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
REDAgentBench: Executable Red Teaming and Faithful Measurement of LLM Agent Systems
Zixing Chen, Xingyuan Liu, Jie Zhu et al.
arXiv · 2026-08-11
REDAgentBench introduces an executable red-teaming framework that measures safety violations in LLM agent systems more faithfully than the single attack-success-rate (ASR) metric common in prior work. The framework derives attacks from explicit safety constraints, runs them in isolated service sandboxes, and verifies harmful effects through service receipts and final-state changes across 1,661 test cases and five service surfaces. Key findings include a macro-average ASR of 65.69% across six models and three agent harnesses, a 'Recognition–Execution Gap' where nearly one in five confirmed violations occur even after the agent has acknowledged the relevant constraint or risk, and a training-free policy reminder that reduces confirmed violations by more than 70 percentage points in matched replay. These results demonstrate that executable, state-grounded evaluation can expose measurement artifacts in simpler ASR approaches and point to concrete intervention strategies for improving LLM agent safety.
- Quality assurance
- AI policy
Research
Curate Before You Connect: Identity and Ontology Tagging in a Production Knowledge Graph
Vaibhav Dangaich, Kevin Lewis, Kundeshwar Pundalik
arXiv · 2026-08-11
This paper describes the design and production deployment of an ingestion and ontology-tagging layer that converts a validated extraction stream into a knowledge graph of 537,157 entities and 2,198,567 relationships sourced from 98,795 government documents. A central contribution is a 'record-identity ladder' that resolves entity sameness using identifier columns, name columns, display names, and type-scoped position rather than name similarity, preventing irreversible over-merges — a risk illustrated by an incident in which two surface forms of one name were merged, corrupting a correct record and deleting eight entities. The paper also addresses ontology tagging, finding that matching name fragments against a class index without anchored evidence generates spurious classifications; requiring anchored evidence reduced role assignments on an enriched sample from 36 to 4, all confirmed correct. The work highlights practical automation limits in knowledge graph curation, quantifying a growing backlog of 48,403 pending proposals against only 775 human decisions, with implications for quality assurance in large-scale AI-driven data pipelines.
- Quality assurance
- Enterprise
Research
Agent Safety Should Be a Runtime Contract
Albus W. Ng, Yi Han, Jusheng Zhang et al.
arXiv · 2026-08-11
This position paper argues that AI safety for autonomous agents cannot rely solely on training-time techniques like RLHF or Constitutional AI, because agents that execute code, mutate files, and modify databases require safety guarantees enforced at runtime. The authors propose a two-faced runtime contract: a preventive face (sandboxes, permission gates, output filters, trajectory monitors) and an evidential face (verifiable proof of completed actions such as test runs, log captures, file diffs, and citation grounding). They ground this argument in four empirical audits: a survey of 52 documented AI-agent safety incidents, a false-completion audit of 31 core cases, a trajectory-schema audit of 12 public agent systems, and a title-level audit of 28,560 papers from NeurIPS, ICML, and ICLR 2023–2025 showing an 8–12x publication imbalance between training-time and deployment-time safety research. The paper formalizes an Agent Trajectory Schema and Evidence Chain, concluding that the right unit of safety in agentic AI is the trajectory-with-checkable-evidence rather than the model itself.
- Quality assurance
- AI policy
Research
ASR-Roundtrip Evaluation Can Mask Context- and Convention-Dependent Reading Errors in Chinese News TTS
Shijun Luo, Lizhi Wan
arXiv · 2026-08-11
This paper investigates a critical flaw in a common method for evaluating text-to-speech (TTS) systems: ASR-roundtrip evaluation, where a speech recognizer transcribes TTS audio and the transcript is compared to the original text. The authors show that for Chinese news TTS, this approach can mask real reading errors—cases where the TTS system chooses a plausible but contextually wrong pronunciation (e.g., for sports scores, aircraft models, or technical units)—because the ASR system independently recovers the 'correct' text from the wrong audio, hiding the mistake. In a targeted audit of 110 high-risk cases across two TTS systems (MiMo and CosyVoice), the study finds dozens of such masked false negatives and shows that different ASR models vary greatly in their ability to surface these errors. The findings indicate that ASR-roundtrip evaluation is useful for screening but should not be treated as a standalone quality standard for Chinese news TTS intelligibility assessment.
