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
Cognitive Monoculture and the Individual-Team Reliance Paradox: A Formal Model of Shared AI Dependence in Organizational Decision Making
Kwan Hong TAN
International Journal of Web of Multidisciplinary Studies · 2026-08-20
This paper develops a formal mathematical model showing that when many employees rely on the same AI model, their judgments become statistically correlated rather than independent—a phenomenon the authors call 'cognitive monoculture.' The central finding is an individual-team reliance paradox: the AI reliance level that minimizes each individual's error can actually increase team-level error because shared AI errors do not diversify away through aggregation. Under an illustrative scenario, a 20-person team using a single shared AI model at individually optimal reliance sees its effective independent judgment count collapse to approximately 1.31, more than doubling team-level error. The paper argues this reframes AI governance away from simply counting human reviewers toward measuring the independence architecture of collective judgment.
- Enterprise
- AI policy
Research
From signals to trust: Multimodal detection and consumer perception of sustainability claims in e-commerce
Seyed Mohammad Sina Mirabdolbaghi, Adel Aazami, Sebastian Kummer
Electronic Commerce Research and Applications · 2026-08-20
This study develops a multimodal AI framework combining natural language processing, image-based color analysis, and machine learning to detect greenwashing in e-commerce by identifying mismatches between seller sustainability claims and consumer-reported experiences. Tested on two large datasets (Amazon and BaSalam), the pipeline achieves 81.6% accuracy on Amazon and 97% on BaSalam, demonstrating practical potential for real-time regulatory and consumer-facing screening. Notably, the study finds that sustainability-related language in reviews is not associated with higher product ratings, suggesting consumers prioritize functional criteria over green claims. The authors discuss implications for platform governance and sustainability communication policy.
- Quality assurance
- AI policy
Research
STEER-FL: A process model for federated learning derived from a dual-perspective systematic literature review
Ben Rachinger, Sven Meier, Jörg Franke et al.
Information and Software Technology · 2026-08-20
This paper presents STEER-FL, a structured end-to-end reference process model for federated learning (FL) projects, derived from a dual-perspective systematic literature review covering FL-specific and general ML/data science process literature. The authors identify recurring gaps in existing approaches—including incomplete lifecycle coverage, missing decision gates, and weak cross-organizational governance guidance—and synthesize a six-phase model with 30 activities, four roles, and explicit decision gates for suitability, feasibility, and production readiness. Eleven domain experts rated the model positively across acceptance, quality, and efficacy dimensions. STEER-FL is tool-agnostic and applicable to both cross-organizational and enterprise-internal FL settings, offering structured guidance for planning and operationalizing distributed AI systems.
- Enterprise
- AI policy
Research
Finance 5.0: Transforming Financial Decision-Making Through Artificial Intelligence and Intelligent Analytics
Sarmad Bin Saeed, Saba Mahmood, Abid Manzoor
Journal of Business Insight and Innovation · 2026-08-20
This study surveyed 420 finance professionals to measure how artificial intelligence and intelligent analytics affect financial decision-making in organizations that have adopted digital financial technologies. Using multiple regression analysis, the researchers found that AI (β = 0.463, p < 0.001) and intelligent analytics (β = 0.387, p < 0.001) each had a significant positive effect on financial decision-making, with the combined model explaining 68.4% of the variance. The findings indicate that Finance 5.0 technologies improve financial planning, forecasting, efficiency, and strategic decision-making. The authors note implications for financial institutions, corporate organizations, and policymakers seeking competitive advantage through digital financial transformation.
- Enterprise
- AI policy
Research
From Data Lakes to Trusted AI Infrastructure: Assessing Data Quality, Governance Maturity, and AI Readiness in U.S. Critical Infrastructure Agencies.
Chidinma Queen Adieze, Fabian Emesiani, Elo-Oghene Imonifano
International Journal of Computer Applications · 2026-08-20
This paper assesses data quality, governance maturity, and AI readiness across the 16 U.S. federally designated critical infrastructure sectors, drawing on GAO audits, OMB memoranda, CISA guidance, and industry reports from 2025. Key findings show that federal agencies nearly doubled reported AI use cases from 571 in 2023 to 1,110 in 2024, yet fundamental governance gaps persist: only 28% of organizations have formally defined AI oversight roles and fewer than 15% of agencies have networks fully optimized for AI workloads. The authors propose a five-tier Trusted AI Infrastructure Maturity Model (TAIMM) along with sector-specific governance recommendations and a strategic roadmap to close the gap between data lake accumulation and operationalized AI capability. The work matters because it provides a structured framework for federal agencies to progress from fragmented data storage toward accountable, enterprise-grade AI infrastructure.
