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
AI-Assisted Test Execution as an Augmentation Layer in Enterprise Quality Engineering
Rejenish Kiran
International Journal of Computer Information Systems and Industrial Management Applications · 2026-08-30
This paper introduces the Probabilistic Augmentation and Governance Model (PAGM), a three-tier framework that formally divides responsibility between AI agents and human testers across autonomous execution, confidence-gated escalation, and human-led verification in regulated enterprise software environments. Applied to a property insurance release pipeline, PAGM enables a full quarterly regression cycle within a five-day SLA while preserving mandatory human accountability for premium calculation and claims workflows. The paper draws on literature showing contextual UI recognition methods achieve a 95% change-handling rate versus 40–80% for conventional tools, and that AI-based classifiers outperform classical approaches by 25–50% across software quality metrics. PAGM positions governance, traceability, and bounded autonomy as core design requirements rather than performance optimizations, offering a governance-ready foundation for AI-assisted testing at enterprise scale.
- Enterprise
- Quality assurance
Research
Fear, power, and superintelligence: A realist reframing of AI catastrophic risk
Carlo Burelli, Federico Formentini
European Journal of Political Theory · 2026-08-30
This paper argues that debates about catastrophic AI risk are too narrowly focused on value alignment—ensuring AI systems follow human intentions—and that this framing misses a deeper structural problem: superintelligence could concentrate autonomous power in ways that dominate all other actors. Drawing on political realism from political theory and international relations, the authors reframe AI safety concerns (instrumental convergence, race dynamics, loss of control) as problems of power under anarchy, noting these already rely on realist assumptions. They contend that solutions emphasizing ethical behavior fail to change the competitive incentives driving AI development, and instead propose that governance should make superintelligence visible as a shared hegemonic threat through mandatory incident reporting, shared evaluations, and institutionalized transparency to coordinate action before power becomes uncontrollable.
- AI policy
Research
Posthumanistic Considerations for the Human-Centric Approach to European Regulation of Artificial Intelligence
Andrej Krištofík
Law Technology and Humans · 2026-08-30
This article critiques the EU AI Act's human-centric regulatory framework, arguing that it reproduces an outdated liberal humanist anthropocentrism built on a human/machine dichotomy that undermines the Act's own stated goals. Using critical posthumanist methodology, the paper proposes an 'ontocentric posthuman jurisprudence' drawing on Floridi's information ethics and Braidotti's critical posthumanities as a more adequate foundation. The authors contend that transcending the human/machine binary would extend regulatory protections to non-human actors and the environment while better serving human values. The work is relevant to how AI governance frameworks are conceptualized and structured at the regulatory level.
- AI policy
Research
Exploring the Impact of Artificial Intelligence Integration on Indonesia Banking Sector
Boy Tjahyono, Muhtosim Arief, Willy Gunadi et al.
Aptisi Transactions On Technopreneurship (ATT) · 2026-08-30
This study surveys 181 senior executives across 30 Indonesian commercial banks to assess the extent and drivers of AI adoption in the banking sector. Findings show that 64.6% of banks have implemented AI, with adoption concentrated in digital operations (65.4%), customer analytics (51.6%), and risk management (23.9%). Larger, better-capitalized banks show significantly higher adoption intensity and maturity, with organizational readiness, capital strength, and ownership structure identified as key predictors. The authors recommend differentiated AI implementation strategies based on each bank's capital capacity and digital maturity, providing empirical evidence of AI's role in improving productivity and financial resilience in an emerging economy context.
- Enterprise
Research
When AI Policies Fail in Practice: Shadow AI as a Structural Policy–Practice Governance Misalignment
Mia Wilson, Ethan Moore
Journal of Management and Informatics · 2026-08-30
This conceptual study argues that Shadow AI—employees' unsanctioned use of AI tools—is not primarily a compliance or security failure but a structural symptom of misalignment between formal AI governance policies and workplace realities. The authors identify three dimensions of misalignment: temporal gaps (governance too slow for operational needs), utility gaps (sanctioned tools poorly matched to actual tasks), and autonomy–control gaps (tension between professional discretion and standardization). Drawing on sociotechnical systems theory and synthetic organizational scenarios, the study proposes adaptive governance models—such as curated AI tool marketplaces and expedited approval pathways—as more effective alternatives to rigid control regimes. The framework reframes Shadow AI as a diagnostic signal of systemic governance design flaws, offering organizations a foundation for more legitimate and responsive AI policy.
