AI Frontier Institute

News & Research

The latest AI research and news with real-world stakes. Each item is sourced, dated and summarized in plain English, tagged by impact area where one fits, and its summary is checked against the text it was written from.

• ResearcharXiv2026-07-01Quality assurance · Algorithms & Automated Decisions

  From Technical Metrics to User Perception: A User Study of a Multimodal Human-Robot Interaction System for Object Detection and Grasping · Jian Song, Tian Zi, Shen Guanting

  This paper investigates whether a 15 percentage point improvement in end-to-end task success (from 75% to 90%) in a multimodal human-robot interaction system for object grasping is noticeable to real users. In a within-subject study with 24 participants, the improved system—swapping Florence-2 for Grounding DINO + SAM and LLaMA 3.1 for Qwen 3.5 9B—was preferred by 70.83% of participants (p = 0.043), and rated significantly higher on perceived speed, reliability, and overall competence and fluency with large to very large effect sizes (p < 0.001). The findings confirm that the technical gains are perceptible during live interaction, underscoring the need to complement benchmark evaluations with user-centred evidence when assessing robotic manipulation pipelines.
• ResearcharXiv2026-07-01AI policy · Algorithms & Automated Decisions

  AI, Trust, and Teaming: The Humans-as-Handlers Approach for Autonomous and Opaque AI Systems · Nathan G. Wood

  This paper proposes a 'humans-as-handlers' framework for governing relationships between people and autonomous, opaque AI systems, particularly in high-stakes domains like medicine and warfighting. Drawing an analogy to the relationship between humans and working animals such as dogs, the author argues that people using these systems should be recast as 'handlers' rather than mere 'users' or 'deployers,' which clarifies traceable lines of human responsibility for AI-driven outcomes. The paper acknowledges limits of the analogy and outlines how the framework can be refined to suit AI contexts, ultimately envisioning a trajectory toward treating AI systems as genuine collaborators rather than tools. This work matters for AI policy and accountability because it offers a concrete conceptual model for assigning human responsibility when AI systems are autonomous and their internal workings are opaque.
• ResearcharXiv2026-07-01Quality assurance

  MolSafeEval: A Benchmark for Uncovering Safety Risks in AI-Generated Molecules · Tong Xu, Xinzhe Cao, Zhihui Zhu et al.

  MolSafeEval is a new benchmark designed to evaluate safety risks in AI-generated molecules, an area largely overlooked by existing molecular generation benchmarks that focus on novelty and property alignment. The system integrates toxicological databases and hazard rules into a structured molecular safety knowledge graph, which then supports large language model-based reasoning to detect and explain unsafe features—such as toxicity or reactivity—in generated compounds. The benchmark covers four generative task types (unconditional generation, property optimization, target protein-based design, and text-based generation) with standardized datasets and evaluation protocols for each. By systematically exposing safety vulnerabilities in current generative approaches, MolSafeEval provides a foundation for more trustworthy and safer AI-driven molecular design.
• ResearcharXiv2026-07-01Quality assurance

  A Penny for Your Prompts: Experiments Detecting and Mitigating LLM Usage by Survey Respondents · Zane Xu, Nathan Malkin

  This study investigates how often survey respondents on crowdsourcing platforms use large language models (LLMs) to answer surveys and tests methods to detect and reduce this behavior. Across a series of surveys (N=250), the researchers found LLM-assisted response rates varied dramatically—under 10% on Prolific but over 80% on Mechanical Turk—depending on platform, survey length, and other conditions. Mitigation measures such as disabling copy-paste and requesting no AI use reduced LLM usage but did not necessarily improve overall data quality. The authors recommend that researchers screen for LLM usage by recording keystroke data and designing questions and instructions specifically aimed at detecting AI-generated responses.
• ResearchProceedings on Privacy Enhancing Technologies2026-07-01Quality assurance · AI policy · +1

  AudAgent: Automated Auditing of Privacy Policy Compliance in AI Agents · Yuyan Zheng, Yimin Chen, Yidan Hu

