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
Digital transformation readiness: evaluating Dutch municipalities’ compliance with the AI Act
Willem Bantema, Trix Mulder, Soner Korucu
Edward Elgar Publishing eBooks · 2026-07-21
This study assesses how prepared Dutch municipalities are to comply with the EU AI Act, which took effect in August 2024 and requires full municipal compliance by 2026. Using survey data and interviews with municipal representatives, the authors find that while awareness of the Act is high, significant gaps remain between regulatory knowledge and actual implementation, with municipalities struggling due to limited legal capacity, uneven digital transformation, and ethical uncertainty. The result is fragmented, largely reactive AI governance at the local level. The authors recommend strengthening AI literacy, clarifying regulations, improving ethical oversight, and developing a coordinated national strategy to bridge the gap between policy goals and local governance.
- AI policy
- Certifications
Research
SafeBoundary-LLM: Measuring Safety Boundary Stability in Local Open-Weight LLMs Through Single-Turn Baselines and Multi-Turn Escalation
Andreea Alexandra Anghel, Cătălin Anghel, Emilia Pecheanu et al.
Computers · 2026-07-21
SafeBoundary-LLM evaluated seven local open-weight large language models across 14 sensitive domains using single-turn prompts and multi-turn escalation conversations, finding that multi-turn failure rates (14.69%) were dramatically higher than single-turn failure rates (0.51%), a rate ratio of 28.80. Boundary collapse occurred specifically at turns 4–5, with role-play bypass accounting for 299 of 456 total failures. The study also found notable over-refusal rates on legitimate queries, particularly in JBB-Behaviors (17.29%). These results argue for evaluation frameworks that combine public benchmarks, controlled multi-turn testing, independent human review, and traceable audit records when deploying LLMs locally in privacy-sensitive contexts.
- Quality assurance
- Certifications
Research
Employees show greater willingness to voice toward algorithmic than human leaders in cognitive tasks through fairness perception and psychological safety
Shiqi Wang, Xiaoling Sun, Suhang Ni et al.
Scientific Reports · 2026-07-21
Three experimental studies conducted in China found that employees are more willing to voice concerns or suggestions to algorithmic leaders than to human leaders, but only in cognitive tasks—the effect disappears for emotional tasks. The mechanism works serially: algorithmic leaders are perceived as fairer, which boosts psychological safety, which in turn increases willingness to voice. These findings offer practical guidance for how organizations deploy and design AI-based management systems.
- Workforce
- Enterprise
Research
Editorial: Legal-Technical Governance and Regulation of Artificial Intelligence in Africa
Chijioke Okorie, Vukosi Marivate
Potchefstroom Electronic Law Journal/Potchefstroomse Elektroniese Regsblad · 2026-07-21
This editorial introduces a special issue on AI governance and regulation in Africa, presenting five peer-reviewed contributions developed through a collaboration between the Data Science Law Lab and the Data Science for Social Impact group at the University of Pretoria. The papers apply a transdisciplinary 'legal-technical' framework that fuses legal and technical perspectives to address gaps in AI regulation, covering topics ranging from African Union soft law and AI treaty development to intellectual property in national AI strategies and algorithmic surveillance in South Africa and Nigeria. The work is relevant to policymakers and regulators seeking to understand how African jurisdictions are approaching AI oversight, including issues of algorithmic sovereignty and predictive policing. The special issue highlights the need to close the legal-technical gap in AI governance across the African continent.
- AI policy
Research
AI Governance and Institutional Readiness in Public Sector Systems: A Comparative Study Between India and Nigeria
Linda John Obiorah, Banavath Ananditha, Mrinalini Menon et al.
Zenodo (CERN European Organization for Nuclear Research) · 2026-07-21
This qualitative comparative case study examines how governance frameworks and institutional capacities shape AI adoption in India and Nigeria, using South Korea as a benchmark. The analysis finds that India has a more developed AI governance ecosystem—anchored by its 2018 National Strategy and 2024 India AI Mission—while Nigeria shows growing but nascent institutional commitment through its 2024 National AI Strategy. Both countries share common challenges including gaps between policy aspirations and implementation, difficulty enacting specific AI legislation, and concerns around accountability, transparency, bias, and digital inclusion. The study argues that meaningful governance progress should be measured against concrete indicators such as enforcement actions and institutional capacity, not policy adoption alone, and calls for specific AI laws, autonomous regulatory bodies, and stronger data governance in the Global South.
