AI Frontier Institute

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

  When Rubrics Fail: Hallucinations Reveal Blind Spots in Medical AI Evaluation

  Griffin Farrow, Lily Sijia Li, Jack Johnson et al.

  arXiv · 2026-09-11

  This paper investigates whether rubric-based evaluation methods—currently the dominant approach for assessing large language models (LLMs) in medicine—reliably detect clinically relevant hallucinations. Using a controlled setting with MedHallu and a novel clinician-validated error-injection pipeline spanning HealthBench, HealthBench Professional, and LiveMedBench, the authors show that medical hallucinations are frequently missed by rubrics, often leaving scores unchanged. Rubrics perform best when explicitly checking specific facts but fail for unanticipated or additional errors. The findings suggest that rubric scores alone are insufficient to establish clinical reliability and that complementary approaches, such as retrieval-based factuality checks, are needed to close these systematic blind spots.

  • Quality assurance
  • Certifications

• Research

  What is the Difference Between Me and You? Benchmarking the Quality Gap Between Human-Written and AI-Generated Code

  Cristina Improta, Pietro Liguori, Domenico Cotroneo

  arXiv · 2026-09-11

  This paper benchmarks code quality differences between human-written and AI-generated code at scale, using 787,562 function pairs across Python, Java, and C produced by GPT models, DeepSeek-Coder, and Qwen2.5-Coder. The study finds that AI-generated code is structurally simpler—roughly half the size and branching complexity of human code—and stylistically more templated, while defect and vulnerability profiles differ by author and language: LLMs produce more and more severe security findings in Python and Java, but fewer high-severity memory-safety issues in C. Naturalness, rather than complexity metrics, is the key differentiator once code size is controlled for. The authors release CQBench, a benchmark of 27,346 issue-prone tasks with evaluation pipelines for quality assurance and security testing, providing a systematic tool for assessing AI coding assistants beyond functional correctness.

  • Quality assurance
  • Enterprise

• Research

  Understanding Game Coaching on Gig Platforms

  Hwijoon Lee, Saiph Savage

  arXiv (Cornell University) · 2026-09-11

  This study examines how freelance game coaches operate on gig platforms like Fiverr, drawing on semi-structured interviews with 20 experienced coaches across 17 competitive games. Despite no shared formal training, coaches converged on common practices around rapport-building, individualized diagnosis, and adaptive feedback. The research identifies two structural challenges unique to this workforce: 'dual precarity' from both gig platform instability and the volatile lifecycles of live-service games, and 'earned authority' requiring coaches to continuously legitimize themselves through visible competitive achievement. Coaches welcomed AI for administrative and analytic tasks but resisted it in live coaching interactions where trust and relational judgment were seen as essential.

  • Workforce

• Research

  When Does AI Augment Work? A Workflow-Level Framework for Human-Agent Collaboration

  CIVIC-AI Collaboration, :, Jiaying Wu et al.

  arXiv (Cornell University) · 2026-09-11

  This paper proposes a workflow-level framework for evaluating when AI genuinely augments human work, rather than simply automating discrete tasks. The authors define AI augmentation through six conditions covering durable net value, meaningful human control, accountability and recovery, and long-term human development including learning, career pathways, and job purpose. They apply this framework to a case study of AI-mediated social surveys and offer guidance for organizations, researchers, and policymakers on understanding the future of work. The framework matters because it shifts the focus from narrow automation metrics to the broader transformation of work through human-agent collaboration.

  • Workforce
  • AI policy

• Research

  Operational Range Bounding in Spectroscopy: A Safety Cage Framework for Machine Learning Models

  Nikki Grens, Luís F. Simões, Kai Hou Yip et al.

  arXiv (Cornell University) · 2026-09-11

  This paper proposes a modular 'safety cage' framework that wraps around black-box machine learning models used in exoplanetary spectroscopy to detect unreliable predictions without modifying the underlying model. The framework combines uncertainty quantification, out-of-domain detection, and influence functions as parallel monitoring indicators, and shows that rejecting only 20% of predictions reduces error by 45–65% across different domains and metrics. The study demonstrates that no single indicator captures all failure modes, making indicator fusion essential. This work advances quality assurance for ML models deployed in safety-critical scientific settings where ground truth is rarely available.