- AI policy
- Enterprise
Research
Arguments About Artificial Intelligence in Social Work: A Critical Perspective
David Hodgson
Australian Social Work · 2026-08-30
This critical essay examines four commonly made arguments about AI's potential benefits and risks in social work: that AI boosts efficiency, enhances clinical practice, complements rather than replaces human skills, and can have its biases corrected by practitioners. The author argues these presumed benefits require careful scrutiny, emphasizing that AI's realistic utility in social work is likely limited to low-stakes, routine, and standardized tasks rather than complex ethical or clinical decision-making. The paper calls for a more informed, critical conversation in the social work field about what is genuinely at stake when deploying AI in practice.
- Workforce
- AI policy
Research
Machine Learning-Enabled Automated Assessment and Grading in Education: A Systematic Literature Review of Techniques, Accuracy, and Fairness
Pritam Kumar
European Journal of Education · 2026-08-30
This systematic literature review synthesizes 27 peer-reviewed studies (2019–2025) on machine learning-based automated grading and assessment in education, examining techniques such as NLP, transformer models, and supervised learning across essay scoring, short-answer assessment, and programming tasks. The review finds automated grading works best for structured tasks, while complex writing and reasoning still require human oversight. The authors recommend institutions deploy these tools as supervised decision-support systems with fairness auditing, validation, transparency, and appeal procedures in place.
- Quality assurance
- AI policy
Research
Offloading cognition to artificial intelligence: A qualitative study of teachers' experiences and perceptions
Şengül Erden, Mehmet Fatih Döğer
e-Kafkas Eğitim Araştırmaları Dergisi · 2026-08-30
This qualitative study of 17 teachers finds that AI assistants are regularly used to offload cognitive tasks such as lesson planning, material design, and exam preparation, with participants reporting reduced mental effort, less time pressure, and higher self-efficacy. However, teachers also expressed concern that over-reliance on AI risks weakening critical thinking and self-evaluation skills—a pattern the authors call 'metacognitive laziness'—and described deliberate strategies like cross-checking outputs and personalizing content to counter this. The study concludes that AI literacy in teacher education needs to move beyond technical skills to include metacognitive awareness and self-regulation. These findings have direct implications for how educators are trained and supported as AI tools become embedded in professional practice.
- Workforce
Research
Public servants’ perspectives on AI adoption and AI literacy in Austrian public administration.
Gregor Eibl, Gabriela Viale Pereira, Valérie Albrecht
arXiv · 2026-08-29
This study investigates how stakeholders in Austrian public administration perceive the barriers and enabling conditions for AI adoption, and what AI literacy competences are needed. Findings show that key concerns include legal uncertainty, data protection, security risks, and accountability, while enablers include clear governance, leadership support, and training opportunities. The paper concludes that successful AI adoption requires both technological resources and context-specific competence development combining skills, knowledge, and attitudes across stakeholder groups.
- Workforce
- AI policy
Research
Detecting Collusion in LLM-based Multi-agent Financial Decision Systems: A Conditional-mutual-information Detector and a Modular Oversight Architecture
Ayman Nait Cherif, Mohamed Youssfi, Omar Bouattane
International journal of intelligent engineering and systems · 2026-08-29
This paper addresses the risk of covert collusion among large language model agents used in parallel financial decision-making, such as credit scoring. The authors develop a conditional mutual information detector that identifies whether two agents' outputs remain statistically dependent after conditioning out their shared input—a signature of coordination rather than independent competence. On synthetic tests the false-positive rate held near 0.05 and detection power reached 0.980 at moderate coordination strength; on real credit data with Claude Haiku 4.5 and GPT-4o, the detector flagged coordination in all 15 coordinated runs and raised no false alarms on honest agreement, outperforming correlation baselines and an LLM auditor. The detector is embedded in a proposed modular oversight architecture covering uncertainty decomposition, calibrated escalation, and incentive-compatible scoring, with implications for trustworthy deployment of multi-agent AI in financial and enterprise settings.
- Enterprise
- Quality assurance
Research
A systematic review of machine learning and explainable artificial intelligence for electrocardiogram based cardiovascular disease prediction
Sabit Ahamed Preanto, Md. Hasan Imam Bijoy, Tapon Paul et al.
Discover Artificial Intelligence · 2026-08-29
This systematic review of 75 studies examines machine learning and deep learning models for ECG-based cardiovascular disease prediction, finding generally high diagnostic performance—accuracy 90–99%, AUROC up to 0.98–0.99—across conditions such as arrhythmia, atrial fibrillation, and myocardial infarction. However, the review identifies critical gaps: few studies conducted external or prospective clinical validation, and challenges including class imbalance, limited generalizability, and poor model interpretability remain significant barriers to routine clinical deployment. The authors also flag ethical and legal concerns—data bias, privacy, and transparency—as prerequisites for responsible adoption. The review calls for more interpretable and clinically validated AI-assisted ECG diagnostic systems before widespread use can be responsibly pursued.
