in-context prompting) via GPT-4o interactions and deployed them across different LLMs via API access. Our findings suggest that in-context prompting consistently yields context-sensitive and theoretically coherent CPS constructs. Chain-of-thought prompting facilitates abstract reasoning but may lead to overgeneralization, while self-prompting demonstrates autonomous logic yet lacks consistency. Expert evaluations using a five-dimensional rubric (clarity, concreteness, objectivity, granularity, specificity) show moderate to high alignment between human and LLM-generated codes, although LLMs tend to overrate clarity and coherence. We further analyse discrepancies between LLMs and academic frameworks such as PISA CPS and ATC21S, and highlight challenges such as terminological drift, low recall, and theoretical misalignment. This work contributes a scalable, human-centered workflow for inductive coding of classroom dialogue and provides ready-to-use prompt templates for educational researchers. The aim of this study is to critically examine whether open-source LLMs can inductively code classroom peer dialogue in collaborative problem-solving tasks, while acknowledging both their potential and limitations in educational practice. We conclude with a dual-pathway strategy for combining practice-oriented, behaviourally grounded constructs with theory-aligned coding schemes, and offer design recommendations for future human-AI collaborative tools in learning analytics and classroom assessment.

As artificial intelligence reshapes the cybersecurity landscape, the demand for a trustworthy, real-time intelligence platform to track security incidents has become mission-critical. This paper proposes AGS-INTEL, an AI-driven platform designed to revolutionize data breach intelligence by providing a credible, real-time repository that consolidates, verifies, and contextualizes global security incidents. Unlike traditional databases, AGS-INTEL employs a validated scoring algorithm and enriched metadata to capture breach dimensions (legal, technical, sectoral, geopolitical), drawing from GDPR/HIPAA disclosures, threat intelligence, dark web forums, and academic reports, among other sources. Utilizing NLP and agentic AI, it extracts structured metadata from unstructured narratives while integrating ethical data scraping, regulatory compliance, and cross-jurisdictional filtering to ensure high fidelity. A visual analytics dashboard empowers stakeholders, including regulators, policymakers, cybersecurity professionals, and journalists, to analyze breach trends by industry, geography, and threat modality, enhancing transparency and risk governance. By delivering authenticated, actionable data, AGS-INTEL addresses critical gaps in existing tools, setting a new standard for ethical AI in breach intelligence and strengthening societal resilience against escalating cyber threats.

As artificial intelligence (AI) technologies become increasingly embedded in higher education, from adaptive learning platforms to algorithmic assessment tools, there is a growing imperative to examine the ethical and governance implications of such systems. While AI holds significant promise for improving efficiency and personalisation in education, its deployment also raises concerns around surveillance, algorithmic bias, data privacy, and equity. This article critically explores how transparent AI governance frameworks can be systematically integrated into academic curricula. It argues that ethical oversight and trust must be embedded not merely at the institutional policy level but within the design, delivery, and evaluation of educational programmes themselves. Drawing on interdisciplinary insights from digital governance, education policy, and data ethics, the paper develops a conceptual model for curriculum reform, articulates the ethical dimensions of AI in education, and presents illustrative case studies of institutions that have begun embedding such oversight into their teaching practices. By aligning curriculum design with emerging ethical principles, this study proposes that universities and higher education institutions can develop graduates who are not only technically proficient but ethically literate, capable of navigating and interrogating the socio-technical dimensions of AI in their future professions.

The deployment of large language models (LLMs) in production environments faces significant challenges due to computational and energy requirements during inference. This paper presents a comprehensive empirical analysis of quantization methods applied to the Qwen3 model family, ranging from 0.6B to 32B parameters. We evaluate six quantization approaches: GPTQ 4-bit, GPTQ 8-bit, AWQ, FP8 W8A8 and INT8 W8A8, and the original FP16 baseline across six established benchmarks (MMLU, HumanEval, TruthfulQA, MetaBench, GSM8K, ARC Challenge). Our analysis examines the relationship between model size, quantization method, accuracy preservation, energy consumption, and inference performance across various context lengths. We demonstrate that larger Qwen3 models exhibit increased resilience to quantization-induced accuracy degradation, while aggressive quantization methods provide substantial energy savings with acceptable trade-offs in model performance. These findings provide crucial insights for optimizing LLM deployments in resource-constrained environments.