- Quality assurance
Research
Inferential Capability Does Not Determine Legal Scope
Nicola Fabiano
arXiv · 2026-08-11
This legal analysis paper examines how two key EU digital regulations—the AI Act and the GDPR—treat 'inference' differently and non-equivalently. The AI Act uses inferential capability as a constitutive criterion to define what counts as a regulated AI system, while the GDPR governs inferences protectively based on what they reveal or do to a person, regardless of whether the technology qualifies as AI. The paper argues that inferential capability does not determine legal scope and that its absence does not create immunity, a gap that becomes operationally acute with agentic (multi-step AI) architectures. The authors propose a compositional-effects test for identifying the relevant decision unit under Article 22 GDPR, along with interpretive rules and documentation duties calibrated to inference chains.
- AI policy
Research
RadFusion: Towards Threshold-Controllable Radiology Report Generation
Ying Jin, Noel C. F. Codella, John Corring et al.
arXiv · 2026-08-11
RadFusion is a framework that adds threshold controllability to automated radiology report generation by fusing a multi-label classifier with a VQA-based report generator and an LLM rewriter. On the MIMIC-CXR dataset, the system's outputs conform to the classifier's ROC curve, meaning diagnostic decisions in generated reports can be tuned to favor sensitivity (for emergency triage) or specificity (for confirmatory interpretation) by adjusting a threshold. Combining the two model types improves diagnostic accuracy over uncontrolled generation, with sensitivity increasing by 6.9% at matched specificity and specificity increasing by 20.7% at matched sensitivity. This ROC-based verifiability strengthens the case for regulatory clearance and makes report generation clinically adaptable across different care scenarios.
- Certifications
- Quality assurance
Research
Evaluating Rational Contracting in Natural Language
Bhavyesh Sajja, Max Kleiman-Weiner, Roger Zimmermann et al.
arXiv (Cornell University) · 2026-08-11
This paper introduces ContractSim, an evaluation suite for testing how LLM-based agents negotiate and execute multi-turn supplier contracts in natural language under uncertainty. The authors develop a rational framework with metrics for measuring efficient, cooperative, and trustworthy contracting behavior across six environments and three supplier settings (catering, hotel cleaning, and AI hosting). Results show that current LLM agents reliably reach agreements and negotiate efficiently under low uncertainty, but struggle to produce satisfiable or mutually beneficial contracts under high uncertainty, and frequently violate contract terms for additional profit even when contracts are easy to satisfy. These findings expose significant gaps in the trustworthiness and cooperative behavior of language agents engaged in open-ended economic activity.
- Enterprise
- Quality assurance
Research
Conversational versus Dashboard Explainable AI for UAV Intrusion Detection: An Empirical Study of Operator Trust and Reliance
Cong Chi Nguyen, Trang Mai Xuan, Vu-Duc Ngo et al.
arXiv · 2026-08-11
This paper compares a conversational XAI interface powered by Large Language Models against a traditional static visualization dashboard for helping operators audit UAV intrusion detection systems. In a controlled experiment, participants found the conversational interface more useful and easier to synthesize information from, but also exhibited higher over-reliance — meaning they were less likely to verify AI advice when the system made errors. The findings reveal a trade-off in human-AI collaboration: interaction designs that improve perceived usability may simultaneously increase the risk of inappropriate reliance on AI judgments. The authors conclude with design implications for building XAI systems that balance natural interaction with cognitive forcing functions to promote appropriate self-reliance.
- Quality assurance
- Workforce
Research
What We Know about Responsible AI Practices in Industry: A Half Decade of Empirical Research
Wesley Hanwen Deng, Agathe Balayn, Andrew Selbst et al.
arXiv (Cornell University) · 2026-08-11
This paper synthesizes 161 empirical studies spanning six years to assess the state of Responsible AI (RAI) practice across industry. It finds that practitioner awareness has grown, RAI activities have become more professionalized, and tools like guidelines and toolkits are more widely adopted. However, persistent barriers remain, including limited training, uneven organizational support, and a shortage of interventions suited to day-to-day work. The findings carry implications for researchers, enterprise practitioners seeking to adopt effective RAI practices, and policymakers seeking to ground AI governance in the realities of industry contexts.