- AI policy
- Enterprise
- Quality assurance
Research
Artificial Intelligence‐Driven Inequality: Gendered and Intersectional Experiences of Workplace Discrimination in Europe
Guðbjörg Linda Rafnsdóttir, Ángel S. Marrero, Carlotta Rigotti et al.
Gender Work and Organization · 2026-08-20
Using survey data from 4,417 individuals across 28 European countries, this study finds that women, older individuals, non-White individuals, those in non-permanent employment, and people identifying as 'other gender' are more likely to perceive or experience discrimination from AI applications in hiring and workplace management. Applying a feminist intersectional framework, the authors show that AI systems risk reproducing and amplifying existing inequalities among vulnerable groups. The findings underscore that whose knowledge and values are embedded in AI tools shapes workplace fairness, and that achieving genuinely trustworthy AI workplaces remains difficult given entrenched inequality.
- Workforce
- AI policy
Research
Transformasi Audit Internal melalui Artificial Intelligence: Systematic Literature Review Periode 2021–2026
Ibnu Hadi, Lidya Primta Surbakti
KINDAI · 2026-08-20
This systematic literature review synthesizes evidence from 21 Scopus-indexed articles (2021–2026) on how AI is transforming internal audit functions. Machine Learning is found to be the most dominant technology, followed by Robotic Process Automation, Hybrid AI, and Natural Language Processing, with primary applications in risk assessment, audit data analytics, and audit decision-making. AI adoption demonstrably improves analysis quality, accuracy, operational efficiency, and decision support, though barriers remain including auditor competency gaps, data quality issues, trust in AI, and explainability concerns. The findings affirm that AI enhances the effectiveness and value of internal audit while identifying research gaps in developing-country contexts, the public sector, and generative AI applications.
- Quality assurance
- Enterprise
Research
Unjustified trust and satisfaction in explainable artificial intelligence: The illusion of transparency persists across user expertise, even in objectively wrong outcomes
Saša Brdnik, Ivona Colakovic, Sašo Karakatič
Engineering Applications of Artificial Intelligence · 2026-08-20
This controlled experiment (n=96) found that SHAP-like explanations attached to AI-driven exam grades produced high user trust and satisfaction even when grades were artificially and unjustly lowered, with no statistically significant difference between fair and unfair outcome groups. Objective understanding of the explanations was low across all participants, and self-reported AI literacy among non-experts was inflated relative to actual comprehension. The findings suggest that XAI explanations can create an illusion of transparency rather than genuine understanding, and that AI expertise offers limited protection against this effect. The study calls for stricter evaluation standards before assuming explanation-based transparency adequately safeguards users in high-stakes domains like education, with direct implications for regulations such as the EU AI Act.
- AI policy
- Quality assurance
Research
Calibrated adaptive framework for trustworthy human and artificial intelligence decision systems
Maytha AL-Ali, Adam Marks, Abdallah A. Mohamed et al.
Scientific Reports · 2026-08-20
This paper introduces CAHAT (Calibrated Adaptive Human-AI Teaming), a closed-loop decision architecture designed for safety-critical industrial settings such as predictive maintenance. The framework combines adaptive trust updating grounded in utility theory, probability calibration via temperature scaling, epistemic uncertainty quantification through Monte Carlo Dropout, and semantic explainability to dynamically regulate human-AI collaboration. Simulation results show the adaptive hybrid system dramatically outperforms a human-only baseline (cumulative reward of 6,887.6 vs. −1,982.8) and achieves near-zero failure rates (0.8%) under realistic operational stress, while ablation analysis identifies miscalibration and human cognitive bias—not model capability—as dominant failure modes. The findings offer design principles for trustworthy AI deployment in high-stakes industrial environments, though the authors note that validation with real operators and operational data remains necessary.
- Enterprise
- Quality assurance
Research
The embodied ethics alignment problem of AI
Andrej Zwitter
Discover Artificial Intelligence · 2026-08-20
This paper argues that mainstream AI alignment research is fundamentally incomplete because it treats ethics as a formalizable rule-set or optimization target, ignoring that human moral agency emerges from embodiment, affect, vulnerability, and social context. The authors introduce the 'Embodied Ethics Alignment Problem' (EEAP), which holds that AI systems lacking these human conditions cannot be ethically aligned in any strong sense—they can only be behaviorally steered toward acceptable outcomes. The paper concludes that the real challenge is not building moral machines but governing non-moral optimization systems within human institutions that can bear genuine ethical responsibility. This has direct implications for how policymakers and institutions should frame AI oversight and governance.