  AudAgent is an automated auditing tool that continuously monitors AI agents' runtime data practices and checks them against stated privacy policies. The system uses a cross-LLM voting mechanism to parse policies into formal models, a lightweight analyzer to detect sensitive data, and ontology-based compliance verification to flag violations in real time. Evaluations show that many privacy policies lack explicit safeguards for highly sensitive data like SSNs, and that agents powered by Claude, Gemini, and DeepSeek do not refuse to process such data through third-party tools. AudAgent proactively blocks such operations, providing transparency and accountability for trustworthy AI agent deployments.
• ResearchHumanities and Social Sciences Communications2026-07-01Enterprise

  Artificial Intelligence Adoption and Organizational Resilience in SMEs: The Roles of Ambidextrous Innovation and Environmental Unpredictability · Peng Peng, Xintian Li, Huanhuan Hu

  This study examines how AI adoption affects organizational resilience in small and medium-sized enterprises (SMEs), using survey data from 275 Chinese SMEs. The findings show that AI adoption significantly boosts organizational resilience, with ambidextrous innovation—particularly exploitative innovation—serving as a key mediating mechanism. Environmental unpredictability further strengthens the relationship between AI adoption and resilience, offering practical guidance for SMEs navigating uncertainty.
• ResearchIPSI Transactions on Internet Research2026-07-01Enterprise · Quality assurance · +1

  Generative AI in Audit Oversight: A Checklist-Based Model Evaluation · Dejana Kresović, Sofija Drulović, Nebojša Đoković et al.

  This paper proposes and tests a checklist-based framework using generative AI (ChatGPT and Gemini) to assist audit oversight by automating the preliminary review of independent auditor's reports and financial statements. Evaluated on 217 sets of documents from Serbian public companies in 2024, both models were benchmarked against expert assessors using metrics such as Cohen's kappa, F1-measure, and ROCAUC. ChatGPT showed higher sensitivity while Gemini was more conservative, and both performed better on formal and structured checklist categories than on areas requiring professional judgment. The findings suggest GenAI can improve consistency and prioritization in preliminary audit review but must be integrated with expert validation.
• ResearchINTERANTIONAL JOURNAL OF SCIENTIFIC RESEARCH IN ENGINEERING AND MANAGEMENT2026-07-01Enterprise · Quality assurance · +1

  “Governing AI Investment: A System-Level Assurance Framework for Financial Institutions” · Ayasha Vadhyani Ayasha Vadhyani

  This paper examines AI governance gaps in financial institutions using data from the 2025 EMEA Model Risk Management Survey, covering 87 banks and 49 insurance companies across Europe, the Middle East, and South Africa. Key challenges identified include transparency and explainability, skills gaps, and regulatory uncertainty. To address these, the authors propose a System-Level Intelligence Assurance framework built on four pillars—architecture, shared memory, continuous validation loop, confidence governance, and auditable evidence chain—extending traditional Model Risk Management to cover agentic AI systems with ongoing, architecture-embedded oversight rather than static documentation. The findings are intended to help financial institutions, regulators, and risk practitioners scale AI safely while maintaining accountability.
• ResearchJournal of Hospitality Marketing & Management2026-07-01Workforce · Enterprise · +1

  AI-Induced occupational uncertainty as a multi-level challenge: reconciling human-centered values with automation in tourism and hospitality · Mahlagha Darvishmotevali, Kevin Kam Fung So, Billy Bai

  This paper examines how AI adoption in tourism and hospitality creates occupational uncertainty across individual, organizational, familial, and societal levels. Using a qualitative multi-level critical synthesis, the authors find that operational gains from AI come at the cost of psychological stress, family relational strain, and deepening societal inequality—a 'double-edged sword' effect. The study introduces a four-level diagnostic framework and calls for agile policy responses including participatory co-design, targeted reskilling, and universal social protections. The findings are directly relevant to policymakers and industry leaders seeking to align technological innovation with human-centered values in hospitality work.
• ResearchZenodo (CERN European Organization for Nuclear Research)2026-07-01Quality assurance · Certifications · +3

  The UK Government Has Deployed AI Across Hundreds of Public Services. The Permanent Secretary Accountable for Its Governance Cannot Certify a Single Output as Constitutionally Verifiable. · Preethi Sharma, Akhil Sharma

  This paper identifies a structural governance gap in the UK government's deployment of AI across hundreds of public services, including benefits fraud detection, healthcare triage, law enforcement, and immigration processing. It documents that despite DSIT's administrative AI governance framework and PAC recommendations, no technical specification exists that would make AI-governed decisions cryptographically verifiable in court proceedings. The central finding is that the Permanent Secretary accountable for AI governance cannot certify any AI output as constitutionally verifiable, meaning the Accounting Officer certification of correct AI operation lacks a legally defensible foundation. The paper argues that a constitutional command architecture is required but has not been technically specified or published.
• Researche-Journal of Nondestructive Testing2026-07-01Workforce · Enterprise · +2

  Generative AI in NDE: Connecting human expertise and data for enhanced inspection support · Marco Induti, Carlo Romito, Stefano Cipolla et al.