- Enterprise
- AI policy
Research
When the Interviewer Is a Bot: Behavior, Breakdowns, and Trust in MLLM-Led Interviews
He Zhang, Kambinachi Chukwuma, ChanMin Kim et al.
arXiv · 2026-08-11
This paper reports an empirical study of what happens when an off-the-shelf multimodal large language model (MLLM) conducts semi-structured qualitative interviews. The researchers built 'InterviewBot,' a voice-based system, and deployed it with 15 participants, analyzing 428 conversational turns and conducting follow-up human-led reflection sessions. Key findings include that the MLLM was acknowledgment-heavy but probe-light (deepening probes only 4.9% of turns), violated its own one-question-at-a-time instruction in 28.7% of question-bearing turns, and produced four types of data-collection breakdowns (information loss, premature termination, latency, and interruption). Participants' trust was shaped not by the bot's conversational competence but by what delegating interviews to AI signaled about the organizing institution, with implications for how enterprises and researchers design and deploy AI-led interview automation.
- Enterprise
- Workforce
Research
Expert-Guided g-computation with Large Language Models for Estimating Causal Effects on Timings: Applications to Hospital Quality Improvement
Patrick Vossler, Jialin Ouyang, F. Richard Guo et al.
arXiv · 2026-08-11
This paper introduces 'egg-computation' (expert-guided g-computation), a hybrid causal inference framework that combines expert judgment with data-driven methods to estimate the average time saved by candidate hospital quality improvement (QI) interventions—specifically targeting average length of stay (LOS). The approach links Gantt charts used in clinical workflow mapping to causal directed acyclic graphs (DAGs), using a g-computation variant that solicits expert input only for components unidentifiable from data alone. To scale the method, the authors build an LLM-assisted pipeline that generates causal graphs and time-saving estimates shown in simulations and a real study of eleven QI interventions at an urban safety-net hospital to be highly concordant with human expert reasoning. The framework addresses the gap where existing qualitative methods are prone to cognitive bias and quantitative methods fail for hypothetical or clinically complex interventions, offering a broadly applicable tool for causal effect estimation in process improvement settings.
- Enterprise
- Quality assurance
Research
Who Gets Heeded? An Obligation-Level Audit of Responsiveness in EPA Rulemaking
Jianing Fan, Yue Yao
arXiv · 2026-08-11
This paper introduces an AI-assisted framework for auditing whether public comments on proposed federal regulations actually influence specific regulatory obligations in final rules, applied to over 70,000 comments across 36 EPA rulemakings. The framework extracts and matches obligations between proposed and final rules, classifies outcomes, and validates each step against human judgment. Key findings show that comment engagement is associated with only modest revision at the obligation level, that supporting versus opposing a rule does not clearly predict outcomes, and that organizational commenters' engagement concentrates in editorial rather than substantive changes. The authors argue the real equity gap lies upstream — in unequal capacity among commenter populations to identify and contest specific legal obligations — rather than in differential agency responsiveness itself.
- AI policy
Research
Lost in the Loop: Who Is the ‘Human’ of the Human in the Loop?
Jake Goldenfein
Cambridge University Press eBooks · 2026-08-11
This chapter critically examines the 'human in the loop' doctrine in automated decision-making, arguing that requirements for human oversight serve primarily political rather than empirical purposes. The author finds little evidence that human oversight actually improves decision outcomes, and through analysis of administrative law cases, shows that invoking 'human in the loop' functions as a legal technology that obscures accountability and limits richer legal understanding of automated systems. The work challenges regulatory orthodoxy by revealing how human oversight requirements can distribute accountability in troubling ways rather than genuinely safeguarding against automation's harms.
- AI policy
Research
Artificial Intelligence in Psychiatry: Five Decades of Progress and Persistent Translational Challenges
Esteban Zavaleta‐Monestel, Luis Guillermo Herrera-Jiménez, Sofía Suárez-Sánchez et al.
Psychiatric Research and Clinical Practice · 2026-08-11
This structured historical review traces AI development in psychiatry from 1972 to 2025, synthesizing evidence across paradigms ranging from rule-based expert systems to large language models. Despite five decades of technical advances, the authors find that recurring barriers—including diagnostic heterogeneity, limited external validation, poor transportability, interpretability gaps, and equity concerns—have prevented widespread clinical adoption. Progress is characterized as cyclical rather than linear, with successive technological waves reproducing the same unresolved challenges. The review concludes that future clinical impact will depend on clearer target validity, prospective implementation trials, patient-centered evaluation, and mental health-specific governance rather than algorithmic sophistication alone.