  • Quality assurance

• Research

  Beyond Establishing the Four-Day Workweek: Understanding Adaptation and Long-Term Survival in an Agile Software Organization

  Michael Neumann, Darja Šmite

  arXiv (Cornell University) · 2026-09-11

  This longitudinal case study follows a software organization operating a 32-hour, four-day workweek (4DWW) across interviews conducted in 2022 and 2026, examining how the arrangement was introduced, adapted, and sustained over time. The research finds that the 4DWW is better understood as an evolving arrangement than a one-time intervention, with teams redesigning agile practices, coordination, and meetings to fit reduced hours. When faced with ownership changes, economic downturns, and AI market pressures, employees engaged in voluntary protective adaptations rather than reverting to a five-day week, anticipating that rollback would harm job satisfaction and employer image. The paper contributes a lifecycle model and a survival matrix explaining how external pressures and management rationale interact to determine rollback risk.

  • Workforce
  • Enterprise

• Research

  Testing Between the Test Cases: Proving End-to-End Steering in Conditions You Never Drove

  Menuka Ghalan, Charles Rodgers, Zachary D. Asher

  arXiv (Cornell University) · 2026-09-11

  This paper applies formal verification (bound propagation) to end-to-end neural network steering models trained in the CARLA simulator, proving mathematically how far steering outputs can drift under unseen disturbance conditions—such as fog, night, and low sun—without requiring additional simulation runs. By reading trained network weights directly, the method covers combinatorial spaces (e.g., 10^133 pose-disturbance combinations on an arterial road) in minutes on a single GPU, far exceeding what simulation testing alone could achieve. Crucially, formal verification identified failure conditions for a clear-weather-only policy that simulation testing missed, offering preliminary evidence of failures 'between the test cases.' The work demonstrates a scalable approach to safety assurance for autonomous steering systems under distribution shift.

  • Quality assurance
  • Certifications

• Research

  PIA-Bench: Towards Automated Privacy Impact Assessment with Large Language Models

  Jiamin Zheng, Hao-Ping Lee, Luo Mai et al.

  arXiv (Cornell University) · 2026-09-11

  This paper introduces PIA-Bench, the first open benchmark for evaluating large language models on privacy impact assessments (PIAs) drawn from 499 expert-authored PIAs published by US federal agencies, with 73 structured PIAs curated for evaluation. The authors find that off-the-shelf LLMs can produce meaningful privacy risk and mitigation assessments, while also identifying gaps that require domain-specific workflows, accountable LLM infrastructure, and new PIA quality standards. The work is directly relevant to automating and scaling a compliance process that currently demands scarce privacy and technical expertise.

  • AI policy
  • Quality assurance

• Research

  NovaFabric: Tamper-Evident, Replayable Evidence for Autonomous AI Agent Runs

  Mohsen Seyedkazemi Ardebili

  arXiv (Cornell University) · 2026-09-11

  NovaFabric is a system for producing tamper-evident, replayable audit records of autonomous AI agent runs without modifying the agent itself. It packages an agent's execution into a portable 'Run Capsule' sealed with cryptographic signatures, timestamps, and Merkle logs, then exports an Evidence Bundle that third parties can verify using standard tooling. The paper evaluates eight research questions and finds that mocked replay works reliably (10/10), tampering is rejected across tested classes, and credential redaction covers 14/14 credential types, though tool-using workload replay and stream completeness have notable gaps. This matters because emerging regulations like the EU AI Act and ISO 42001 presume that independent parties can inspect records of AI actions, and NovaFabric directly addresses that audit infrastructure need.

  • Certifications
  • AI policy

• Research

  Voice AI in Firms: A Natural Field Experiment on Automated Job Interviews

  Brian Jabarian, Luca Henkel

  CESifo · 2026-09-11

  This natural field experiment with 70,000 real job applicants randomly assigned to human recruiters or AI voice agents finds that AI-interviewed applicants are 12% more likely to receive job offers, with downstream gains in job starts and worker retention and no productivity decline among hires. Analysis of interview transcripts shows AI agents produce more structured, consistent interviews while still adapting to individual applicants, yielding higher-quality, decision-relevant information. The study argues that AI's key advantage emerges in settings where many human workers execute the same task repeatedly, because human variance introduces noise that AI reduces through adaptive standardization.