  This paper describes the development of IVA (Inspector Virtual Assistant), a generative AI tool built on Retrieval Augmented Generation (RAG) architecture to assist nondestructive testing (NDT) inspectors during preparation, execution, and reporting. IVA retrieves contextual information from reports, standards, drawings, and certificates to ground answers in trusted documents rather than relying on general foundation model knowledge. The system is being deployed first within the Pressure Vessel Inspectorate (Kesselinspektorat) to address workforce challenges including an ageing inspector population, shortages of qualified personnel, and difficulty transferring tacit knowledge. The work demonstrates how enterprise AI tools can reduce administrative burden and support consistent, knowledge-grounded inspection decisions in regulated TIC (Testing, Inspection and Certification) sectors.
• ResearchRevista Española de Educación Comparada2026-07-01AI policy · Education

  Regulating artificial intelligence in Higher Education: · Ángela Martínez Rojas, Cristobal Suárez Guerrero, Vladimir E. Martínez-Bello

  This study analyzes AI regulations established by universities in the United States and Spain, using a qualitative approach to compare how higher education institutions govern AI use in learning, teaching, and research. The paper finds that while the two countries have different legal frameworks, both seek to minimize potential harms through ethics-based guidelines rather than penalties or restrictions. Notably, neither country has developed solid policy addressing the epistemological challenges AI poses to research. The findings are relevant to higher education policy and quality assurance, highlighting gaps in institutional governance of AI.
• ResearchSelected Issues Papers2026-07-01Workforce

  The Impact of Artificial Intelligence on Israel’s Labor Market · Ece Ozge Emeksiz

  Using occupational microdata, this paper analyzes how Generative AI could reshape Israel's labor market relative to selected European economies. The findings indicate that while most Israeli workers are likely to benefit from AI adoption through productivity gains, approximately one-fifth of the workforce faces high AI exposure with low complementarity, making them vulnerable to displacement. The authors recommend a comprehensive lifelong learning strategy—including reskilling, upskilling, and mid-career training—to support workers at risk.
• ResearcharXiv2026-07-01Workforce · AI policy

  AI AND THE TRANSFORMATION OF THE LABOR MARKET: THE SOCIAL CONSEQUENCES OF AUTOMATION AND THE NEW EMPLOYMENT UNCERTAINTY · Nurlan Baigabylov, Alimzhan Yessenovabylov

  This study examines the socio-economic effects of AI and generative AI on global and national labor markets during 2025–2026, drawing on secondary quantitative data from WEF, ILO, McKinsey, PwC, and Kazakhstan's Center for Human Resources Development. Key findings include a projected global net gain of 78 million jobs by 2030, but with 22% of employment undergoing structural change and 39% of current skills becoming obsolete, while women in high-income countries face an automation risk nearly three times higher than men. The study introduces the concept of 'Precariousness 2.0,' describing a state of manufactured uncertainty and chronic anxiety among workers, and calls for gender-sensitive retraining, regional R&D equity, and mitigation of 'cultural debt' to support the emerging 'AI precariat.' Kazakhstan's 'Law on AI' (2026) and 'Alem.AI' ecosystem are highlighted as proactive policy responses to the potential transformation of 2.2 million workers.
• ResearchLegal Issues in the Digital Age2026-07-01AI policy

  Artificial Intelligence. Law. Industries · Irina Bogdanovskaya, Алексей Волос, Nikita Danilov et al.