- AI policy
Research
Centering Intellectual Property in Artificial Intelligence Strategies in Africa Through a Techno-legal Analysis
Desmond Oriakhogba
Potchefstroom Electronic Law Journal/Potchefstroomse Elektroniese Regsblad · 2026-07-21
This paper analyzes how intellectual property (IP) should be integrated into African AI strategies at both national and continental levels, using a techno-legal and comparative policy methodology. Reviewing existing African Union and national AI strategies alongside developments in the US, China, Australia, and Europe, it argues that whether new IP rules are needed—or existing ones should be revised or simply applied—depends on the specific AI-IP issue at hand. The paper advocates for a human-centered, development-focused approach that balances AI innovation, data access, IP protection, and fair rewards for creators and indigenous communities. Its findings are directly relevant to policymakers shaping the regulatory and legal frameworks governing AI across Africa.
- AI policy
Research
Human-centered AI in healthcare teams: integrating clinical oversight, EHR auditability, and reimbursement pathways for responsible adoption
Michael C. Changaris, Franca V. Niameh
Frontiers in Digital Health · 2026-07-21
This narrative synthesis reviews clinical, economic, regulatory, and implementation-science literature from 2022–2025 to identify the structural conditions needed for responsible AI adoption in healthcare teams. The paper finds that AI tools show benefits in diagnostic accuracy, decision support, and documentation efficiency—particularly in radiology, cardiology, and EHR-integrated workflows—but that adoption is stalled by absent reimbursement pathways for clinician-reviewed AI outputs, high implementation costs falling on health systems, and limited EHR governance standards. The authors conclude that scaling AI safely requires aligned financial incentives, auditable outputs, equity-centered validation, and coordinated oversight, drawing parallels to lessons from national EHR rollout.
- Workforce
- AI policy
Research
Data Governance and Policy Support for Secure AI-Driven Corporate Digital Transformation
Fang Sun
Journal of Reliable and Secure Computing · 2026-07-21
This paper develops a unified security-aware governance framework for AI-driven corporate digital transformation, addressing both data governance (classification, provenance, access, privacy) and AI governance (model validation, robustness, auditability). It identifies six core dilemmas—including data sharing versus protection, adversarial AI risks, cloud-edge-IoT vulnerabilities, and compliance gaps between large firms and SMEs—and proposes an integrated agenda spanning tiered governance, zero-trust security, privacy-enhancing collaboration, algorithmic audits, and regulatory sandboxes. The study argues that technical controls, organizational routines, and policy support must be combined to enable trustworthy AI adoption across firms of different sizes and sectors. The findings are relevant to enterprise AI deployment, regulatory compliance, and the design of certification and policy instruments.
- Enterprise
- AI policy
- Certifications
Research
Artificial intelligence use among healthcare professionals in healthcare facilities in Lagos, Nigeria: a cross-sectional study
Anirejuoritse Alero Chima-Oduko, Blossom Maduafokwa, Kamaldeen Sunkanmi Abdulraheem et al.
BMC Digital Health · 2026-07-21
A cross-sectional survey of 415 healthcare professionals across 14 public facilities in Lagos, Nigeria found that 86.5% reported using AI, yet only 19.8% had received formal AI training. Over half (51.3%) used AI specifically for professional duties such as diagnostic support, clinical reasoning, and medication-interaction checks, with postgraduate-qualified professionals twice as likely to use AI professionally compared to those with only basic degrees. Key barriers included privacy concerns, algorithmic distrust, and limited formal training, pointing to an urgent need for structured AI literacy programs and institutional governance frameworks to support safe and equitable AI integration in Nigeria's health system.
- Workforce
- AI policy
Research
Beyond Human Oversight: Cognitive Sovereignty in Global Governance Frameworks for Agentic Artificial Intelligence
Kwan Hong TAN
arXiv · 2026-07-21
This qualitative study compares ten major AI governance frameworks (from UNESCO, OECD, EU, Council of Europe, NIST, UK, G7, and Singapore) and finds that while they broadly endorse human oversight, they fail to specify what capabilities people must retain when AI systems take autonomous, multi-step actions with real-world consequences. Through provision-level coding and cross-framework analysis, the paper identifies six recurring gaps—including oversight without empowerment, a reversibility deficit, and temporal-capability asymmetry—and develops the concept of 'cognitive sovereignty' to describe the practically exercisable capacity to understand, authorize, interrupt, contest, restore, and assign responsibility for AI-delegated processes. It then proposes the CLEAR² framework (Comprehension, Legitimate authorization, Effective intervention, Appeal and contestation, Restoration and reversibility, Responsibility and remedy) as a structured lifecycle control architecture. The work matters because it shifts AI governance analysis from human presence to preserved human agency, offering organizations a maturity model and audit questions for responsible deployment of agentic AI.