  • Workforce
  • Enterprise

• Research

  ORQA: An Occupation-Realistic Question and Answer Framework for LLM Professional Knowledge

  Shreyas Krishnan, Serina Chang, Abhishek Nagaraj

  arXiv (Cornell University) · 2026-09-11

  ORQA introduces a benchmarking framework that evaluates large language models on occupation-specific professional knowledge by linking O*NET occupations to trusted sources such as regulatory agencies, licensing bodies, and government publications, then converting that content into traceable question-answer pairs. The benchmark spans 116 occupations across all 21 major SOC groups, with 480 questions drawn from 187 websites. Top frontier models (Claude Opus, GPT-5, Claude Sonnet) score only 58–62%, while smaller open-weight models reach 33–41%, with near-zero performance on some occupations like Sheet Metal Workers and Fish and Game Wardens. The findings highlight significant gaps in LLM professional knowledge and offer a scalable, source-grounded method for evaluating AI readiness across occupational domains.

  • Workforce
  • Certifications

• Research

  From Review to Reuse: How Post-Task Workflow Can Support Human-AI Agent Interaction

  Zekun Wu, Xinru Wang, Rock Yuren Pang et al.

  arXiv (Cornell University) · 2026-09-11

  This paper investigates 'post-task workflows'—editable, graph-based representations of an AI agent's completed execution—as a way to help users understand, validate, and reuse automated multi-step processes. The researchers analyzed over 10,000 real-world workflow templates and built a tool called Trace2Flow, then tested it with 20 participants reviewing agent executions containing errors. They found that post-task workflows improved user understanding and error detection compared to reviewing only the original prompt, and that adapting a workflow for follow-up tasks was as effective and often preferred over rewriting the prompt.

  • Workforce
  • Enterprise

• Research

  How Do Data Collection Strategy and Data Quality Influence the Outcomes of Digital Technology Adoption?

  Xuejiao Li, Cheng Yang

  arXiv (Cornell University) · 2026-09-11

  Drawing on survey data from 86 manufacturing firms, this study uses PLS-SEM to show that data collection strategy and data quality jointly drive successful digital technology adoption. Data quality fully mediates the relationship between data collection strategy and implementation performance, meaning upstream data practices affect digital outcomes only through the quality of data they produce. The findings offer practical guidance for managers designing data governance frameworks, and call for future research integrating quality assurance, standardization, and lifecycle management into data-driven transformation strategies.

  • Enterprise
  • Quality assurance

• Research

  Human Touch vs. Tech Talk: Mental Health Clinicians' Perspectives on Artificial Intelligence in Systemic Practice

  Merranda Marin, Marcel Montañez, Kourtney Vaillancourt

  Journal of Marital and Family Therapy · 2026-09-11

  This mixed-methods study surveyed 89 U.S. mental health professionals about their familiarity with and attitudes toward AI tools in clinical practice. About half reported using AI professionally, most commonly for notetaking and transcription, and respondents expressed conditional acceptance—valuing AI for clinician support tasks while raising concerns about confidentiality, professional judgment, and the erosion of human connection. The findings highlight a pressing need for ethically grounded, relationally informed frameworks for AI integration in mental health care.

  • Workforce
  • AI policy

• Research

  Responsible Algorithmization in the Public Sector: A “Natural Perspective” Based on Ethnographic Research in Regulation, Policing, and Healthcare

  Albert Meijer, Lukas Lorenz, Isabelle Donatz-Fest et al.

  Public Administration · 2026-09-11

  This paper examines how algorithms are actually adopted in Dutch public-sector organizations across regulation, policing, and healthcare, using ethnographic fieldwork. It finds that algorithmization in practice is emergent, political, open, and fragmented—contrasting sharply with the rational, pre-structured frameworks promoted by bodies like the EU and OECD. The authors argue that existing responsible AI frameworks focus too heavily on passive responsibility (compliance with formal rules) and must also address active responsibility—how organizations navigate messy, real-world decision-making around algorithmic tools. The findings have direct implications for how governments design and govern AI policy in public institutions.

  • AI policy

• Research

  Current Standards of Monitoring Models in Healthcare Settings

  Alan Kay, Daljit Takher, Wenting Liu et al.