  This paper reports on the XIV International Scientific and Practical Conference 'Law in the Digital Age' held at HSE in October 2025, which convened scholars and practitioners to discuss AI and law across multiple domains including civil law, intellectual property, healthcare, education, and digital platforms. The conference also included an international panel on AI and law developments in BRICS countries. The event aimed to integrate theoretical and applied perspectives on AI regulation for a more comprehensive understanding of the field. Its coverage of AI regulatory policies as a factor in national education systems and cross-national governance makes it relevant to both policy and certification discussions.
• ResearchProblemy Ekorozwoju2026-07-01AI policy

  Artificial Intelligence in Global Libraries: Pathways and Policies for Sustainable Development · Shuang Zheng, Qianbai Dai, R. Abubakar

  This systematic literature review examines how AI is reshaping libraries globally from 2019 to 2025, identifying four stages of AI integration and showing that tools such as automated classification, chatbots, and analytics dashboards have improved user services and broadened information access in ways linked to UN SDGs 4, 8, and 9. The study finds persistent obstacles including algorithmic bias, data ownership disputes, labor disruption, and uneven AI adoption across nations. A cross-national comparison reveals that China's centralized model enables rapid deployment while Western nations prioritize ethical oversight and equity, often at the cost of slower rollout. The paper proposes a policy framework centered on layered governance, built-in ethical checks, institutional skill-building, and eco-responsible AI operations for policymakers and library leaders.
• ResearchZenodo (CERN European Organization for Nuclear Research)2026-07-01Quality assurance · Certifications · +3

  The UK Government Has Deployed AI Across Hundreds of Public Services. The Permanent Secretary Accountable for Its Governance Cannot Certify a Single Output as Constitutionally Verifiable. · Preethi Sharma, Akhil Sharma

  This paper documents a structural accountability gap in the UK government's deployment of AI across hundreds of public services, including benefits fraud detection, healthcare triage, law enforcement, and immigration processing. Despite parliamentary oversight—including PAC recommendations and NAO findings—no governance document specifies a constitutional command architecture that would make AI-governed decisions cryptographically verifiable in court proceedings. The paper argues that DSIT's administrative AI governance framework lacks the technical specification required for the Permanent Secretary's Accounting Officer certification of correct AI operation to be legally defensible. This matters because AI systems are making high-stakes public service decisions without a published or legally defensible standard for verifying their outputs.
• ResearchZenodo (CERN European Organization for Nuclear Research)2026-07-01Enterprise · Quality assurance · +2

  Paper 10.1 – Decision Continuity Engineering Compliance Standard (Standard Edition) · Joel Monasterial

  This paper establishes a formal compliance framework for 'Sovereign AI' governance, derived from a broader charter document. It specifies testable requirements and evidence structures for verifying properties such as continuity, identity preservation, intent stability, multi-agent coherence, and governance boundary integrity across AI systems. The standard is positioned as a foundational reference for enterprises, regulators, and governance bodies implementing Decision Continuity Engineering, and defines baseline criteria for Sovereign AI compliance certification. Its relevance spans enterprise AI governance, quality assurance processes, and emerging certification and policy frameworks for AI systems.
• ResearchProceedings of the ... International Conference on Business Excellence2026-07-01Enterprise · AI policy

  Do AI and Digital Technologies Curb Greenwashing in ESG Reporting? · Artem SHAPOSHNIKOV, Svetlana RATNER, Inna Choban de Sousa Paiva et al.

  This paper synthesizes evidence from 76 empirical studies to assess whether AI and digital technologies reduce corporate greenwashing in ESG reporting. Using the PRISMA framework and quantitative synthesis of panel regression results, the authors find that AI and digital technology adoption is associated with a statistically significant but modest reduction in ESG disclosure-performance gaps (standardized β range: –0.17 to –0.03). The effect is stronger in state-owned enterprises, high-pollution industries, and larger firms, and is amplified by governance mechanisms such as institutional ownership and Big4 audit quality. The findings suggest these technologies primarily work by improving regulatory compliance efficiency, reducing information asymmetry, and enabling better resource allocation.
• ResearchAcademy of Management Proceedings2026-07-01Workforce · Enterprise

  Does AI Adoption Reduce Overtime? Empirical Evidence from Annual Reports and Satellite Data in China · Anqi Hu, xueyan li