- AI policy
Research
Would you trust and accept judges using generative artificial intelligence if you became victim of a crime?
Marvin Walczok, Friederike Funk
Technology Mind and Behavior · 2026-08-20
Two vignette-based online experiments with German participants (N=282 and N=464) examined how crime victims perceive judges who use generative AI in court decisions. The only statistically significant negative effect found was on perceived integrity (d=0.46 in Study 1), while Bayesian analyses supported the null hypothesis for trustworthiness, trust, and acceptance across most conditions. The authors conclude that while AI use by judges may trigger some unintended integrity concerns, the overall limited impact suggests hybrid human-AI judicial decision making could be viable. These findings are directly relevant to policy debates around AI deployment in legal and governmental institutions.
- AI policy
Research
Unlocking research output with ChatGPT- 4 and SciSpace Ai through the mediating and moderating roles of research orientation
Samuel Owusu, Mathew Thomas Gil, Bernard Tutu-Boahene et al.
Discover Artificial Intelligence · 2026-08-20
This study surveys 503 academic researchers across 20 universities in Ghana to test an AI–Research Output model examining how ChatGPT-4 and SciSpace AI affect research productivity. Structural Equation Modeling finds both tools are positively associated with research output (ChatGPT-4 β=0.387; SciSpace AI β=0.182), and that researchers' cognitive/methodological orientation mediates and differentially moderates these effects. The results suggest that AI tool effectiveness depends not just on the technology itself but on alignment with researchers' orientations, pointing to the need for targeted training and structured AI governance frameworks in academic institutions.
- Workforce
- AI policy
Research
A systematic review of artificial intelligence in small ruminant production systems: applications, performance outcomes, and reported implementation challenges
Abdullah Ghazy, Ibrahim Atta Abu El-Naser
BMC Veterinary Research · 2026-08-20
This systematic review synthesizes 92 peer-reviewed studies (2020–2025) examining AI applications in sheep and goat production across six domains, including behavior recognition (mean accuracy 92.4%), individual animal identification (mean accuracy 97.3%), and health/disease detection (mean accuracy 89.7%). Convolutional Neural Networks, YOLO-based models, and Random Forest algorithms were the most commonly used approaches. Despite strong performance under controlled conditions, the review finds that fewer than half of studies used fully independent external validation, and real-world translation is constrained by limited dataset diversity, class imbalance, environmental complexity, and hardware constraints. The authors conclude that standardized public datasets, transparent metric reporting, and deployment-focused validation across farms and breeds are needed before AI tools can be sustainably adopted in extensive livestock systems.
- Enterprise
- Quality assurance
Research
The constitutional foundation of explanation rights in EU digital regulation: An integrated approach
Melanie Fink, Simona Demková
Computer law & security review · 2026-08-20
This article examines how explanation rights embedded in EU digital regulations—the GDPR, DSA, and AI Act—lack a coherent constitutional foundation, producing two forms of fragmentation: overlapping obligations across instruments for a single actor, and distributed explanation duties within a single regime that separate legal responsibility from the expertise needed to fulfill it. The author proposes an integrated approach that anchors sectoral explanation rights in the constitutional duty to state reasons, operating at levels of implementation, adjudication, and legislative design. Two practical scenarios illustrate how this framework resolves inter- and intra-instrumental fragmentation. The article argues that future explanation requirements must be grounded in constitutional reasoning principles and coordinated with parallel sectoral provisions from the outset to achieve legal coherence and mutual reinforcement.
- AI policy
Research
Responsible and innovative AI for mental health care: five priority themes
Martin P. Paulus, John Torous, Roy H. Perlis et al.
NPP—Digital Psychiatry and Neuroscience · 2026-08-20
This paper synthesizes insights from a 2026 American College of Neuropsychopharmacology study group to outline five priorities for responsibly translating AI into clinical mental health care. The authors argue that the central barriers to impact are no longer computational but infrastructural—including unreliable measurement systems, incomplete governance frameworks, and insufficient standards for clinical evidence. Key priorities include building robust data and phenotyping pipelines, deploying clinician-facing tools like ambient documentation and decision-support systems, rigorously evaluating patient-facing AI for safety risks, extending governance frameworks to address bias and digital equity, and shifting toward causal and mechanistic approaches in precision psychiatry. The paper concludes that real-world effectiveness depends on disciplined integration into clinical workflows and governance structures, not model performance alone.
