News & Research
The latest AI research and news with real-world stakes — each item sourced, dated, summarized in plain English, and tagged by impact area. Every item is checked against its source before it appears.
Research
SkillFortify: Formal Analysis and Supply Chain Security for Agentic AI Skills
Varun Pratap Bhardwaj
Open MIND · 2026-08-05
SkillFortify is a formal security analysis framework designed to protect agentic AI skill supply chains—ecosystems like OpenClaw and Anthropic Agent Skills—from malicious skill injection, as demonstrated by real-world attacks such as the ClawHavoc campaign and the MalTool dataset of thousands of malicious tools. The framework combines a Dolev–Yao attacker model, sound static analysis via abstract interpretation, capability confinement proofs, SAT-based dependency resolution, and a trust score algebra, achieving 96.15% F1 with 100% precision and 0% false positives on a 540-skill benchmark, with dependency graphs resolved in a median 27 ms. A notable negative result is also reported: information flow analysis adds no detection coverage beyond pattern matching on the evaluated corpus. This work matters for enterprise and quality-assurance teams deploying agentic AI systems, as it provides formal guarantees—rather than heuristics—against supply chain attacks targeting AI skill marketplaces.
- Enterprise
- Quality assurance
Research
Local Violation Certification for Linear Predict-Then-Optimize Pipelines
Ş. İlker Birbil, Wenhao Chi
arXiv (Cornell University) · 2026-08-05
This paper presents a framework for certifying whether AI-driven predict-then-optimize decision pipelines comply with safety, fairness, and reliability constraints. Rather than relying on expensive repeated random testing, the authors show that for linear pipelines, the local risk of failure can be computed in closed form via a single optimization solve, along with exact feature-level attributions identifying which inputs drive non-compliance. The approach is demonstrated on an economic power dispatch system under emissions regulations, delivering auditable risk assessments at a fraction of traditional computational cost.
- Quality assurance
- Certifications
Research
Responsible Adoption of Artificial Intelligence in Institutional Review Board Proposal Review. A Narrative Review
Mostafa Kofi, Nawaf Alkhayat, Fahad Alhusain et al.
Scientia. Technology, science and society. · 2026-08-05
This narrative review synthesizes evidence from 15 peer-reviewed publications to evaluate how AI tools, including large language models, can responsibly be integrated into Institutional Review Board (IRB) proposal review. The authors find that AI can reduce specific review task times by approximately 25% and support consistency, with high accuracy on well-defined tasks (recall 0.90, precision 0.87, and 91% acceptance of machine-generated risk-of-bias judgements) when human oversight is maintained. The framework prioritizes lower-risk entry points such as completeness screening and protocol summarization, while requiring stricter validation for decision support and consent-related applications. The review concludes that a phased, pilot-driven adoption strategy with human-in-the-loop safeguards offers a responsible path to making IRB review faster and more consistent without replacing human ethical judgement.
- Quality assurance
- AI policy
Research
Authorless copyright: the special relevance of subject matter other than works in the era of artificial intelligence
Timothy Nugent
Griffith Law Review · 2026-08-05
This article examines how Australian copyright law applies to AI-generated content, focusing on 'subject matter other than works' (such as films and sound recordings) protected under Part IV of the Copyright Act 1968 (Cth). It argues that existing 'entrepreneurial copyright' frameworks—which do not require a human author—offer a sound conceptual foundation for protecting AI-generated outputs. Based on this analysis, the author calls for AI-generated works to be recognized as a distinct category within Part IV of the Copyright Act. The paper is directly relevant to how legal and policy frameworks may need to adapt to accommodate authorless AI-generated creative content.
- AI policy
Research
Implementing AI-enabled chest X-ray for community-based integrated screening for tuberculosis, chronic respiratory diseases, and cardiovascular diseases in Nigeria
Chidimma Okoye, Jude ILOZUMBA, John Oko et al.
BMC Global and Public Health · 2026-08-05
This study evaluated an AI-enabled portable chest X-ray screening programme across five Nigerian communities, screening 9,585 individuals for tuberculosis, cardiovascular disease, and chronic respiratory disease. AI flagged 33% of chest radiographs as abnormal; 204 people received bacteriologically confirmed TB diagnoses and 95% of those were initiated on treatment, demonstrating strong TB care linkage. However, among the 2,367 individuals with radiographic features suggestive of CVD or CRD, only 12% completed referral to tertiary facilities, highlighting that integrated screening is constrained by limited decentralisation of non-communicable disease services. The authors conclude that AI-enabled screening is feasible for TB case finding but must be paired with strengthened primary healthcare capacity and context-specific engagement to address non-TB conditions.
