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
Implementation readiness of artificial intelligence in in vitro fertilization: a multi-framework appraisal with special reference to advanced reproductive age patients
Chia Lin Chang
Frontiers in Endocrinology · 2026-09-07
This paper evaluates nine AI applications across the IVF clinical workflow using four implementation frameworks and evidence grading, organizing them into three deployment waves based on readiness. Wave 1 technologies—outcome prediction, stimulation dosing, and RFID quality tracking—are immediately deployable as decision-support tools, while embryo selection, despite high visibility, is not yet the most deployment-ready. A key finding is that AI embryo-selection discriminatory performance (AUC) rises substantially with maternal age, from 0.596 in patients under 35 to 0.768 in those 43 and older, making advanced reproductive age patients the group most likely to benefit. The authors conclude that success depends not on algorithmic sophistication but on rigorous validation, regulatory harmonization, ethical governance, and workforce and IT infrastructure investment.
- Quality assurance
- Certifications
- AI policy
Research
Public Responses to Disaster Warnings from Government Versus AI: Evidence from Behavioral Experiments
Lin Lei, Jing Tan, Di Zheng
International Journal of Disaster Risk Science · 2026-09-07
Using two online behavioral experiments with 599 residents in China, this study examines how people respond to disaster warnings issued by government authorities versus AI systems. Results show government warnings generate higher trust than AI warnings, while consistent messaging from both sources boosts trust and protective intentions further. When government and AI warnings conflict, anticipated regret drives people to follow the higher-level warning. The findings offer practical guidance for designing multi-source disaster warning systems that blend institutional credibility with AI capabilities.
- AI policy
Research
A study of the employment challenges and countermeasures for broadcasting and hosting students in Guizhou higher vocational colleges in the era of digital intelligence
Qifu Wan, Zhanyu Wang, Qirui Chen
Journal of Education and Educational Policy Studies · 2026-09-07
This qualitative study examines how digital and AI-driven media transformation is creating employment challenges for broadcasting and hosting graduates at higher vocational colleges in Guizhou, China. Drawing on Social Cognitive Career Theory, the researchers conducted semi-structured interviews with 12 graduating students and analyzed the data using NVivo thematic coding. The findings identify three interacting factors behind students' employment difficulties: self-efficacy conflicts, lagging educational provision, and rapid changes in the media industry ecosystem. The paper recommends structural reforms in education, competency rebuilding, closer industry alignment, and stronger policy coordination to address these challenges.
- Workforce
- AI policy
Research
Desafíos para la responsabilidad civil a propósito del riesgo y de la autonomía
Marcos Santos Vaquero
Ius et Praxis · 2026-09-07
This paper examines how the European Union is structuring civil liability for damages caused by AI systems, using risk level as the organizing principle. It analyzes how different degrees of AI risk affect legal regimes, burden of proof, damage traceability, and complementary protections like compulsory insurance and compensation funds. The study finds that EU regulation is moving toward a differentiated liability model that seeks to balance victim protection, legal certainty, and technological innovation.
- AI policy
Research
A dynamic recommendation algorithm for regional industry-adapted higher vocational courses integrating counterfactual causal inference and graph neural networks
Xianglong Xiao, Dongmei Xia
Scientific Reports · 2026-09-07
CCIG-DRec is a dynamic course recommendation system for higher vocational education that combines heterogeneous graph neural networks with counterfactual causal inference to better align curricula with regional industry skill demands. Built on a time-evolving graph of students, courses, occupations, skills, and industries drawn from the Chengdu–Chongqing Economic Circle, it uses causal intervention to remove popularity bias and generates counterfactual student trajectories for both training and explainability. Against the strongest baseline, it achieves a 12.7% improvement in Recall@10 and 11.8% in NDCG@10, while narrowing a counterfactual fairness gap by 30.4%. The system's auditable interventional justifications are presented as suitable for program-level curricular governance, with implications for how vocational institutions can adapt course offerings to shifting labor market needs.