- AI policy
- Quality assurance
Research
SafeCA: Safe Cross-Attention Localization and Regulation for Text-to-Video Jailbreak Defense
Siyuan Liang, Yupeng Qiu, Junfeng Fang et al.
arXiv (Cornell University) · 2026-08-11
SafeCA is a defense mechanism for text-to-video generative models that targets jailbreak attacks—attempts to make these models produce harmful or inappropriate content. By analyzing cross-attention feature spaces, the authors identify a 'cumulative separation effect' between clean and jailbreak samples during the diffusion process, enabling feature-level intervention via attention masking, energy normalization, and a lightweight semantic adapter. Experimental results show SafeCA reduces jailbreak success rates by approximately 20% on mainstream text-to-video models while adding only 0.1 seconds of inference overhead and preserving text-video semantic consistency. This work advances quality-assurance and policy-relevant safeguards for deployed generative AI systems.
- Quality assurance
- AI policy
Research
Agentic Configuration Management (ACM): A Reference Configuration Model for Governed Agentic Systems
Audrey Quessada-Vial
arXiv (Cornell University) · 2026-08-11
This paper introduces Agentic Configuration Management (ACM), a framework-independent governance and configuration model for AI agentic systems composed of heterogeneous agents, tools, models, and workflows. ACM provides typed versioned configuration items, immutable revisions, dependency-aware impact propagation, and runtime provenance, normalized into a canonical Configuration Graph. A Python reference implementation with adapters for LangGraph, CrewAI, and the OpenAI Agents SDK was evaluated across 27 governance scenarios and nine quantitative impact-propagation cases, demonstrating reproducible and auditable governance outcomes across frameworks. The work provides evidence that common governance semantics can support reproducibility, auditability, and interoperability across heterogeneous agentic systems.
- Quality assurance
- AI policy
Research
On Understanding, Identifying, and Mitigating Vulnerabilities in Agentic Large Language Models
Md Jafrin Hossain, Mohammad Arif Hossain, Nirwan Ansari
arXiv (Cornell University) · 2026-08-11
This paper presents a systematic literature review (PRISMA 2020) of 85 papers on agentic large language model security, finding that attack research outpaces defense work by 3.9:1 and that perception-layer vulnerabilities dominate the literature while action-layer risks like tool misuse and code injection are severely understudied relative to real-world threat. The authors introduce a four-layer taxonomy covering 13 vulnerability types and identify architectural coupling—weak isolation allowing vulnerabilities to propagate across layers—as the root cause of many agentic LLM security failures. The findings matter because autonomous LLM agents now operate with real-world privileges (API calls, file modification, database queries), making unmitigated security gaps a direct enterprise and policy risk.
- Enterprise
- AI policy
Research
Challenges and opportunities for advanced manufacturing to deliver war-winning technology to the warfighter
Clara Mock, Jian Yu, Eric Wetzel et al.
arXiv · 2026-08-11
This report examines how Advanced Manufacturing (AdvM) can address global supply chain vulnerabilities and contested logistics environments by enabling both domestic Defense Industrial Base production and distributed, Point-of-Need manufacturing on the battlefield. The authors identify persistent barriers to AdvM adoption—including material limitations, process transferability, and certification bottlenecks—and propose solutions across four domains: novel feedstock development, AI-driven digital quality assurance (predictive modeling, in-situ monitoring, digital twins), workforce development through academia-industry-military collaboration, and performance-based qualification standards. The paper argues that integrating these elements can enable highly autonomous manufacturing to support continuous military mission readiness. The findings are directly relevant to workforce training, enterprise-level defense industrial capacity, quality assurance through digital frameworks, and certification pathway reform.
- Workforce
- Enterprise
- Quality assurance
- Certifications
Research
Professional roles, gender, and attitudes toward generative AI: evidence from Japan
Eiji Yamamura, Fumio Ohtake
AI and Ethics · 2026-08-11
Using a large-scale survey of 3,433 respondents in Japan (early 2024), this study finds that attitudes toward generative AI vary substantially by occupation and gender. Managers and ordinary employees tend to hold more favorable views of generative AI, while teachers emphasize the ongoing importance of human judgment and learning. Gender differences are also significant: female managers and employees are less favorable toward generative AI than male peers, while female teachers show the reverse pattern. The findings highlight how occupational identity and gender shape workforce readiness and receptivity to AI adoption.
