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
VERA-8B: Evidence-Grounded Audit Risk Reasoning from SEC Filings
Menghan Liu, Elynn Chen
arXiv (Cornell University) · 2026-08-28
VERA-8B is an end-to-end AI reasoning system designed to identify audit risks from SEC filings before enforcement actions occur. The system combines supervised fine-tuning (SFT) and group relative policy optimization (GRPO) under a unified evidence standard, and introduces abstention and uncertainty qualification to avoid unsupported claims. An AuditBridge component transforms raw filings into reviewer-ready reports. The authors report that VERA-8B surpasses all evaluated baselines, representing the first machine learning approach targeting pre-enforcement audit prediction with evidence-grounded outputs.
- Quality assurance
- Enterprise
Research
Sustainability of Open-Source Machine Learning Robustness Assessment Tools: A Repository Mining Study
Joshua Owotogbe, Indika Kumara, Willem-Jan van den Heuvel et al.
arXiv (Cornell University) · 2026-08-28
This empirical study mines 28 open-source machine-learning robustness-assessment repositories on GitHub to evaluate their maintenance health, community engagement, and longevity. The authors find that sustained activity is heavily concentrated in a small minority of tools: as of January 2026, only 5 of 28 repositories were classified as active, 22 were inactive, and 1 was archived. The findings matter for AI assurance and quality-assurance practices because practitioners who rely on unmaintained robustness tools risk reproducibility failures and weakened evidence for safety claims. The paper calls for treating robustness tools as evolving software systems requiring ongoing stewardship.
- Quality assurance
- Certifications
Research
CURA: Certified Runtime Alarms for Computer-Use Agents
Divake Kumar, Sina Tayebati, Devashri Naik et al.
arXiv (Cornell University) · 2026-08-28
CURA (Certified Runtime Alarms for Computer-Use Agents) is an external monitoring system that detects when AI computer-use agents are failing at tasks without admitting it. The paper shows that a capable agent pipeline achieves 82.9 mean task score on OSWorld benchmarks but self-reports success in 90% of its failures, demonstrating that self-report is a deeply unreliable oversight mechanism. CURA addresses this by analyzing only externally visible telemetry—no model internals or extra LLM calls—and applies sequential statistical testing with certified false-alarm control, detecting 42.3% of failures a median of 31 steps before termination. When used to gate mid-execution human or AI oversight, CURA improves overall task completion from 82.9 to 86.8 mean score, recovering 23 of 70 failures, making it a practical tool for deployers who need reliable failure detection in agentic AI systems.
- Enterprise
- Quality assurance
Research
SENTINEL: A cascade-aware temporal framework with formally verified alert logic for early ICU mortality prediction
V. Sindhuja, S. Venkatraman
Array · 2026-08-28
SENTINEL is a temporal framework for early ICU mortality prediction that combines a cascade deterioration metric, gradient-boosted ensemble, and formally verified alert logic encoded as a timed automaton with guaranteed bounded latency in critical states. Evaluated on over 256,000 ICU stays across 209 hospitals, it achieves an AUROC of 0.864 internally and 0.801 externally, outperforming XGBoost (0.778), with an 84.4% catch rate and a median lead time of 46.2 hours. The authors explicitly note that without a formal landmark analysis the system cannot be validated as a deterioration detector distinct from a mortality predictor, and prospective multicenter trials are required before clinical deployment. The work matters for quality assurance because it introduces formal verification of alert logic and clearly scopes the safety guarantees, illustrating both the promise and current limitations of AI-driven clinical decision support.
- Quality assurance
- Enterprise
Research
From Intelligent Application to Green Performance: How Artificial Intelligence Reshapes the Carbon Emission Pathways of Energy Enterprises
Xuelong Zhang, Xiaoling He, Mei Li et al.
Sustainability · 2026-08-28
Using panel data from listed Chinese energy firms (2010–2023) and fixed-effect models, this study finds that AI adoption significantly reduces carbon emission intensity in the energy sector. The effect is amplified by executives with digital backgrounds and by low-carbon city pilot policies, and operates through mechanisms including higher capacity utilization, improved supply-chain efficiency, and greater production efficiency. Impacts are stronger in eastern and central China and among state-owned and conventional energy firms. The findings offer firm-level evidence that AI can accelerate the low-carbon transition and inform both policy design and corporate strategy.