- AI policy
- Workforce
Research
THE STATUS OF ARTIFICIAL INTELLIGENCE IN THE DEVELOPER COMMUNITY: A LITERATURE REVIEW OF STATISTICS, TRENDS, AND EXPECTATIONS CURRENT ADOPTION PATTERNS, EMERGING TRENDS, AND FUTURE OUTLOOK (2023–2026)
S. Colafranceschi
ShodhAI Journal of Artificial Intelligence · 2026-09-07
This literature review synthesizes evidence from four large industry surveys and multiple randomized controlled trials to assess AI adoption among software developers from 2023 to 2026. It finds that 84% of developers now use or plan to use AI tools and 51% use them daily, yet trust in AI accuracy has fallen to 29–33%, creating an 'adoption–trust paradox.' Critically, the most rigorous field experiment found AI access slowed experienced developers by 19% on real maintenance tasks, contrasting with earlier studies showing speed gains, while additional evidence links AI-assisted coding to reduced security, higher code churn, and skill deficits in junior programmers. The review concludes that AI's net effect on software quality, security, and professional skill development remains contested and depends heavily on task complexity, codebase familiarity, and developer experience.
- Workforce
- Quality assurance
Research
The Governance-Experience Gap in Generative AI: A Cross-Level Analysis of Regulatory Frameworks and Interactional Frictions
Hae Sun Jung, Haein Lee
International Journal of Human-Computer Interaction · 2026-09-07
This study investigates whether major AI governance frameworks—including the EU AI Act, NIST AI Risk Management Framework, and OECD Principles—align with the real-world problems end-users encounter when interacting with generative AI applications. By analyzing mobile app reviews using transformer-based topic modeling and comparing findings to governance dimensions derived through semantic harmonization, the researchers found a significant 'governance-experience gap': only 36.6% of user-reported issues corresponded to formal governance categories, while 61.0% reflected concerns not captured by existing frameworks. The findings suggest that current system-centric governance approaches may miss a large share of user-experienced friction, and that AI policy could benefit from incorporating user-centered quality metrics alongside traditional risk management approaches.
- AI policy
- Quality assurance
Research
Beyond resistance: an integrated psychological organizational model of employee response to digital transformation
Junyan Zhao, Khadija Tahir, Ahmed Muneeb Mehta et al.
Scientific Reports · 2026-09-07
This study develops an integrated psychological-organizational model to explain why employees resist digital transformation initiatives. Drawing on Innovation Resistance Theory and Social Cognitive Theory, the authors find that technology anxiety directly increases resistance, while perceived risk and change fatigue operate indirectly by reducing digital competence, which in turn drives resistance. Perceived organizational support moderates these pathways, meaning stronger organizational backing can buffer against resistance. The findings, validated via PLS-SEM on quantitative survey data, offer practical guidance for organizations on how to manage employee resistance and improve digital transformation outcomes.
- Workforce
- Enterprise
Research
From Risk Recognition to Regulatory Proof: The Missing Longitudinal Assurance Layer for AI Companions Across Five Jurisdictions
AREG MARTIROSYAN
Zenodo (CERN European Organization for Nuclear Research) · 2026-09-07
This article compares AI-companion regulation across five jurisdictions—the US, China, EU, UK, and Australia—and identifies a critical 'longitudinal assurance gap': while regulators increasingly recognize harms from sustained human–AI relationships (such as emotional dependency and distorted social norms), no publicly adopted standardized method yet connects changes in user autonomy or social engagement to specific system behaviors and legal duties in a form suitable for independent audit. The paper distinguishes between risk recognition (which regulators are advancing) and regulatory proof (which requires validated human constructs, behavioral traces, causal inference, and jurisdiction-specific legal interpretation working together). It sets out minimum requirements any credible longitudinal assurance framework would need to meet, calling for cooperation among regulators, behavioral scientists, auditors, and developers to close this gap.
- AI policy
- Certifications
Research
Influence of Technology on Labor Market: A Systematic Review
Smridhi Vohra
International Journal for Research in Applied Science and Engineering Technology · 2026-09-07
This systematic review examines how automation, AI, digital platforms, and advanced communication technologies reshape labor markets across industries. The study identifies three core mechanisms: displacement of routine tasks, productivity-enhancing complementarities that create new work, and growth of platform-based and remote employment. While technology can drive job polarization and wage inequality, it also opens pathways for skill development and flexible work arrangements, with institutional factors like education and labor regulation mediating outcomes. The authors call for adaptive policy strategies to promote inclusive growth and cushion disruptions from technological change.
- Workforce
- AI policy
Research
Revisiting judging reliability in taekwondo freestyle Poomsae: implications for AI-supported evaluation
Min-woo Jeon, Kim Hong-Suk, Park Ji-Yong et al.