- AI policy
- Quality assurance
Research
Dynamics of Employee Adaptation, Anxiety, and Role Change Under AI Integration in Indonesia
Richa Afriana Munthe, Imran Al Ucok Nasution
International Journal of Applied Business and International Management · 2026-08-20
This qualitative study examines how employees in Indonesian banking, manufacturing, and digital-service organizations experience AI integration in their workplaces. Drawing on 18 semi-structured interviews with employees and HR practitioners, the research identifies five key themes: ambivalent perceptions of AI, anxieties over job displacement and surveillance, uneven readiness tied to digital literacy and psychological safety, role shifts toward analytical and relational work, and the need for transparent communication and learning culture. The findings indicate that successful AI adoption hinges not just on technological implementation but also on organizational support, psychological safety, and participatory change management. The study matters for understanding how AI is reshaping worker identity, job roles, and the conditions needed to manage workforce transitions effectively.
- Workforce
Research
‘ A Legal Whack‐a‐Mole’ . Escaping the Rider Law in Spain: Platform Strategies and the Hollowing of Enforcement Under Platform Capitalism
Martí López Andreu, Oriol Barranco
New Technology Work and Employment · 2026-08-20
This article investigates why Spain's Rider Law—a major regulatory effort to govern gig-economy platform work—has had limited impact on actual employment relations. Drawing on qualitative interviews with trade unions and rider organizations, the study finds that platform firms responded through organizational restructuring, algorithmic opacity, and performative compliance, effectively maintaining algorithmic management while formally satisfying legal requirements. The authors characterize enforcement as a contested, relational process undermined by digital technologies, weak workplace organization, and limited state capacity, contributing to debates on labor regulation and platform governance.
- Workforce
- AI policy
Research
Standardizing AI Model Transparency in U.S. Federal Agencies: A Framework for Implementing AI Model Cards and Data Provenance Auditing
Chidinma Queen Adieze, Fabian Emesiani, Taiwo Paul Onyekwuluje et al.
International Journal of Computer Applications · 2026-08-20
This paper proposes a five-tier standardization framework for implementing AI model cards and data provenance auditing across U.S. federal agencies. Drawing on recent policy developments—including OMB Memoranda M-25-21 and M-25-22, Executive Order 14179, and the NIST AI Risk Management Framework—the authors analyze the 2024 Federal AI Use Case Inventory, which disclosed over 2,133 agency AI deployments including 227 rights- and safety-impacting use cases. Their analysis reveals significant inconsistencies in current transparency reporting, gaps in third-party auditing capacity, and a need for interoperable provenance infrastructure. The paper concludes with policy recommendations directed at Congress, OMB, NIST, and Chief AI Officers to address these shortcomings.
- AI policy
- Certifications
- Quality assurance
Research
Delegating for the Greater Green? How competence and autonomy shape acceptance of agentic AI for carbon-intensive choices
Mariana Gaytan Camarillo, Charlie Wilson
Behaviour and Information Technology · 2026-08-20
This study investigates whether people are willing to delegate control to agentic AI systems to reduce carbon emissions, using a 3×2 online experiment with 1,522 participants across food shopping, digital usage, and email management tasks under full-autonomy or human-in-the-loop conditions. Delegation intentions were highest for digital usage and lowest for food shopping, where participants also showed a preference for human-in-the-loop over fully autonomous AI. Competence satisfaction—users' sense that AI credibly enhances their ability to achieve goals—emerged as the strongest psychological mediator of acceptance, while identity-relevance and hedonic value in food contexts exerted influence beyond the standard Self-Determination Theory framework. The findings suggest that sustainable AI design should prioritize utilitarian domains, avoid full automation in identity-relevant contexts, and jointly support user competence and autonomy.
- AI policy
- Enterprise
Research
The use of Artificial Intelligence in higher education admissions: a scoping review
Berna Sena Civan
Assessment in Education Principles Policy and Practice · 2026-08-20
This scoping review synthesizes 25 peer-reviewed studies (2020–2026) on how AI is being integrated into higher education admissions processes, drawn from searches across ERIC, Scopus, and Web of Science. The review identifies three themes: AI used by institutions in admissions systems, AI used by applicants in performance-based assessments, and issues of fairness and governance. The authors argue that AI in admissions goes beyond efficiency, reshaping how merit, performance, and fairness are defined and justified in high-stakes decision-making. The research base is predominantly quantitative and concentrated in the United States.