- Enterprise
- AI policy
Research
ARTIFICIAL INTELLIGENCE, HUMAN RESOURCE CAPABILITY AND SUPPLY CHAIN AGILITY FOR RICE SUPPLY CHAIN RESILIENCE
Sonya Mamoriska Mulia Harahap, Nur Damayanti, Dalili Izni Shafie
Jurnal Teknologi Industri Pertanian · 2026-08-28
This study surveyed 210 logistics service providers in Indonesia's rice supply chain to examine how human resource (HR) capabilities and AI adoption affect supply chain resilience (SCR), with supply chain agility (SCA) as a mediating factor. Results show both HR capabilities and AI adoption positively influence SCA, which in turn significantly enhances SCR, with SCA fully mediating both relationships. The findings demonstrate that combining workforce development with AI-enabled digital transformation builds the agility needed to handle seasonal harvest fluctuations and distribution variability in rice logistics. The study extends Resource-Based View and Dynamic Capabilities Theory to rice supply chain logistics in an Indonesian context.
- Workforce
- Enterprise
Research
A Study on the Employment Status of Vocational College Graduates in the Era of Artificial Intelligence
Lijun Deng, Xue Huang
Scientific Journal Of Humanities and Social Sciences · 2026-08-28
This study surveyed 298 vocational college graduates across five institutions to examine how AI literacy, practical skills, and adaptability relate to employability in an AI-driven labor market. Using descriptive statistics, correlation analysis, and multiple regression, the researchers found that all three factors were positively associated with employability, with practical skills showing the strongest effect. The findings suggest that vocational training programs need to evolve alongside AI-driven changes in workplace requirements to better prepare graduates for shifting labor market demands.
- Workforce
Research
Artificial Professionalism: An Evaluation of Six Large Language Models on the UK Multi-Specialty Recruitment Assessment Professional Dilemmas Paper
Mohammad Jakir Ahmed, Munsif M Mansoor, Hamza Mahmood et al.
Cureus · 2026-08-28
This study evaluated six publicly available large language models (ChatGPT, Microsoft Copilot, Claude, Gemini, DeepSeek, and Grok) on the professional dilemmas section of the UK Multi-Specialty Recruitment Assessment, a situational judgement test used to shortlist postgraduate medical training candidates. All six models scored between 59% and 71%, with no statistically significant differences among them, and all performed substantially better on multiple-choice items (80.3%) than on ranking items (57.2%). The findings suggest that current LLMs provide only moderate and uneven performance on this high-stakes medical assessment, meaning candidates who rely on them for exam preparation risk receiving unreliable guidance, particularly for ranking-style questions.
- Certifications
- Workforce
Research
Artificial intelligence-supported therapeutic interventions for autism spectrum disorder: a systematic review
Julia Kuca, Magdalena Stencel, Błażej Pilarski et al.
Frontiers in Psychiatry · 2026-08-28
This systematic review examined 14 studies of AI-supported interventions (including robotics, virtual reality, and wearables) for autism spectrum disorder, finding that these tools show promise as social mediators that improve social-emotional outcomes by reducing cognitive load through real-time task adaptation and behavioral monitoring. However, the authors found significant heterogeneity in intervention dosage, predominantly small and male-dominated samples (most with N < 20), and a complete absence of adult participants, limiting generalizability. The review concludes that more RCTs and broader demographic representation—particularly for adult and female ASD phenotypes—are needed before AI-based interventions can be reliably implemented in clinical care.
- Quality assurance
- AI policy
Research
Postgraduate Education in Artificial Intelligence: A Proposal from the Artificial Intelligence Commission of the Portuguese Medical Association
João Frutuoso, Ana María, Helena Donato et al.
Acta Médica Portuguesa · 2026-08-28
This state-of-the-art review from the Portuguese Medical Association's AI Commission proposes a three-tier postgraduate medical curriculum for AI competencies, spanning baseline literacy for all physicians, proficient skills for those involved in AI appraisal and implementation, and advanced competencies for clinician-scientists and governance leaders. The proposal identifies six core domains—foundational AI literacy, critical appraisal, safe clinical application, ethics/law/governance, patient communication, and health data literacy—drawing on literature published since 2022. It is explicitly framed around the European AI Act's February 2025 requirement that AI deployers ensure adequate AI literacy among staff operating such systems. The authors present this as a conceptual framework intended for multidisciplinary consensus validation within the Portuguese postgraduate training context.