Frontiers in Psychology · 2026-09-07
This study examined the reliability of human judging in Taekwondo freestyle Poomsae to identify which performance components can be scored consistently before AI-based evaluation systems are introduced. Ten internationally certified referees scored ten competition videos twice, one week apart, revealing systematic score inflation across sessions and very low inter-rater agreement—single-rating absolute-agreement ICCs for the total score were near zero or negative. The authors conclude that evaluation components with poor inter-rater reliability require clearer operational definitions and further validation before computational or AI-assisted scoring is applied. The findings directly inform the prerequisites for developing trustworthy AI judging systems in competitive sports.
- Quality assurance
- Certifications
Research
Artificial Intelligence in Financial Reporting Fraud Detection: An Empirical Investigation of Auditor Trust as a Mediating Mechanism in U.S. Markets
Mohammad, Zahurul, Abu, Faysal, Ataur, Mohammad, Samirul Mia, Islam, Sayed, Ahmed, Rahman, Aziz, Islam
Journal of Applied Finance and Banking · 2026-09-07
This study investigates how auditor trust mediates the relationship between AI system characteristics and fraud detection effectiveness in U.S. financial reporting contexts. Using survey data from 450 auditing professionals analyzed via PLS-SEM, the researchers found that predictive capability, explainability, data quality, and governance all significantly predicted auditor trust in AI (R²=.697), which in turn strongly predicted perceived fraud detection effectiveness (R²=.524; β=.724). The results demonstrate that AI-enabled fraud detection is a socio-technical outcome dependent on explainability, data integrity, governance structures, and user trust—not algorithmic accuracy alone. The study offers practical guidance for audit firms and regulators on designing trustworthy AI systems for financial oversight.
- Enterprise
- Quality assurance
Research
Agro-Algorithmic Stockpiles, Food Security and the CAP: Europe’s New Governance of Uncertainty in Agricultural Markets
Antonioluigi Costato
European Journal of Risk Regulation · 2026-09-07
This article introduces the concept of 'agro-algorithmic stockpiles' to describe how EU agricultural market stabilization is shifting from managing physical food reserves to governing data-driven predictive systems. The author argues that while the EU Data Act creates a legal framework for agricultural data access, the exclusion of agrifood systems from the AI Act's high-risk classification leaves a regulatory gap exposing markets to systemic risks such as algorithmic monocultures and pro-cyclical dynamics. The paper concludes that food security should be treated as critical infrastructure and that this gap represents a key test case for the EU's risk-based AI regulation and the future role of the Common Agricultural Policy in the digital era.
- AI policy
Research
QCR Composite Ratio: A Companion Metric to the Quantum Confidence Rating Standard
Ilyes Tarik MAZARI
Zenodo (CERN European Organization for Nuclear Research) · 2026-09-07
This paper introduces the Composite Ratio (QCR-C), a companion metric to the Quantum Confidence Rating (QCR) standard for evaluating AI systems' cryptographic compliance with post-quantum standards. The QCR standard scores primitives verified through composite constructions (classical plus post-quantum algorithms) identically to those using pure post-quantum constructions, obscuring how conservatively an organization migrated to post-quantum cryptography. QCR-C addresses this gap by reporting what proportion of verified primitives rely on composite versus pure post-quantum constructions, giving regulators, auditors, and insurers clearer visibility into migration conservatism without altering existing QCR scores or certifications.
- Certifications
- AI policy
Research
From Risk Recognition to Regulatory Proof: The Missing Longitudinal Assurance Layer for AI Companions Across Five Jurisdictions
AREG MARTIROSYAN
Zenodo (CERN European Organization for Nuclear Research) · 2026-09-07
This article examines how five major jurisdictions—the United States, China, the European Union, the United Kingdom, and Australia—regulate AI companion systems and finds a significant 'longitudinal assurance gap': while regulators increasingly recognize risks such as emotional dependency and harm to real-world relationships, no publicly adopted standardized method exists to connect changes in user autonomy or social engagement to specific system behaviors in a form suitable for independent audit. The paper distinguishes between risk recognition (which regulators are achieving) and regulatory proof (which requires validated human constructs, behavioral traces, causal inference, and jurisdiction-specific legal interpretation working together). The authors argue that closing this gap demands cooperation among regulators, behavioral scientists, auditors, and AI developers to build credible, reproducible evidentiary pathways. This matters because AI-companion regulation is outpacing the assurance methods needed to verify compliance and protect users over time.