- AI policy
- Certifications
Research
dspy-security-bench: reproducible security, authorization, and mission assurance evidence for tool-using AI agents
Imran Ahamed
Open MIND · 2026-08-20
This paper presents dspy-security-bench, a Python benchmarking harness designed to measure how well language-model agents resist indirect prompt injection and related security threats. It introduces a suite of reproducible, auditable measurement protocols—including controlled policy-on/off experiments, repeated-execution uncertainty quantification, and cryptographic provenance via GitHub/Sigstore—spanning task utility, attack resistance, authorization, and mission assurance. The framework bridges major agent frameworks (OpenAI Agents SDK, LangChain, CrewAI, AutoGen, etc.) to a common evaluation contract and exports results in formats such as OSCAL 1.2.2 for federal compliance use. It matters because it provides verifiable, offline-auditable evidence for AI agent security and authorization decisions, directly supporting policy and certification processes for tool-using AI systems.
- Quality assurance
- Certifications
- AI policy
Research
dspy-security-bench: reproducible security, authorization, and mission assurance evidence for tool-using AI agents
Imran Ahamed
Zenodo (CERN European Organization for Nuclear Research) · 2026-08-20
This paper presents dspy-security-bench, a Python benchmarking harness for measuring how well language-model agents resist indirect prompt injection attacks. The framework wraps existing AgentDojo task environments and adds frozen, hashed measurement protocols, joint reporting of task utility alongside attack resistance, and cluster-bootstrap confidence intervals to determine when results are stable enough to make claims. It includes modules for policy-on/off comparisons, cyber-incident digital twins, delegated-authorization testing, and exports to federal compliance formats such as OSCAL 1.2.2, making security evidence auditable and reproducible across major agent frameworks including OpenAI Agents SDK, LangChain, CrewAI, and AutoGen. The work matters because it provides a standardized, provenance-backed method for generating verifiable security and authorization evidence for AI agents, directly supporting quality assurance and policy compliance assessments.
- Quality assurance
- AI policy
- Certifications
Research
Artificial Intelligence-Powered Accounting Systems and Accountant’s Job Security: Evidence from Selected Deposit Money Banks in Nigeria
Cynthia Onyinyechukwu Nwambe
IIARD INTERNATIONAL JOURNAL OF BANKING AND FINANCE RESEARCH · 2026-08-20
This survey-based study examined how AI-powered accounting systems affect the job security and employment stability of accountants at four major Nigerian deposit money banks (Access Bank, Zenith Bank, First Bank, and GTCO). Using regression analysis, the researchers found that AI adoption significantly reshapes—but does not eliminate—accounting roles, shifting responsibilities from routine manual tasks toward more strategic and analytical functions while increasing demand for digital skills. The study concludes that continuous technological training and professional adaptation are essential for accountants to remain relevant in an AI-driven environment.
- Workforce
Research
Five-year cost-effectiveness of AI for adult diabetic eye exams—a health system perspective
Mahnoor Ahmed, Michael D. Abramoff, Harold P. Lehmann et al.
npj Digital Medicine · 2026-08-20
This paper uses a 5-year Markov model to compare the cost-effectiveness of AI-based retinal screening, teleophthalmology, and traditional eye care professional (ECP) screening for diabetic retinal disease across U.S. health systems. AI-based strategies completed 3 times more screenings, identified 3.6–3.8 times more true positives, and led to 7.5–8.0 times more patients initiating treatment compared to ECP. The preferred strategy depends on health system scale and willingness-to-pay thresholds, with handheld and stationary AI becoming cost-effective at larger patient volumes due to real-time diagnostic feedback enabling earlier detection and reducing costs of advanced disease management. The findings provide concrete economic guidance for health systems considering AI-based diabetic eye screening programs.
- Enterprise
- AI policy
Research
AI adoption and innovation in MSMEs for enhancing competitiveness and firm performance
Mohammad Falahat, T. Ramayah, Pureheart Ogheneogaga Irikefe et al.
Benchmarking An International Journal · 2026-08-20
This study of 268 Malaysian MSMEs finds that environmental conditions and technological readiness significantly drive AI adoption, which in turn fuels AI-driven innovation, competitiveness, and multidimensional firm performance (technological, economic, and sustainability). Organisational support alone is not a sufficient adoption driver, and high firm performance requires minimum threshold levels across the entire AI adoption-to-innovation-to-competitiveness chain. The authors develop the TEC-AIP Framework and recommend that managers treat AI as a catalyst for genuine innovation rather than a symbolic upgrade, while policymakers design coordinated support systems combining infrastructure, skills, financing, and ecosystem confidence.
- Enterprise
- AI policy