- AI policy
- Quality assurance
Research
Tools facilitate cheating, or partner supports learning? GenAI and academic integrity issues from students’ perspectives
Feyisa Mulisa, Tesfa Mezgebu
International Journal for Educational Integrity · 2026-08-05
This qualitative study explores undergraduate students' perspectives on generative AI and academic integrity, focusing on plagiarism, originality, and assessment fairness. Using reflexive thematic analysis with students at Ambo University, the study finds a paradox: students perceive GenAI as beneficial for overcoming language barriers and improving academic support, yet many view using it for coursework as only a minor violation. The research also identifies equity concerns, as GenAI creates uneven playing fields between students who complete work independently and those who submit AI-generated content. The authors conclude that students' attitudes toward GenAI shape their ethical use of it, calling for awareness-raising, policy frameworks, and enforcement mechanisms.
- AI policy
- Quality assurance
Research
Career Trajectories of Graduates of Educational Programs in the Field of Artificial Intelligence: Results of a Qualitative Study of a Regional University
A. О. Averyanov, Anna V. Simakova, Irina Stepus
Высшее образование в России · 2026-08-05
This qualitative study of 25 graduates from a regional university's AI master's programs found that only 7 (28%) were employed in AI-related roles after graduation. Key barriers included insufficient practical experience, prior IT employment that predated AI interest, and low initial motivation to work with AI technologies. The study also found that a graduate's professional identity as a researcher significantly boosted AI-field employment, and recommends strengthening the practical focus of AI curricula to improve workforce outcomes.
- Workforce
Research
Artificial Intelligence-Related Literacy and Fears Among Critical Care Nurses in Oman: A National Study
Shreedevi Balachandran, Joshua Kanaabi Muliira, Eilean Rathinasamy Lazarus et al.
Sci · 2026-08-05
A nationwide cross-sectional survey of 477 critical care nurses in Oman found low overall AI literacy (mean score 146.62 ± 84.03) alongside moderate fear of AI, particularly around job displacement and ethical concerns. Significant predictors of AI literacy included age, education level, marital status, prior IT or AI training, and work experience. The authors conclude that structured, hands-on continuing education programs with integrated ethical reflection are needed to build AI competency and self-efficacy among this workforce, especially for older, more experienced nurses with limited formal education.
- Workforce
Research
Acceptance of generative AI–assisted medical decision-making among Chinese physicians and patients and its ethical determinants: a cross-sectional survey
Zhen Qian, Xizheng Li, Yingxue Li et al.
Frontiers in Public Health · 2026-08-05
This cross-sectional survey of 533 Chinese physicians, patients, and healthcare stakeholders found moderate overall acceptance of generative AI-assisted medical decision-making (mean 3.68/5), with perceived functional value and ethical governance expectations as the strongest positive predictors of acceptance intention. Healthcare professionals perceived higher ethical risk than patients, and education level shaped governance expectations, with postgraduate respondents scoring highest. The findings suggest that ethical governance frameworks—not just technical capability—are a key lever for broader clinical adoption of AI decision-support tools.
- AI policy
- Workforce
Research
Human-Robot Collaboration and Organizational Citizenship Behavior: The Psychological and Emotional Mediating Pathways
Wen-Yan Duan, X.Z. Cui, Tung‐Ju Wu
Psychological Reports · 2026-08-05
Drawing on conservation of resources theory, this two-study field investigation examines how human-robot collaboration in manufacturing and logistics affects employees' organizational citizenship behavior (OCB). Results from both a two-wave survey (N=220) and a three-wave survey (N=294) show that working alongside robots increases OCB through the mediating mechanisms of cobot identity and reduced emotional exhaustion, with employees high in openness to experience showing amplified effects. The findings offer practical guidance for organizations seeking to structure human-robot collaboration in ways that encourage employees to reinvest their freed-up resources into prosocial workplace behaviors.