- AI policy
- Enterprise
Research
The sustainability paradox: A systematic review of carbon footprints and decarbonization strategies for artificial intelligence in climate science
Nohman Khan, Huma Sikandar, Mohammad Falahat et al.
Sustainable Environment · 2026-07-21
This systematic review of 59 peer-reviewed studies examines the 'sustainability paradox' of using carbon-intensive AI systems to address climate change. The meta-analysis finds that current methods systematically underestimate AI carbon footprints by 300–500% due to excluded inference phases and hardware lifecycle impacts, while also showing that integrated technical mitigation strategies can achieve a pooled carbon reduction of roughly 32–38%. The authors identify critical governance gaps—including inconsistent measurement boundaries in 77% of studies and only 32% disclosure compliance—and propose the Sustainable AI Carbon Accounting Protocol (SAICAP) with mandatory disclosure categories and a compliance structure to standardize measurement and oversight.
- AI policy
- Quality assurance
Research
A Model-Agnostic Framework for Transparent, Fair, and Reproducible Automated Essay Scoring
Ahsan Javed
Journal of Global Social Transformation · 2026-07-21
This paper presents a model-agnostic framework for automated essay scoring that prioritizes transparency, fairness, and reproducibility alongside accuracy. The authors benchmark classic and transformer-based models—including a fine-tuned DeBERTa-v3-large achieving macro-F1 of 0.898 on the ASAP2.0 dataset—within a reproducible pipeline using deterministic controls and artifact tracking. SHAP and LIME are integrated to explain model decisions and measure demographic disparities, while a prompt-aware binning strategy normalizes scores into proficiency tiers to improve fairness analysis. The framework aims to make AI-driven essay scoring more auditable and trustworthy for students, educators, and policymakers.
- Quality assurance
- Certifications
- AI policy
Research
AI Governance and Institutional Readiness in Public Sector Systems: A Comparative Study Between India and Nigeria
Linda John Obiorah, Banavath Ananditha, Mrinalini Menon et al.
Zenodo (CERN European Organization for Nuclear Research) · 2026-07-21
This paper uses a qualitative comparative case study of India, Nigeria, and South Korea to assess how governance frameworks and institutional capacity shape AI adoption in public sector systems. It finds that India has a more mature AI governance ecosystem—anchored by its 2018 National Strategy and 2024 India AI Mission—while Nigeria shows growing but nascent institutional commitment through its 2024 National AI Strategy, and South Korea serves as an advanced benchmark. Both developing countries share common challenges including gaps between policy intent and implementation, weak AI-specific legislation, and deficits in accountability, transparency, and digital inclusion. The study calls for dedicated AI laws, independent regulatory bodies, and concrete governance metrics beyond policy adoption to build responsible and sustainable AI governance in the Global South.
- AI policy
Research
Towards a Transdisciplinary Governance of Artificial Intelligence in Africa
Chijioke Okorie
Potchefstroom Electronic Law Journal/Potchefstroomse Elektroniese Regsblad · 2026-07-21
This article introduces 'legal-technical governance' as an analytical framework for understanding how AI is regulated across Africa, arguing that governance is already distributed across data protection laws, fintech guidelines, cybersecurity frameworks, and platform-imposed content moderation policies. Drawing on Dooyeweerd's modal aspects as a philosophical lens, the authors reveal how institutional fragmentation, infrastructural dependency, and global platform dominance undermine state-centered regulatory models. The framework highlights the co-constitutive relationship between legal norms and technical operations—such as data labeling, model training, and algorithmic auditing—and accounts for a broad range of actors including multinational firms, standards bodies, and affected communities. The article offers a systems-level analytical tool intended to guide both scholarship and policymaking at the intersection of law, technology, and development in Africa.
- AI policy
- Certifications
- Enterprise
Research
Editorial: Legal-Technical Governance and Regulation of Artificial Intelligence in Africa
Chijioke Okorie, Vukosi Marivate
Potchefstroom Electronic Law Journal/Potchefstroomse Elektroniese Regsblad · 2026-07-21
This editorial introduces a special issue focused on the legal-technical governance and regulation of AI in Africa, produced through collaboration between the Data Science Law Lab and the Data Science for Social Impact research group at the University of Pretoria. It presents a transdisciplinary framework that fuses legal and technical perspectives to close gaps in AI governance, covering topics ranging from African Union soft law and potential AI treaty development, to national intellectual property policy in African AI strategies, algorithmic sovereignty in surveillance systems, and the legal status of predictive policing for cybercrime in Nigeria. The work matters because it provides both conceptual and applied frameworks for how African nations and continental bodies can develop more coherent, context-specific AI regulation. This has direct implications for policy design and governance practices across the continent.