- AI policy
- Certifications
Research
How rejection letters from AI recruiters shape recruiter's prestige and experienced respect
Kyriaki Fousiani, Pieter A. Minnigh, Brian Lewis
Frontiers in Organizational Psychology · 2026-09-07
This paper investigates how AI-generated rejection letters—either considerate or typified—affect job applicants' perceptions of the AI recruiter's prestige and their own sense of being respected. Across two experiments, considerate rejection letters reduced mechanistic dehumanization compared to typified ones, which in turn positively influenced perceived recruiter prestige and experienced respect. Unemployment status and attitudes toward unemployment moderated some effects, suggesting vulnerable applicants may be especially sensitive to how AI systems communicate unfavorable outcomes. The findings highlight that the tone and quality of AI-driven rejection communications have meaningful psychological consequences for applicants.
- Workforce
- Enterprise
Research
A large language model-enhanced knowledge graph framework for text-implied public health policy gap screening: digital health executability, behavioral accessibility, and service-support coverage
Xinyi Wang, Jiao Lu
Frontiers in Public Health · 2026-09-07
This paper presents PRP-KG, a large language model-enhanced knowledge graph framework designed to screen public health policy documents for structural gaps in implementation, accessibility, and service-support coverage. The system segments policy text into clauses, extracts entities and relations via a predefined schema, and organizes them into a three-layer knowledge graph with a consistency feedback mechanism to catch missing or inconsistent elements. Experiments on annotated public health policy benchmarks show PRP-KG achieves lower prediction error and more accurate identification of high-priority review signals than most comparison methods. The framework produces evidence-linked outputs that human policy reviewers can inspect, making it a practical tool for systematic policy gap analysis.
- AI policy
- Quality assurance
Research
QCR Composite Ratio: A Companion Metric to the Quantum Confidence Rating Standard
Ilyes Tarik MAZARI
Zenodo (CERN European Organization for Nuclear Research) · 2026-09-07
This paper introduces the QCR Composite Ratio (QCR-C), a companion metric to the Quantum Confidence Rating standard for AI systems. While the existing QCR standard scores cryptographic primitives verified against post-quantum standards without distinguishing between composite constructions (classical plus post-quantum algorithms combined) and pure post-quantum constructions, QCR-C fills this gap by reporting what proportion of verified primitives rely on composite rather than pure post-quantum approaches. The metric is designed to give regulators, auditors, and insurers clearer visibility into how conservatively an organization has approached its post-quantum migration, without modifying the underlying QCR formula, tier structure, or any previously issued certifications.
- Certifications
- AI policy
Research
Artificial intelligence adoption among Swiss adult educators: usage patterns, determinants, barriers, and perceptions
Attila Güler
Frontiers in Education · 2026-09-07
A cross-sectional survey of 123 Swiss adult educators found that 82.1% already use AI tools, with adoption frequency driven primarily by self-reported AI knowledge rather than demographics. Educators are broadly optimistic but cite privacy and security concerns and limited institutional support as the main barriers. Current use is concentrated in efficiency tasks like lesson preparation rather than deeper pedagogical integration, and educators worry about impacts on learners' critical thinking. The findings point to a sector with significant professional development needs and institutional support gaps that constrain more meaningful AI adoption.
- Workforce
Research
Human-Centered AI for Shared Prosperity: A Multi-Stakeholder Framework Bridging Innovation, Ethics, and Inclusion in Southeast Asia
Ahmad Rizki Apriansyah, Faqih Wildan Hakim, Ridwansyah
Proceeding of International Conference on Digital Social and Science · 2026-09-07
This paper develops a multi-stakeholder, human-centered AI governance framework tailored to Southeast Asia, drawing on interdisciplinary review and comparative policy analysis across Indonesia, Malaysia, Singapore, Thailand, and Vietnam. The authors identify three key governance deficits in the region: fragmented institutional coordination, reliance on non-binding guidelines without enforcement mechanisms, and insufficient civil society participation. The proposed framework rests on four pillars—rights-based regulation with AI impact assessments, structured multi-stakeholder engagement, localization of global best practices, and capacity building—framed as enablers of sustainable development rather than barriers to innovation. The study positions itself as the first comprehensive human-centered AI governance framework specifically designed for the Southeast Asian context, offering actionable guidance for policymakers balancing technological advancement, human rights, and equitable development.