- Workforce
- Enterprise
Research
AI-Powered personalization vs. blockchain-based privacy: a systematic literature review of consumer trade-offs in digital marketing
Ebtisam LABIB
Frontiers in Blockchain · 2026-08-05
This systematic literature review applies the PRISMA framework to 56 peer-reviewed papers (2015–2024) examining the tension between AI-driven personalization and blockchain-based privacy in digital marketing. Three core themes emerged: AI enhances hyper-personalization and ROI, blockchain improves data security and GDPR compliance, and consumers trade convenience for privacy—accepting personalized marketing when it is transparent and user-controlled. Blockchain reduces some AI-related ethical risks such as data exploitation and lack of auditability, but does not address algorithmic bias or scalability. The authors call on policymakers to develop hybrid regulatory frameworks with GDPR-consistent consent mechanisms and global interoperability standards, while marketers adopt blockchain-audited AI systems to build consumer trust.
- AI policy
- Enterprise
Research
Could ambient artificial intelligence scribes offer a promising digital health solution to reduce physician burnout and documentation burden in Poland? A narrative review
Praveen Kumar Malik, Adam Bednarek
Polish Journal of Public Health · 2026-08-05
This narrative review examines the evidence for ambient AI scribes—tools that automatically convert clinician–patient conversations into structured clinical notes—as a solution to physician burnout and documentation burden in Poland. International studies, including randomised controlled trials, show these tools reduce documentation time, administrative load, and burnout, though patient acceptance varies with digital literacy and data-privacy trust. No peer-reviewed Polish implementation studies currently exist, but hospital pilots and national digital health strategies suggest growing readiness. The authors conclude that prospective Polish pilots are needed, and that deployment must comply with GDPR and the EU AI Act.
- Workforce
- AI policy
Research
Artificial Intelligence, External Audit Quality and Financial Accountability of Deposit Money Banks in Lagos, Nigeria
Tosin Olayemi Adeeko, Adedipe Oluwaseyi Ayodele
JOURNAL OF ACCOUNTING AND FINANCIAL MANAGEMENT · 2026-08-05
This survey-based study of 912 external auditors and audit specialists at the Big Four firms in Lagos, Nigeria finds that AI adoption in external auditing significantly improves financial accountability of deposit money banks. Four factors—AI adoption level, AI effectiveness in fraud detection, audit reliability/accuracy, and audit timeliness/efficiency—each positively and significantly predict financial accountability, with the combined model explaining 71.4% of variance (R²=0.714). AI effectiveness in fraud detection was the strongest single predictor (β=0.624, R²=0.573). The study recommends that the Central Bank of Nigeria establish a policy framework to guide responsible AI use in auditing while enhancing bank accountability.
- Quality assurance
- AI policy
Research
Mapping the landscape of AI governance in higher education
Magezi Samuel Khoza, Samuel Fosso Wamba, Serge Nyawa
Assessment & Evaluation in Higher Education · 2026-08-05
This study analyzes AI governance frameworks from 35 universities worldwide, examining 43 publicly accessible governance documents through qualitative thematic analysis. It finds increasing convergence around core ethical principles—fairness, transparency, and accountability—and a shift from reactive regulation to proactive governance. Universities are establishing formal oversight structures and adopting adaptable 'living' policies to keep pace with technological change. The findings point to growing international alignment and a need for collaborative, flexible governance frameworks that balance innovation with ethical safeguards.
- AI policy
Research
AI-supported neuroeducational methodologies and logical-mathematical thinking in teacher training
Margarita Narváez Ríos, Mayra Fernanda Quiñonez Bedón, Daniel Morocho-Lara et al.
Infinity Journal · 2026-08-05
This quasi-experimental study (n=480 pre-service teachers in Ecuador) tested a 14-week AI-supported neuroeducational intervention against conventional ICT-based mathematics instruction. ANCOVA results showed a significant group effect on post-test logical-mathematical thinking (F(1,477)=206, p<0.001, η²p=0.302), and linear regression confirmed that membership in the experimental group was a strong predictor of final performance (B=0.810, β=0.882). The findings suggest that AI-enhanced neurodidactic methodologies can meaningfully improve logical-relational reasoning, abstraction, and strategic problem solving in initial teacher training programs.