- AI policy
- Certifications
Research
When Does Machine Learning Beat Value Sorting? A Three-Dataset Diagnostic of Exposure-Weighted Shipment Prioritization
Jize Li
arXiv · 2026-07-20
This paper investigates whether machine learning models can outperform a simple baseline—sorting shipments by value—when prioritizing which shipments a manager should review first under limited capacity. Across three real supply-chain datasets (SCMS procurement, DataCo logistics, and Olist e-commerce), the authors find that ML beats value sorting only when delay severity is learnable (as measured by R² and calibration bias), and fails in two of the three contexts. Rather than proposing a new algorithm, the paper offers a deployment diagnostic and evaluation protocol, recommending that value sorting remain a permanent benchmark and that ML be deployed only after passing learnability and calibration audits under leakage-controlled rolling-origin evaluation. The findings have direct implications for enterprise supply-chain operations and quality-assurance processes around model deployment decisions.
- Enterprise
- Quality assurance
Research
Attacking Graph Foundation Models Through Their Shared Representation
Pankaj Kumar, Subhankar Mishra
arXiv · 2026-07-20
This paper identifies the 'alignment layer' in graph foundation models — the component that maps diverse graph inputs into a shared representation — as a previously unstudied attack surface distinct from standard graph neural network vulnerabilities. The authors demonstrate that directed perturbations in representation space can collapse six publicly available graph foundation models at inference time, with no access to training data, and that one model (OpenGraph) is especially fragile due to its spectral tokenizer rather than its decoder. Realizable input-space attacks (editing edges, features, or text) eliminate at least half of correct predictions on three of the six models, with attack effectiveness tracking the decoder's local Lipschitz sensitivity rather than clean task accuracy. These findings matter for quality assurance and enterprise deployment of graph foundation models, revealing that architectural choices in the alignment layer introduce exploitable fragilities that standard robustness metrics do not capture.
- Quality assurance
- Enterprise
Research
The Story Shapes the Agent: Narrative Priors in LLM Behavior
Yixuan Wang, James Lester, Shashank Srivastava
arXiv · 2026-07-20
This paper investigates how the narrative framing of a task affects large language model (LLM) agent behavior, finding that story context can matter far more than the assigned persona. Using three structurally identical text-based investigation games (disease investigation, IT troubleshooting, and murder mystery) that differ only in narrative, the researchers ran 1,890 sessions across 3 models and 10 personas, identifying 'narrative priors'—systematic behavioral tendencies driven by story framing that explain 5–31x more behavioral variance than persona and are negatively associated with task success in two of three domains. Persona effects that do transfer across narratives stem from 'behavioral anchors,' concrete action-linked language in persona descriptions, and removing those anchor words reduces cross-narrative consistency by 95%. These findings have important implications for enterprise and quality-assurance applications of LLM agents, suggesting that reliable behavior requires grounding in concrete actions rather than abstract persona descriptions.
- Enterprise
- Quality assurance
Research
Engineering Trustworthy Agentic AI for Critical Systems
Omar Al-Refai, Ibrahim Shahbaz, Adam Ali Husseinat et al.
arXiv (Cornell University) · 2026-07-20
This survey addresses how agentic AI systems—capable of autonomous perception, planning, tool use, and multi-step action—can be made trustworthy for critical engineering domains where decisions carry physical, operational, or economic consequences. The authors define trustworthiness as a first-class engineering property organized around five dimensions: safety and constraint satisfaction, robustness and reliability, transparency and interpretability, accountability and auditability, and privacy and security. The framework is applied across four engineering domains—power systems, autonomous vehicles/robotics/UAVs, high-performance computing, and communication networks—identifying shared failure modes and design patterns. The paper argues that agentic AI trustworthiness is a single cross-domain problem and proposes a path toward a reusable assurance framework analogous to graded certification regimes in mature safety-critical fields.
- Certifications
- Quality assurance
- AI policy
- Enterprise
Research
Censoring-Aware In-Context Learning for Generalized Supplier Lead Time Estimation in Supply Chain Planning
Christopher Wang, Sebastien Ouellet, Behrouz Haji Soleimani et al.
arXiv · 2026-07-20
This paper introduces LeadTime-ICL (LT-ICL), a censoring-aware in-context learning model that combines a transformer backbone with a conditional normalizing-flow head to produce probabilistic forecasts of supplier lead times in supply chains. A key challenge addressed is that many industrial lead time datasets are right-censored—some orders haven't arrived when forecasts are needed—and standard regression or classification models discard this information. Pretrained on synthetic right-censored tasks, LT-ICL adapts to new industrial datasets without task-specific parameter updates, achieving the lowest point-forecasting error on 15 of 24 proprietary supply-chain datasets and the lowest probabilistic forecasting error on 14 of 24. The results demonstrate that pretrained in-context models can deliver accurate, low-adaptation-cost lead time forecasting relevant to material requirements planning, inventory optimization, and supply chain risk management.