- Quality assurance
- AI policy
Research
Unsupervised Ensemble Learning for Active Drug‐Induced Liver Injury Surveillance: Integrating Pharmacokinetic Burden With Enzyme Trajectories
Thawatchai Nakkaratniyom, Sanita Hirunrassamee, Thapana Boonchoo et al.
Pharmacoepidemiology and Drug Safety · 2026-08-29
This study developed an unsupervised machine-learning ensemble to detect drug-induced liver injury (DILI) earlier than traditional static laboratory thresholds. Applied to over 616,000 treatment episodes in Thailand's national health records (2021–2024), the framework identified 17 sub-threshold cases missed by standard criteria, with over half clinically confirmed as DILI, and achieved a number needed to review of 1.43—indicating high triage efficiency. The system's negative predictive value of 92.6% and clinical plausibility rate of 69.8% (via blinded expert review) suggest it can serve as a scalable, complementary pharmacovigilance tool for proactive adverse drug reaction surveillance.
- Quality assurance
- AI policy
Research
Beyond the Default: How Customizable Artificial Intelligence Agents Can Attenuate Stereotypical Preferences
Marius Claudy, Anshu Suri, Shengnan Ren et al.
Psychology and Marketing · 2026-08-29
This paper investigates how the design of AI agents—specifically the salience of gender cues in the choice interface—can reduce consumers' tendency to prefer gender-stereotypical AI agents (e.g., female voices for support roles). Across four experiments with 2,530 participants, the authors find that when gender is made salient through design choices or explicit bias warnings, consumers are more likely to choose counterstereotypical agents, a shift mediated by action-efficacy beliefs—the sense that one's choice can promote gender equality. These effects were strongest among women in male-typed contexts. The findings offer actionable guidance for managers and policymakers designing AI systems that reduce rather than reinforce occupational gender stereotypes.
- AI policy
- Enterprise
Research
Governing automated credit after explainability: From transparency to contestability
Tariq K. Alhasan, Mohammed Alqaisi, Faisal Alabdallat
Social Sciences & Humanities Open · 2026-08-29
This article examines how explainability-focused governance of automated credit scoring can produce 'legible injustice'—decisions that appear procedurally transparent but remain systematically exclusionary in practice. The author identifies three mechanisms—proxy concentration, defensive rationing, and review theatre—through which compliance with transparency requirements may actually entrench rather than reduce bias. Drawing on EU, US, and UK regulatory frameworks, the article proposes a six-element standard for meaningful human oversight and argues for a right of contestability backed by outcome-facing supervision using distributional dashboards and proxy-use diagnostics.
- AI policy
- Enterprise
Research
Data sacrifices and the ‘third way’ toward AI: justification and critique in local conflicts over automated surveillance
Philipp Knopp
AI & Society · 2026-08-29
This paper analyzes parliamentary debates in Hamburg, Germany over a police AI surveillance project to show how Europe's 'third way' AI framework—which claims to balance innovation with public values—is used to justify large-scale data extraction for training CCTV surveillance systems. Using Critical Discourse Analysis and the concept of 'data sacrifices,' the authors reveal how civic values like privacy and equality are rhetorically internalized to legitimize surveillance industrialization while silencing certain critics. The case culminated in the first German law explicitly authorizing police transfer of surveillance data—including non-anonymized data—to external partners for machine learning. The findings expose the paradoxical nature of third-way compromises, which can simultaneously invoke democratic values and undermine them.
- AI policy
Research
Humanoid Colleagues and the Rheumatology Workforce: Preparing for a New Era
Adam Kilian, Laura Upton, Joshua Samec et al.
ACR Open Rheumatology · 2026-08-29
This review paper proposes a profession-centered governance framework for integrating humanoid robots into rheumatology clinical training, framed as anticipatory oversight before such technology is deployable. The authors conceptualize humanoid systems as supervised trainee analogs progressing through four training stages—observation, competency assessment, supervised patient contact, and parallel evaluation—rather than as autonomous clinicians, with a fixed boundary that they never achieve independent practice. The framework pairs each stage with technology-readiness and regulatory trigger milestones, with dual oversight from FDA device regulation and fellowship-anchored privileging. The authors argue that without proactive professional stewardship, AI integration into rheumatology will be shaped primarily by commercial and regulatory forces, potentially undermining patient trust and workforce sustainability.