- AI policy
Research
Can Artificial Intelligence Enhance Corporate Green Productivity? Evidence from Chinese Listed Firms
Yunji Zhang, Yang Yi, Zipan Cai
Sustainability · 2026-09-07
Using 33,017 firm-year observations from 4,079 Chinese A-share listed firms (2015–2024), this study finds that AI adoption is significantly and positively associated with corporate green total factor productivity (GTFP), a measure that accounts for undesirable environmental outputs. R&D intensity is identified as a key transmission channel, and the effect is strongest among firms with tighter financing constraints, non-polluting industries, and non-state-owned enterprises. The findings suggest AI-enabled innovation can advance both environmental efficiency and firm value in emerging economies.
- Enterprise
Research
Generative AI usage and employee innovation performance: the chain mediation mechanism of resource acquisition, integration, and activation
Da Teng, Yaoyao Shen, Ruixue Yuan et al.
Frontiers in Psychology · 2026-09-07
A survey study of 612 knowledge workers finds that generative AI use is positively associated with employee innovation performance, with 72.9% of that effect explained by a three-stage cognitive chain: resource acquisition (cognitive divergence), resource integration (cognitive elaboration), and resource activation (challenge appraisal). Work stress moderates this pathway in a stage-dependent way, strengthening mediation at the integration and activation stages but not at the acquisition stage. The findings suggest organizations can improve innovation outcomes by targeting specific cognitive stages rather than focusing solely on AI tool adoption.
- Workforce
- Enterprise
Research
SerenAI: State-transition system inspired by text-based world AI models
Elvin Babayev, Artem Sinitsa, Arash Hajisharifi et al.
arXiv (Cornell University) · 2026-09-06
SerenAI is a state-transition system built on large language models that outputs structured, verifiable predictions—including causal deltas, next states, validity rewards, and termination signals—rather than unconstrained free text. The system uses parameter-efficient fine-tuning followed by verifier-based reinforcement learning on 50,000 cause-and-effect examples across 12 environments, substantially improving structured output metrics such as schema validity (55% to 84%) and reward exact match (1% to 80.5%) over an 8B baseline. The authors frame this as a step toward making LLM outputs auditable in legal, operational, and financial workflows, while explicitly noting it does not yet meet legal-grade reliability. The paper also outlines a validation protocol covering human oversight, calibration, and on-premise deployment for regulated use cases.
- Enterprise
- Quality assurance
Research
A Unified Policy Architecture (UPA): The Governance Kernel for Enterprise AI Operating Systems
Prabhu Raghav, Balamurugan Pandi, Arul Vivek et al.
arXiv (Cornell University) · 2026-09-06
This paper introduces the Unified Policy Architecture (UPA), a governance framework designed to address fragmented authorization, security, and compliance mechanisms in enterprise AI systems where autonomous agents plan, reason, use tools, and collaborate. UPA provides a unified policy model covering AI agents, tools, workflows, memory, enterprise resources, and agent-to-agent interactions, extending beyond simple authorization to include runtime obligations, human approvals, audit evidence, and compliance evaluation. The framework includes a declarative policy language, extensible plugins, industry policy packs, and an evaluation framework tailored to enterprise governance needs. It matters because enterprises deploying autonomous AI lack a coherent governance architecture, and UPA aims to provide a foundation for secure, accountable, and auditable AI operating systems.
- Enterprise
- AI policy
Research
Agentic Artificial Intelligence in Business Decision-Making: A Framework for Human–AI Collaborative Governance and Strategic Value Creation
P V. Amutha, M. Bhuvaneswari
International Journal of Research Publication and Reviews · 2026-09-06
This paper develops and tests the Human–AI Collaborative Decision Governance (HACDG) model using a mixed-methods study of 168 managers and 14 senior executives across four sectors. Findings show that AI involvement reliably improves decision speed, but only improves perceived decision quality when paired with moderate-to-high governance maturity; without it, heavy AI reliance correlates with lower confidence and higher post-decision regret. The framework helps organizations calibrate the degree of AI autonomy against their governance capacity, with direct implications for management education, internal audit, and enterprise risk functions.
- Enterprise
- AI policy