- Enterprise
- Workforce
Research
EduPanel: A Three-Agent LLM Judge for Teaching Videos -- Reliability, Complementarity, and Human Trust Calibration
Jia-Kai Dong, Yi-Cheng Lin, Hung-yi Lee
arXiv · 2026-07-20
EduPanel is a three-agent LLM-based system that evaluates the pedagogical quality of teaching videos using rubric-grounded, learner-conditioned assessments. The system decomposes evaluation across specialized agents to handle multimodal evidence and produces interpretable feedback tailored to different learner personas. In expert studies, EduPanel achieves reliability comparable to a median human expert and improves scoring accuracy (MAE from 0.87 to 0.73), while human experts retain the ability to detect unreliable outputs (AUC = 0.77). The findings suggest EduPanel can serve as a scalable assistant for educational video evaluation without replacing human expert judgment.
- Quality assurance
- Workforce
Research
Querying Multimodal Scientific Papers with AI: Practices and Preferences Across Blind, Low-Vision, and Sighted Scientists
Arnavi Chheda-Kothary, Lucy Lu Wang, Joseph Chee Chang et al.
arXiv · 2026-07-20
This paper investigates how blind, low-vision, and sighted scientists use AI tools—specifically ChatGPT and Gemini—to query visual elements such as figures, diagrams, and tables in scientific documents. Through interviews with ten scientists across STEM fields, the authors find that vague or incorrect AI-generated descriptions of images lead both blind/low-vision and sighted users to abandon AI workflows, highlighting reliability as a critical barrier. The study contributes a dataset of 115 real queries and AI responses and surfaces design implications for building more accessible and accurate AI-powered scientific question-answering systems. The findings matter for quality assurance of AI tools and for enterprise and workforce considerations around equitable access to scientific knowledge.
- Workforce
- Enterprise
- Quality assurance
Research
Towards an Automated Test of LLM Security Knowledge
Shufan Chai, Liangliang Sun, Jessica Staddon
arXiv · 2026-07-20
This paper presents a partially-automated method for assessing LLMs' knowledge of security topics by using authoritative information from Consumer Protection Agencies (CPAs) to detect instability in LLM responses—instability that can signal knowledge gaps. The approach is demonstrated across two security topics (identity theft and impostor scams) and five LLMs from the Gemini and GPT families, using publicly available CPA information. The method successfully distinguishes between models with sufficient and insufficient knowledge to accurately identify security topics in text narratives. This matters for quality assurance and certification of AI systems used in security-sensitive contexts, offering a more scalable alternative to manually curated benchmarks.
- Quality assurance
- Certifications
- AI policy
Research
Structured Output Collapses Answer Diversity Across 44 Language Models
Tapan Parikh
arXiv · 2026-07-20
This paper investigates how requesting structured output formats (like JSON) from language models affects the diversity of answers they produce. Across 44 models and 31 open-ended prompts, adding a simple format instruction ('Reply with JSON only') caused the most common answer to rise from 41% to 64% of responses, while distinct answers fell from 52 to 36 and mean answer-choice surprisal dropped from 1.80 to 1.58 bits. The compression effect is specific to formats models are trained on for tool use (JSON and XML) and is not explained by decoder-level schema enforcement — it lives in the model's learned response to the register itself. This matters because software systems typically consume language models through structured output interfaces, meaning the models being deployed in real enterprise pipelines are measurably more homogeneous than the chat-surface models that benchmarks, comparisons, and purchasing decisions are based on.
- Enterprise
- Quality assurance
- Certifications
- AI policy
Research
Intelligent Cause Prioritisation? An Analysis of AI Policy Priorities and Governance in Africa
Osaremen Iluobe, Kisso Selvan
arXiv (Cornell University) · 2026-07-20
This paper examines how African governments are framing and prioritizing AI in their national strategies and public communications. Drawing on speeches, press releases, public statements, and national AI strategies, the authors find that while African policymakers are highly attentive to AI's economic and developmental opportunities, they devote comparatively little attention to AI safety. The paper argues that Africa's position as largely a consumer rather than a producer of frontier AI systems, combined with pressing development challenges, shapes a policy orientation focused on catching up economically. The findings highlight a governance gap that could influence who benefits from AI and who bears its risks on the continent.
- AI policy