  Advanced Intelligent Discovery · 2026-09-11

  This paper examines the regulatory and operational gap between FDA-approved AI/ML clinical devices and the broader landscape of healthcare AI research, finding that only 3.4% of approved models have predetermined update plans. The authors argue that limited trust, verifiability, and inadequate monitoring frameworks are major barriers to broader clinical deployment of AI, particularly deep learning methods. They review current device monitoring frameworks and identify key challenges for implementing model monitoring in high-risk medical settings.

  • Certifications
  • Quality assurance

• Research

  Smart contracts and AI agents for secure last-mile pharmaceutical delivery through crowdsourcing

  Kadim Lahcen Nadime, Doha Haidar, Rajaa Benabbou et al.

  Discover Artificial Intelligence · 2026-09-11

  This paper proposes a ledger-grounded AI agent architecture for crowdsourced pharmaceutical last-mile delivery, where blockchain smart contracts serve as the authoritative record for identity, assignments, and delivery evidence while AI agents handle forecasting, routing, carrier selection, and dispatch using verified on-chain data. Evaluated against simulated human-operated workflows across 1,600 orders, the agent-assisted system reduced route time by 25.0% and planning time by 78.6%, with improvements in urgent-order handling and proof completeness. The findings suggest that combining smart contracts with AI agents can enhance operational efficiency in pharmaceutical distribution without sacrificing traceability or accountability, though field validation remains needed.

  • Enterprise
  • Quality assurance

• Research

  Intention to Use Generative AI for Vocational College Administration

  Porntep Chooppawa, Potsirin Limpinan, Thada Jantakoon

  World Journal of Education · 2026-09-11

  This study surveyed 330 administrators at private vocational colleges in Thailand to identify what drives their intention to use and actual use of Generative AI in administrative tasks. Using an extended UTAUT model with trust, privacy concern, and institutional policy fit added, the research found that social influence, trust, privacy concern, and policy alignment were the key predictors of adoption intent, explaining 82.8% of the variance in behavioral intention and 76.1% in actual use behavior. Notably, traditional factors like perceived usefulness and ease of use were not significant, suggesting that governance and policy considerations outweigh technical perceptions in this context. The findings carry direct implications for institutional AI governance, policy design, and responsible AI implementation in vocational education.

  • AI policy
  • Workforce

• Research

  Competence, Career, and Compliance: An Integrated Framework Aligning Teacher AI Literacy with National Performance, UNESCO Standards, and Data-Protection Law in Thailand

  Paritchaya Sarakan, Anucha Somabut, Lan Thi Nguyen et al.

  Journal of Education and Training Studies · 2026-09-11

  This paper develops an integrated framework aligning teacher AI literacy competencies with Thailand's national teacher performance and career systems, UNESCO's 2024 AI Competency Framework for Teachers, and the country's Personal Data Protection Act. Using policy document analysis and crosswalk synthesis, the authors produce an alignment matrix, developmental progression map, data-governance overlay, and phased implementation roadmap. An expert panel (n=5) rated the framework favorably (mean=4.68) with moderate inter-rater concordance (Kendall's W=0.42). The study argues that embedding AI literacy professional development within existing accountability and legal structures is more likely to achieve durable adoption than standalone training programs, and suggests the design logic may transfer to comparable jurisdictions.

  • Workforce
  • AI policy
  • Certifications

• Research

  Video-to-report generation for cataract surgery using procedurally grounded vision–language models

  Tao Yu, Kaikai Zhao, Vitalii Prudyus et al.

  Communications Medicine · 2026-09-11

  LensNarrate is a vision-language AI system that converts unedited cataract surgery videos into temporally structured operative reports with visual evidence. Tested on videos from European and Chinese hospitals, it achieved internal temporal frame accuracy of 83.2% and segmental F1@50 of 58.0%, substantially outperforming a retrieval baseline, though cross-site performance was lower. The system addresses the problem of retrospectively written operative reports that may miss short events or use inconsistent descriptions, and the authors identify cross-site variation and short-phase boundary localization as remaining challenges before clinical adoption.

  • Quality assurance

• Research

  Justice by humans, assisted by AI

  Asif Khan, Aftab Haider, Asif Salim

  Oñati Socio-legal Series · 2026-09-11

  Examining 47 statutes and judgments, 142 studies, and seven datasets across five jurisdictions from 2015–2025, this article identifies three systemic faults when AI tools—including machine learning, large language models, and online dispute resolution—are used in bail, sentencing, and legal pleadings: opacity of proprietary models, the substitution of statistical prediction for case-specific explanation, and the concentration of error risk on the least-resourced litigants. The authors propose a duty to explain AI-assisted decisions, pre-deployment legal impact assessments, and an audit standard for fair process, arguing that the goal is not to exclude AI from courts but to preserve human judgment at the center of adjudication. The paper is directly relevant to certification and policy frameworks governing AI in public legal institutions.