- Workforce
- AI policy
Research
Navigating regulatory fragmentation in the convergence of synthetic biology, artificial intelligence, and automation
Ross K. J. McLennan, Paul S. Freemont, Isak S. Pretorius
Nature Communications · 2026-08-29
This paper analyzes how the convergence of synthetic biology, AI, and automation (SynBioxAI) creates regulatory challenges across biosecurity, AI governance, export control, and data sovereignty frameworks simultaneously. Through a cross-jurisdictional analysis of sixteen nations and seven realistic collaboration scenarios, the authors find that regulatory friction compounds multiplicatively rather than additively. They propose the SynBioxAI Regulatory Interoperability Toolkit (RIOT), a seven-lens institutional framework designed to help research institutions navigate divergent and fragmented regulatory landscapes efficiently and transparently.
- AI policy
- Certifications
Research
A Study into the Evolving Challenges in Regulating Artificial Intelligence and Machine Learning in the Future Legal Profession: Analysing Regulatory Gaps, Ethical Dilemmas and Adaptive Strategies
Sushanta Kumar Das, Shantanu Ganguly
International Journal of Law Management & Humanities · 2026-08-29
This qualitative study examines how existing regulatory frameworks for the legal profession have failed to keep pace with AI and ML tools such as predictive analytics, e-discovery platforms, generative-AI drafting assistants, and algorithmic dispute-resolution systems. Drawing on statutes, bar association guidance, judicial pronouncements, and peer-reviewed scholarship, the authors find that current regulatory architectures are largely reactive and fragmented across jurisdictions, leaving gaps around accountability, bias, confidentiality, and the unauthorized practice of law. The study recommends a principle-based, risk-tiered regulatory model supported by continuing legal education, algorithmic auditing, and cross-border cooperation to reconcile technological innovation with core professional duties.
- AI policy
- Certifications
Research
تأثير استخدام الذكاء الاصطناعي في تعزيز تنافسية الشركات الناشئة: دراسة كمية وصفية في السياق العربي
وداد محمد اللغبي, نوره علي الحقبان, وداد سعد الشهراني et al.
مجلة العلوم الإقتصادية و الإدارية و القانونية · 2026-08-29
This quantitative-descriptive study examined how AI adoption affects the competitiveness of startups in the Arab context, surveying 118 startups. Results showed that partial AI adoption was the most common pattern (48.3%), with recommendation and personalization systems rated most effective, while investment cost, change resistance, and regulatory constraints were the leading barriers. Statistical analysis found significant differences in effectiveness across adoption levels (F(3,114)=5.567, p=0.001) and a meaningful regression model (R²=0.195), though data-quality mediation and institutional-support interaction effects were not statistically significant; governance, however, strongly predicted internal readiness (R²=0.637). The findings highlight that financial and regulatory obstacles limit systematic AI adoption among Arab startups, underscoring the importance of supportive governance frameworks for improving competitive capacity.
- Enterprise
- AI policy
Research
A Study on the Role of Artificial Intelligence in E-Commerce and Its Cyber Law Implications
Shantanu Ganguly, Sushanta Kumar Das
International Journal of Law Management & Humanities · 2026-08-29
This study examines how AI technologies—including personalized recommendations, fraud detection, dynamic pricing, and automated logistics—are reshaping e-commerce and creating regulatory gaps under Indian cyber law. Using doctrinal-qualitative analysis of Indian statutes (IT Act 2000, Consumer Protection Act 2019, DPDP Act 2023), judicial precedents, and comparative instruments like the EU AI Act and GDPR, the authors find that India's current legal framework offers only partial, fragmented coverage of AI-driven e-commerce. The paper calls for a comprehensive, risk-calibrated legal framework addressing algorithmic transparency, liability for autonomous decisions, data protection, and cross-border enforcement to safeguard consumer rights and constitutional privacy guarantees.
- AI policy
- Enterprise
Research
Governing Artificial Intelligence in a Fragmented World: Toward a Multi-Level Policy Framework for Global AI Governance
Asher Odhiambo Ojuok, Julius Murumba, Elyjoy Micheni
East African Journal of Information Technology · 2026-08-29
This policy analysis paper examines how global AI governance can be structured across national, regional, and multilateral levels given growing regulatory fragmentation between the EU, US, and China. Drawing on primary legal documents including the EU AI Act, OECD AI Recommendations, a 2025 UN Resolution, and the Global Digital Compact, the paper finds that multilateral instruments currently lack binding enforcement and that low- and middle-income countries face capacity constraints even when proactively developing national strategies like Kenya's 2025-2030 AI Strategy. The paper proposes a three-tier 'layered subsidiarity' architecture—global normative coordination, regional regulatory clusters, and national implementation—arguing this approach, combined with capacity-building financing and interoperable technical standards, is more likely to achieve effective global AI governance than a binding treaty.