- Workforce
Research
Fra KI-adopsjon til agentberedskap: Tre målbare gap – og hvordan kunstige agenter endrer premissene for ledelse
Benja Fagerland
Magma · 2026-08-05
This paper analyzes open firm-level data and recent research to identify three measurable gaps—adoption, scaling, and capability—that separate current AI use from genuine organizational readiness for agentic AI systems. Key statistics cited include 20.2% firm-level AI adoption (OECD), 20.0% (EU), and 30% (Norway) in 2025, with large firms adopting at nearly three times the rate of small firms (52.0% vs. 17.4%). The authors argue that as AI systems gain greater agency in workflows and decision support, the core leadership challenge shifts from technology implementation to organization design, governance, and responsible oversight. The paper is relevant to enterprise leaders and policymakers navigating the transition from experimental AI tool use to broad, accountable deployment of agentic systems.
- Enterprise
- AI policy
Research
A Blockchain–Artificial Intelligence Hybrid Framework for Secure Assessment, Quality Assurance, and Student Trust in Distance Education
Mulugeta Tilahun Bekele
International Journal of Computer Science and Artificial Intelligence · 2026-08-05
This paper proposes a hybrid framework combining blockchain and artificial intelligence to address security, integrity, and trust challenges in distance education assessment. The system pairs blockchain's immutable record-keeping with AI-driven automated grading, anomaly detection, and plagiarism identification, and was evaluated against conventional cloud-based e-learning systems. Reported results include 99.8% assessment data integrity, 98.6% grading accuracy, 97.9% misconduct detection accuracy, a 64% reduction in verification time, and 96.7% student trust satisfaction. The authors conclude the framework offers a scalable and transparent alternative to centralized LMS architectures for improving institutional quality assurance and learner confidence.
- Quality assurance
- Certifications
Research
Artificial Intelligence, Customer Relationship Management, And Patronage Retention: Evidence from Real Estate Enterprises in Lagos, Nigeria
Akinsanya Alade Mohammed, Igbayilola Emmanuel Oladejo
IIARD INTERNATIONAL JOURNAL OF ECONOMICS AND BUSINESS MANAGEMENT · 2026-08-05
This study examines how AI-enabled customer relationship management (CRM) affects customer retention among real estate firms in Lagos, Nigeria. Surveying 202 marketing practitioners across 30 firms, regression analysis found that AI-powered CRM capability strongly predicts patronage retention (β = 0.671, p < 0.001), explaining 45% of the variance in retention outcomes. Real-time engagement, personalized communication, and automated response systems were the strongest drivers of sustained patronage. The findings offer practical guidance for real estate enterprises in African emerging markets seeking to use AI as a competitive advantage.
- Enterprise
Research
SAP S/4HANA Intercompany Matching and Reconciliation: Concepts, Processes, Benefits, Challenges, and AI Footprint
Jitender Sahu
American Journal of Technology · 2026-08-05
This study examines SAP S/4HANA's Intercompany Matching and Reconciliation (ICMR) module as a financial control tool for multinational organizations, measuring outcomes before and after implementation. Key reported results include up to a 50% reduction in reconciliation cycle times, approximately 30% reduction in manual labor, a 10–30 percentage-point increase in auto-match rates after rule refinement, and a 25–60 percentage-point reduction in exception aging. The paper also details how AI capabilities—including predictive matching, anomaly detection, and natural language reference interpretation—transform intercompany reconciliation from a manual bottleneck into an auditable, automated process. The findings suggest significant enterprise value in adopting AI-enhanced ICMR for faster financial closes and improved compliance reporting.
- Enterprise
- Quality assurance
Research
Türkiye’de akademik dergilerin yapay zekâ politikaları: DergiPark Akademik dergileri üzerine bir değerlendirme
Kasım Binici, Mehmet Ali Akkaya, Coşkun Polat
Bilgi Dünyası · 2026-08-05
This study examines how well academic journals hosted on DergiPark Akademik (DPA), Turkey's largest academic journal platform, have adopted AI use policies in their article acceptance processes. Analyzing 340 journals from a stratified sample of 2,181 journals via document analysis, the research finds that Turkish academic journals are significantly lacking in AI-related policies, guidelines, and information for authors, editors, and reviewers. Where policies do exist, they tend to be shallow in structure and author-centered, with weak institutional-level guidance for publishers, editors, and reviewers. The study calls on the Turkish academic community to more seriously adopt the AI ethics and policy standards established by international publishing organizations such as COPE, ICMJE, and major publishers.