  • AI policy
  • Certifications

• Research

  ProofLayer: Control Coverage Screening for Government Cybersecurity Policies via Multi-LLM Orchestration

  Devharsh Trivedi

  Journal of Cybersecurity Digital Forensics and Jurisprudence · 2026-09-11

  ProofLayer is a two-stage system that screens government cybersecurity policy documents against a 22-control catalogue crosswalked to NIST SP 800-53 Rev. 5, using keyword matching followed by LLM-based resolution. Evaluated on ten Maryland state and local government policy documents, the system finds a mean control coverage of 62.8% when documents are read in full, compared to 49.1% when truncated at fifteen pages—a gap that itself reveals how sensitive screening metrics are to extraction depth. The paper identifies notable failure modes including negation cues near matches (18.1% of covered pairs) and single-keyword dependencies, and honestly distinguishes what is measured from what is only described, declining to use model-generated labels as ground truth. This work matters for policy and quality-assurance practitioners who need transparent, auditable tools for assessing cybersecurity policy coverage at scale.

  • AI policy
  • Quality assurance

• Research

  Technostress and Productivity: Evidence from Business Process Outsourcing (BPO) Workers

  Fredrick Michael Ogore, Patrick Kanyi Wamuyu, Gerald Chegee

  African Journal of Commercial Studies · 2026-09-11

  This study of Kenya's Business Process Outsourcing (BPO) sector finds that technostress significantly reduces worker productivity, with primary technostress factors negatively correlated with productivity (r = -0.403) and explaining 16.2% of variance. Technology insecurity—driven by fears of job displacement from automation and AI—was identified as the most severe stressor, followed by technology overload, complexity, invasion, and uncertainty. The findings call on BPO organizations to implement comprehensive, multi-dimensional mitigation strategies including transparent communication about AI plans, career development pathways, and reskilling programs. The study matters because it provides quantitative and qualitative evidence that AI-driven workplace change poses measurable mental health and performance risks for a large segment of knowledge workers in an emerging economy.

  • Workforce
  • Enterprise

• Research

  Cognitive reshaping and resurgence of humanness: restructuring the medical education continuum in the era of generative AI

  Ya Liu, Lutuo Han, Linlin Che et al.

  Frontiers in Medicine · 2026-09-11

  This narrative review examines how generative AI is reshaping medical education by introducing three cognitive vulnerabilities: deskilling among advanced learners, never-skilling in junior trainees who fail to build foundational mental models, and automation bias from uncritical reliance on AI outputs. The authors propose restructuring the full medical education continuum—undergraduate, graduate, and continuing medical education—to prioritize pathophysiological reasoning, human-AI collaboration, and deliberate reflection, while shifting assessment away from memory-based multiple-choice questions. The paper argues that medical education should cultivate 'augmented clinicians' with high AI literacy and humanistic competencies, framing the goal as ensuring AI enhances rather than displaces relational patient care.

  • Workforce
  • Certifications

• Research

  Artificial Intelligence Readiness in Emergency Medicine: Expert Consensus Opinion for Preparing the Workforce

  Debadutta Dash, Joyce Macalalad, Donald L. Lum et al.

  Journal of the American College of Emergency Physicians Open · 2026-09-11

  This expert consensus opinion from the American College of Emergency Physicians (ACEP) AI Task Force, formally endorsed by the ACEP Board of Directors in 2025, addresses the gap between rapid AI adoption in emergency departments and clinicians' ability to safely evaluate and oversee these tools. The paper proposes three coordinated priorities: a standardized AI education framework spanning residency through continuing medical education, a structured three-stage clinician-led framework for vetting AI tools before deployment, and the creation of a national Emergency Medicine AI Advisory Council to provide shared terminology and best-practice guidance. Together, these recommendations aim to make AI integration in emergency medicine safe, equitable, and clinically effective across academic, community, and critical access settings.

  • Workforce
  • AI policy

← Back to the dashboard