- Certifications
- AI policy
Research
Artificial Intelligence for Automated Semen Analysis from Microscopy to Clinical Translation
Shimaa B. bahy, Mostafa Ahmed, Ahmed A. Elngar
Journal of Smart Algorithms and Applications (JSAA) · 2026-08-28
This review paper synthesizes AI and deep-learning methods applied to automated semen analysis, covering sperm detection, segmentation, morphology classification, motility assessment, and hybrid computational pipelines. While many studies report strong performance on curated internal datasets, the review finds that external validation, calibration, uncertainty estimation, and prospective clinical utility assessments remain uncommon, limiting readiness for routine clinical deployment. The authors conclude that translation to clinical practice will require harmonized multicenter datasets, consensus-based annotation, leakage-aware validation, standardized reporting, and regulatory planning appropriate to the intended jurisdiction.
- Quality assurance
- Certifications
Research
Software quality assurance in the era of Agentic AI: a systematic mapping study
Randa Ouaarous, Imane Hilal, Abdellatif Mezrioui
Frontiers in Computer Science · 2026-08-28
This systematic mapping study analyzes how Agentic AI—characterized by autonomous decision-making and adaptive behavior—is being integrated into Software Quality Assurance (SQA). Reviewing 37 primary studies from five digital libraries, the authors find that applications cluster heavily around test generation and defect management, while process planning and managerial roles receive far less coverage. Most agents show moderate autonomy with goal-oriented or diagnostic behaviors, delivering benefits in efficiency, accuracy, and maintainability, but facing limitations in scalability, generalizability, and transparency. The study identifies trustworthiness, validation, and leadership support as priority research gaps for building process-aware Agentic SQA systems.
- Quality assurance
Research
Comparative analysis of artificial intelligence-enabled medical devices in the state registries of the USA, China, and Russia
D. I. Korabelnikov, A. I. Lamotkin
FARMAKOEKONOMIKA Modern Pharmacoeconomics and Pharmacoepidemiology · 2026-08-28
This comparative study analyzes AI-enabled medical device (AIeMD) registries across the USA (FDA), China (NMPA), and Russia (Roszdravnadzor), finding that the FDA registry leads with 1,524 AIeMDs, roughly 10 times more than China's 154 and 27 times more than Russia's 57. Across all three countries, medical image analysis—especially radiology—dominates AIeMD approvals, accounting for 76–84% of registrations, though the FDA registry has begun diversifying toward cardiology and neurology since 2022. The study also highlights structural differences in manufacturers: the USA is led by multinationals with 57.7% foreign developers, China by specialized national AI firms concentrated in four cities, and Russia almost entirely by domestic software developers (91.2%) centered in Moscow. The findings are relevant for health technology policy and regulatory benchmarking of AI-based medical devices internationally.
- AI policy
- Certifications
Research
Embracing Large Language Models for Medical Applications, Part II: Building a Framework for Clinical Stewardship
Shiv Patil, Mert Karabacak, Matthew Southerby et al.
Cureus · 2026-08-28
This viewpoint paper proposes a practical framework for the responsible clinical adoption of large language models (LLMs) in medicine, arguing that demonstrated model performance alone does not guarantee improved physician reasoning, safer decisions, or better patient outcomes. The authors distinguish lower-risk administrative and communication tasks from higher-risk diagnostic, triage, and treatment tasks, and call for tiered evidence standards, local validation, human oversight, equity auditing, patient-centered consent, and continuous post-deployment surveillance. LLMs are framed as workflow-embedded sociotechnical systems whose real-world performance depends on the interaction among models, clinicians, patients, interfaces, and governance structures. The paper contends that realizing the promise of medical LLMs requires a shift from fascination with fluent outputs toward disciplined stewardship of clinical performance, accountability, equity, and trust.
- AI policy
- Quality assurance
Research
Worker Power for the AI Economy: Portable Benefits and the Future of Employment Support
Joseph Evans
IPPR Progressive Review · 2026-08-28
This paper argues that AI-driven disruption to the UK labour market demands a new employment support framework, as the current system is ill-equipped for rapid, unpredictable changes in labour demand. The author proposes 'portable benefits'—flexible, worker-held entitlements covering training credits, income security, and legal support—modelled on the French Compte Personnel de Formation and emerging US pilots. The proposal is paired with stronger employer obligations to train and redeploy workers, and reinforced cooperative labour market institutions. The paper is relevant to workforce and policy impacts of AI adoption.