  This study examines whether AI adoption by Chinese listed firms reduces overtime work, using annual report text analysis to measure AI adoption and satellite nighttime light data as a proxy for overtime intensity. Analyzing data from 2012–2019, the authors find that AI adoption significantly lowers overtime intensity, primarily by shifting human capital composition away from low-skill toward high-skill labor. The effect is especially pronounced among small and medium-sized enterprises and holds across varied industry and regional contexts. These findings suggest AI reshapes labor organization in ways that could inform workforce and labor policy decisions.
• ResearchZenodo (CERN European Organization for Nuclear Research)2026-07-01Enterprise · Quality assurance · +2

  Paper 10.1 – Decision Continuity Engineering Compliance Standard (Standard Edition) · Joel Monasterial

  This paper defines an operational compliance framework for 'Sovereign AI' governance, derived from a parent charter document. It specifies testable requirements and evidence structures for verifying continuity, identity preservation, intent stability, multi-agent coherence, and governance boundary integrity across AI systems. The standard is positioned as a foundational reference for enterprises, regulators, and governance bodies seeking certification in Decision Continuity Engineering. It establishes baseline criteria for Sovereign AI compliance certification, making it directly relevant to enterprise AI governance, quality assurance, and certification processes.
• Researche-Journal of Nondestructive Testing2026-07-01Enterprise · Quality assurance · +3

  AI in NDT: Hype, History and Realistic Adoption Pathways · Glenn Tubrett

  This paper examines the realistic adoption trajectory of artificial intelligence in non-destructive testing (NDT), arguing that change will be slower and more uneven than current narratives suggest. Drawing on historical technology hype cycles—such as early EVs, MOOCs, and Google Glass—the authors contend that safety culture, regulatory requirements, liability, data quality, and human acceptance will shape AI adoption more than technical capability alone. AI is expected to perform well in well-defined, repeatable inspection tasks but face slower uptake in complex or low-volume environments, with human certification, standards, and judgment remaining central. The paper's key message is that AI's impact on NDT will be evolutionary rather than instantly revolutionary.
• ResearchNational Bureau of Economic Research2026-07-01Workforce

  How Might Fiscal Policy Respond to the Rise of Artificial Intelligence? · Karen Dynan, Douglas Elmendorf, Louise Sheiner

  This paper analyzes how U.S. fiscal policy might need to adapt to different long-term economic scenarios driven by artificial intelligence, including faster productivity growth, rising income inequality, job displacement, and a higher capital share of income. For each scenario, the authors assess implications for federal debt and evaluate potential policy responses related to economic growth, income distribution, worker support, and capital taxation. The paper emphasizes that because AI's economic effects are deeply uncertain, policies that are robust across multiple scenarios would be especially valuable.
• ResearchAfrican Journal of Management and Business Research2026-07-01Enterprise · Quality assurance

  Adoption of AI-Based Accounting Systems and Audit Efficiency · Glory Ihuaku Ndukwe

  This study examines how adopting AI-based accounting systems affects audit efficiency in 15 Nigerian deposit money banks from 2019 to 2023. Using panel data and fixed effects regression, the authors find that AI adoption significantly improves overall audit efficiency, boosts error detection rates, reduces audit costs, and speeds up audit completion. The findings suggest AI is a key driver of audit transformation in emerging banking systems and support calls for greater investment in AI infrastructure and regulatory backing for audit digitisation.
• ResearcharXiv (Cornell University)2026-07-01Quality assurance

  Risk Architecture for AI-Native Engineering Teams: An Organizational Framework for Agentic System Governance · Laxmipriya Ganesh Iyer

  This paper develops an organizational framework for managing risk in engineering teams that build and operate agentic AI systems, arguing that traditional software risk frameworks—which assume deterministic behavior and clear component ownership—break down when applied to probabilistic, autonomous AI systems. The authors contribute a seven-dimension team profile taxonomy, a six-cluster failure-mode taxonomy (including a novel 'dependency-boundary determinism mismatch' cluster), and a methodology for scoring how well a team's risk architecture detects and escalates failures. Key findings show that risk coverage degrades monotonically as teams move from traditional software engineering to AI-native operation, with the most severe and least-covered failures occurring at organizational boundaries where AI outputs are consumed by systems that assume deterministic behavior. The work fills a gap between high-level policy frameworks (like NIST AI RMF and ISO/IEC 42001) and low-level threat taxonomies by addressing the management layer of roles, decision rights, and escalation structures.

← Back to the dashboard