- AI policy
Research
Reproducibility artifact: EviRoute - Auditable Fail-Closed Routing for Sensor-Based Edge AI
Ngoc-Phuong Doan, The-Vinh Nguyen, Thi-Dung Nguyen et al.
Zenodo (CERN European Organization for Nuclear Research) · 2026-08-29
EviRoute (also called EdgeGuard) is a contract-based architecture for auditable, fail-closed routing of sensor-based edge AI inference, combining input-integrity checking, predictive-risk monitoring, action-feasibility gating, and deterministically replayable evidence logs. The reproducibility artifact provides full experimental support for the manuscript, including 78 automated tests, on-device benchmarks from a Samsung Galaxy Note20 and Pixel Watch 2, and a 48,000-decision assurance report. The headline result is a 66.71% reduction in wrong-accepts (40,016 to 13,320) under a label oracle at 47.45% local coverage on a human-activity-recognition dataset, though accepted faulted outputs remain approximately 58% wrong and integrity detectors average only 0.15 recall across a 36-condition graded battery. The authors explicitly scope this as a weak-model architecture case study, not deployment assurance, making it most directly relevant to quality-assurance and certification of edge AI systems.
- Quality assurance
- Certifications
Research
Artificial Intelligence (AI) on Construction Projects: Regulatory Position and Governance Gaps
Reihaneh Samsami
arXiv · 2026-08-29
This paper analyzes the regulatory landscape governing AI—particularly Large Language Models and multi-agent systems—deployed on construction projects, finding that governance frameworks have not kept pace with these technologies. Under EU AI Act (Regulation 2024/1689 as amended), the paper identifies five distinct regulatory positions for a single five-agent platform, ranging from prohibited practices to functions no reviewed instrument covers. U.S. frameworks such as NBIS, NCEES model law, and ASCE Policy Statement 573 constrain AI only through personal obligations of qualified engineers, not the software itself. Both traditions rely on human oversight as a primary control, yet the paper notes that four decades of automation bias research show this control fails when not properly specified and sized.
- AI policy
- Certifications
Research
Administrative Burden Documented in Medicaid Care Coordination
Sanjay Basu, Aaron Baum, Kiiera Robinson et al.
JAMA Health Forum · 2026-08-28
This retrospective cohort study used natural language processing to identify and quantify four types of administrative burdens—scheduling difficulties, transportation problems, paperwork requirements, and prior authorization delays—documented in Medicaid care coordination notes across nearly 50,000 beneficiaries in Washington, Virginia, and Ohio. Paperwork was the most prevalent burden (25.3% of engaged beneficiaries), while transportation carried the highest per-patient time cost ($47.58 at a clinician-equivalent wage rate), with documented burdens totaling 18,822 patient-hours and $628,665 cohort-wide. African American beneficiaries showed a 22% higher unadjusted burden prevalence than White beneficiaries, though payer-stratified analysis suggested this reflected enrollment in higher-burden plans rather than within-plan disparities. The study demonstrates that NLP applied to existing care coordination notes offers a scalable approach for managed care plans and state Medicaid agencies to monitor and address the most costly administrative burdens.
- AI policy
- Workforce
Research
Offline-Verifiable Accountability for Cross-Organization Agent Messaging: A Preserved Evidence-Bundle Approach
Adil Alshammari, Hayretdin Bahşi
arXiv (Cornell University) · 2026-08-28
This paper proposes a preserved evidence-bundle model that enables offline, verifiable accountability for agent-to-agent messaging across organizations. Each bundle captures policy-required evidence—including sender authentication, delegation authorization, signed checkpoints, append-only log continuity, and receiver-signed receipts—allowing a later auditor or dispute reviewer to assess evidence sufficiency without access to live systems. In a prototype evaluation across 300 workflows and 1,200 evidence bundles, the offline verifier rejected all corrupted or policy-insufficient bundles with no false acceptances observed. The approach is relevant to enterprise multi-organization workflows and policy-governed audit processes where platform-neutral, tamper-evident records are required.
- Enterprise
- AI policy