- Workforce
- AI policy
Research
Technological Polarization and Unequal Growth in the Era of Generative Artificial Intelligence
Kyra Mahindru
International Journal of Computer Information Systems and Industrial Management Applications · 2026-08-28
This mixed-methods study examines how Generative AI and Agentic AI reshape labor markets, finding that broader AI adoption increases wage inequality, disproportionately disrupts cognitive work, and polarizes the labor market while boosting productivity and wages for AI-skilled workers. Drawing on a PRISMA-based review of 94 articles and qualitative case studies of Universal Basic Income, data dividend laws, and Industry 5.0 implementations, the authors show that redistribution policies and human-centric governance can mitigate these harms. The paper proposes the EQUATE framework—grounded in Rogers' Innovation Diffusion Theory—covering Equity, Quality Augmentation, UBI Policy Calibration, Agentic Alignment, Technology Access, and Ethical Governance as a roadmap for balancing innovation with equitable economic development. The findings are directly relevant to workforce policy and the governance of AI-driven labor disruption.
- Workforce
- AI policy
Research
Beyond Human-in-the-Loop: Designing Accountable Organizations for AI-Augmented Decision-Making — The HILO-A Framework
Ignacio Bonasa Alzuria
arXiv · 2026-08-28
This paper introduces the Human-in-the-Loop Organizational Architecture (HILO-A), a management framework for allocating decision rights and accountability in organizations that use AI systems. It defines five layers (Data, Algorithm, Human Judgment, Decision, and Accountability) and four human-AI interaction modes ranging from assisted intelligence to autonomous execution, complemented by a Decision Criticality Matrix assessing impact, reversibility, uncertainty, and ethical sensitivity. The framework is explicitly positioned alongside the NIST AI Risk Management Framework, OECD AI Principles, and EU AI Act, arguing that mature AI organizations will be defined not by maximum automation but by how clearly they assign responsibility across hybrid human-AI decision chains. It offers managers a practical vocabulary and implementation protocol for making AI governance explicit, auditable, and adaptable.
- Enterprise
- AI policy
Research
Unpacking English Language Teachers' Professional Learning Needs in the Age of AI : A Descriptive Qualitative Analysis of Perceptions, Expectations and Suggested Supports
Yuanyuan Zhang, Ali Derakhshan, Yujong Park
European Journal of Education · 2026-08-28
This qualitative study surveyed 20 experienced EFL teachers to understand how AI technologies are reshaping their professional learning needs. Using inductive thematic analysis and the METUX model, the researchers found that teachers need support around AI understanding, curriculum integration, ethical implications, pedagogical alignment, and assessment clarity. Teachers expected training that goes beyond basics and called for workshops, collaborative discussions, and sustained institutional support. The findings offer practical guidance for teacher educators designing AI-focused professional development programs.
- Workforce
Research
Adapting to generative AI in creative work: a technological frames perspective on creative advertising
Georg von Richthofen, Sonja Köhne, Maja Golf Papez
Journal of Business Research · 2026-08-28
Drawing on three years of netnographic research in online advertising communities and interviews with advertising creatives, this study finds that workers hold competing interpretive frames about generative AI—concerning its agency, job impact, and value creation—that produce distinct adaptive responses: AI skilling, reskilling, and 'deep skilling' (deliberately cultivating human capabilities GenAI cannot replicate). The research shows that how creatives interpret GenAI shapes whether and how they update their skills, with implications for managers navigating GenAI adoption in creative contexts.
- Workforce
- Enterprise
Research
A responsible artificial intelligence framework for translational readiness of machine learning biomarker models in clinical decision support
Vahid Habibzadehomran
Discover Artificial Intelligence · 2026-08-28
This structured narrative review synthesizes evidence from 70 publications to propose a five-domain framework—covering methodological, clinical, operational, governance, and post-deployment readiness—for translating machine learning biomarker models into clinical decision-support systems. The authors identify recurring gaps in external validation, calibration, feature stability, subgroup assessment, and monitoring for data and performance drift, and distinguish predictive performance from broader implementation readiness. The framework highlights modality-specific risks such as assay variability in molecular models and missingness patterns in electronic health record models. It is offered as a structured evaluation aid, with prospective testing needed to confirm whether it improves governance decisions and patient safety.
- Quality assurance
- Certifications
- AI policy
Research
Governing synthetic biology and artificial intelligence (AI) convergence: emerging biosecurity priorities for Africa
Keletso Masisi, Marie Atsama Amougou, Ilodia Amilcar Zacarias et al.