- AI policy
- Quality assurance
Research
Certificate-Gated Clearance in Change-Impact Analysis: A Fail-Closed Architecture
Ankur Garg, Jeffrey Esposito
Qeios · 2026-08-05
RCII² is a knowledge-based architecture for automating regulatory change-impact analysis that only clears a governed artifact as unaffected when it can construct and independently verify a machine-checkable certificate; cases without an accepted certificate escalate rather than being cleared. On a synthetic corpus of 2,020 decisions, the system cleared 850 cases with zero false clearances (95% CI [0.0000, 0.0083]) and rejected all 600 tampered certificates across twelve mutation classes, compared to a 12.17% false-clearance rate from a supervised baseline clearing the same number of cases. The architecture distinguishes logically sufficient 'certifying' grounds from merely 'suggestive' grounds, ensuring that a negative decision can never be authorized by weak or unverifiable evidence. This matters for regulated organizations because an incorrect clearance silently removes a compliance obligation from review, and the fail-closed design prevents that failure mode.
- Certifications
- Quality assurance
- AI policy
Research
The three‐step test in international copyright—a global framework for generative AI training
Nicola Lucchi, Tim W. Dornis, Pascal T. Sierek
American Business Law Journal · 2026-08-05
This article examines how international copyright law's 'three-step test'—a binding treaty requirement that copyright exceptions not unfairly conflict with rights holders' economic interests—applies to the use of copyrighted materials in generative AI training datasets. The author argues that both the U.S. fair use doctrine and the EU's text-and-data-mining exceptions under the DSM Directive must operate within these international constraints, which have been largely overlooked in policy debates. The article critiques broad copyright exceptions for AI training and proposes conditions including transparency, audits, and compensation mechanisms to bring national approaches into compliance with international norms. The findings matter for AI policy because they suggest current and proposed legislative frameworks in both jurisdictions may need targeted reform to lawfully accommodate generative AI training.
- AI policy
Research
The Effect of Source Disclosure on Trust in an AI-Based Knowledge Management System
Ina Schiedermair, Marco Baumgartner, Elena Kick et al.
International Journal of Human-Computer Interaction · 2026-08-05
This vignette experiment with 590 employees tested how disclosing the human source of knowledge in an AI-based knowledge management system affects employee trust. Results show that source disclosure alone does not raise overall trust, but it allows users to calibrate trust according to the perceived credibility of the source. Personality traits such as risk affinity, people-pleasing propensity, and perfectionism, along with decision criticality, further shape employees' intentions to use the system. The findings have practical implications for designing transparent, human-centered AI tools in organizational settings.
- Enterprise
- Workforce
Research
GOVERNING THE ALGORITHM: THE REGULATORY FRAMEWORK FOR AI, COPYRIGHTS & CULTURAL CREATIVE INDUSTRIES FROM THE PERSPECTIVE OF INTERNATIONAL ORGANISATIONS
Berivan Aslan
Hacettepe hukuk fakültesi dergisi · 2026-08-05
This paper examines how international organizations—including WIPO, the EU, the Council of Europe, UNESCO, the OECD, and the UN—are developing regulatory frameworks to address the copyright and cultural challenges posed by Generative AI. It finds that most bodies currently favor non-binding soft-law instruments, while the 2024 EU AI Act stands out as the most decisive regional framework by requiring generative AI models to disclose summaries of copyrighted training data. The paper argues that the sustainability of cultural and creative industries depends on a shift toward partially binding international regulations to prevent cultural homogenization, protect local cultures, and ensure fair remuneration for human creators. A potential UNESCO Additional Protocol to the 2005 Convention is identified as a promising future step.
- AI policy
Research
Integration of Robotics Service on Guest Experience in Fine Dining Restaurants: A Case of Kileleshwa in Nairobi City County Kenya
Benard Binyanya
Journal of Hospitality and Tourism Management · 2026-08-05
This study examined how robotics and AI technologies affect guest experiences and employee reception in fine dining restaurants in Nairobi, Kenya. Using a census of 37 respondents (25 employees and 12 guests), findings showed that 92% of employees felt they could work alongside robotic systems and 84% reported improved staff efficiency, while all guests agreed that robotics improved dining satisfaction through accurate order delivery and 91.7% noted better experiences due to timely service. However, 66.7% of guests preferred human interaction over robotic service, acceptance was lower among elderly guests, and responses regarding accessibility for persons with disabilities were mixed. The study concludes that a hybrid service model combining robotics with retained human service is optimal, and that accessible design accommodating diverse guest needs should be prioritized.
- Workforce
- Enterprise