Frontiers in Bioengineering and Biotechnology · 2026-08-28
This paper examines how the convergence of synthetic biology and AI is creating governance gaps in African biosecurity systems. Through analysis of African biosafety regulatory bodies and national frameworks, the authors identify 'structural exposure points'—systemic conditions where emerging technological capabilities outpace oversight—including fragmented institutional mandates, uneven regulatory capacity, and limited AI literacy in biosafety systems. The paper proposes context-sensitive governance priorities such as embedding AI considerations into existing biosafety review processes, risk-based oversight, and DNA synthesis screening baselines, advancing an adaptive governance model focused on institutional integration rather than regulatory expansion. The findings matter because they offer a practical pathway for African institutions to govern bio-digital convergence while supporting responsible scientific innovation.
- AI policy
Research
AI-driven talent acquisition in Indian IT firms: A qualitative study of automation, impacts, and skill requirements
Itam Urmila Jagadeeswari, Nidhi Shukla, Mercy Toni et al.
SN Business & Economics · 2026-08-28
This qualitative study examines how AI is reshaping talent acquisition in Indian IT firms, drawing on semi-structured interviews with 14 HR and IT professionals across four companies. Findings show that tools such as automated resume screening, chatbots, and predictive analytics are accelerating and scaling hiring processes, while also surfacing concerns around data privacy, system integration, and resistance to change. The study highlights that HR professionals need to upskill in data literacy and AI management, and that the organizational changes from AI adoption vary depending on a firm's stage of digital HR maturity. Results offer practical guidance for HR managers seeking to balance automation benefits with ethical safeguards and human oversight.
- Workforce
- Enterprise
Research
Human-Governed Validation of Artificial Intelligence-Generated Medical Assessment Artifacts: A Technical Report
Vinícius Côgo Destefani, Matheus Feliciano C Ferreira, Afrânio Côgo Destêfani
Cureus · 2026-08-28
This technical report proposes a seven-gate human-governed validation workflow for reviewing AI-generated medical assessment materials—such as multiple-choice questions, clinical vignettes, and distractor sets—before they are used in scored or consequential assessments. The authors argue that linguistic fluency from generative AI does not guarantee measurement validity, and that risks related to clinical accuracy, fairness, blueprint alignment, and score interpretation require staged expert review combined with empirical psychometric evaluation. The workflow separates clinical content review from psychometric evidence (e.g., DIF analysis, distractor analysis) while preserving documented human accountability at every stage. The paper recommends treating AI-generated items as preliminary candidates subject to governance rather than ready-made deployment materials.
- Quality assurance
- Certifications
Research
From technological closed-loop to human-machine trust: a systematic review of ethical and communication challenges of brain-computer interface in elderly care
Kun Fu, Peize Li, Peng XinXi et al.
Frontiers in Digital Health · 2026-08-28
This systematic review of 177 studies examines the ethical and communication challenges of deploying brain-computer interfaces (BCIs) in elderly care. The authors identify six core ethical themes—including privacy, informed consent, personhood, safety, equity, and risk-benefit trade-offs—and construct a three-level analytical framework showing how these risks manifest as communication breakdowns across BCI processes. Key findings include that decoding uncertainty can cause care misjudgment, neural data raises privacy risks beyond information leakage, and technology-mediated communication creates structural tensions in trust and accountability. The review concludes that responsible BCI deployment requires embedding ethical principles into care processes at the interface, institutional, and interdisciplinary levels.
- AI policy
- Quality assurance
Research
The numb efficiency paradox: AI work pressure, affective numbing, and professional judgment in organizationally digitally mediated work
Shiyao Yin, Wenxuan Hu, Zirong Tian et al.
Frontiers in Psychology · 2026-08-28
This three-wave longitudinal panel study of accounting and auditing professionals (n=512 at baseline) introduces the 'numb efficiency paradox,' where AI-related work pressure can lead to affective numbing that undermines ethical vigilance and professional skepticism even as task efficiency is maintained. Results show that while occupational resilience supported sustained work efficiency, affective numbing was negatively associated with both ethical vigilance and professional skepticism, meaning professionals may appear productive while becoming less attentive to warning signs and professional consequences. The findings suggest that AI implementation in accounting should be evaluated not just on efficiency metrics but on whether professionals retain the judgment quality required for auditing integrity.
- Workforce
- Quality assurance