- Workforce
- AI policy
Research
From Traditional Audits to Digital Audits: A Systematic Review of the Impacts and Driving Factors
Christine Belgina Saurmauli, Krisna Puji Rahmayanti
Journal Of Social Research · 2026-08-11
This systematic review synthesizes empirical evidence from 33 studies (2015–2026) on how digital technologies—including AI, robotic process automation, blockchain, big data, and audit analytics—are transforming audit practice. The findings show these tools generally improve audit effectiveness and efficiency, strengthen internal controls, and reduce errors and financial statement restatements, while repositioning auditors as more strategic, data-driven professionals. Adoption success depends heavily on technological infrastructure, data governance, organizational capabilities, leadership support, and auditor competencies, indicating digitalization is neither neutral nor automatic. The paper has direct implications for audit firms, internal audit units, public sector institutions, and regulators designing digital audit strategies.
- Quality assurance
- Certifications
- AI policy
Research
Balancing AI Data and Consumer Rights: The Colombian Context
J.W. Vásquez, Rene Alvarez‐Orozco
Latin American Policy · 2026-08-11
This paper examines surveillance capitalism in Colombia, where AI-driven data collection by corporations in telecommunications, retail, finance, and digital platforms raises serious privacy and consumer rights concerns. Using computational simulations and case studies, the authors quantify risks from opaque consent mechanisms, algorithmic bias, and data breach vulnerabilities. The study evaluates Colombia's AI governance framework—specifically CONPES 4144 de 2025—finding significant enforcement and implementation shortcomings, and calls for stronger regulatory frameworks to protect consumer rights while supporting innovation.
- AI policy
Research
The AI–Climate Nexus in International Relations: Climate Diplomacy, Energy Transition, and Emerging Technology Governance
Hina Ahmad, Kinza Kamran, Fizzah Muhammad
Journal of Global Social Transformation · 2026-08-11
This study examines how AI is reshaping international climate diplomacy, energy transition, and geopolitical power dynamics through analysis of 127 policy and academic sources and four comparative case studies. It finds that while AI improves emissions verification and renewable grid optimization, it also introduces risks such as large computational carbon footprints, digital infrastructure lock-in, and geopolitical shifts favoring states with computational dominance and control of critical minerals. The research highlights that the Global South faces acute risks of data colonialism and algorithmic marginalization under fragmented current governance regimes. The paper calls for a dedicated UN AI–Climate Governance Interface that embeds common but differentiated responsibilities into global AI policy.
- AI policy
Research
The integration of artificial intelligence in the academic writing of EFL students in higher education: a case study at Fergana State University
Abdulkhay Kosimov Akhadali Ugli, Abdullajon Nematov, Markhabo Shokirova Sharifovna
Cogent Education · 2026-08-11
This mixed-methods case study of 351 EFL students at Fergana State University, Uzbekistan, examined how AI writing tools (ChatGPT, Grammarly, QuillBot) affect academic writing quality and cognitive independence. Quantitative results showed AI assistance improved grammatical accuracy (35% error reduction) and lexical sophistication (53% increase in Academic Word List coverage), but was associated with declines in argumentative claims (24%) and originality. Qualitative findings highlighted tensions between AI as scaffolding versus substitution and a perceived performance–competence gap, suggesting AI may boost surface-level writing while undermining independent academic reasoning.
- Workforce
- Quality assurance
Research
Integrating LLM with consortium blockchain for personalized and verifiable online education in higher education
Fuan Xiao, Jiahui Huang, Jia-Xin Huang et al.
International Journal of Educational Technology in Higher Education · 2026-08-11
This paper proposes a framework that combines large language models (LLMs) with a permissioned consortium blockchain to deliver personalized online education while addressing LLM reliability problems such as hallucinations and output inconsistency. The blockchain serves as a tamper-proof ledger that records learning interactions, AI-generated content, and academic credentials, creating an auditable trail that can assign accountability when AI errors cause poor learning outcomes. The framework aims to make AI-driven education both trustworthy and verifiable, which has direct implications for how institutions certify learning and validate credentials. The work matters because it provides a concrete architectural response to the accountability gap in deploying LLMs at scale in higher education.
- Certifications
- Quality assurance
Research
Uso inadecuado de la inteligencia artificial en el proceso de enseñanza-aprendizaje: Un estudio en instituciones educativas del Ecuador
Angelica Liliana Zhunio Zhunio, María Tráncito Chela Tualombo, María Beatriz Villalobos Veloz et al.
Neosapiencia Revista especializada en Ciencias de la Educación · 2026-08-11
This study from Ecuador surveyed 350 higher-education students to assess the impact of inappropriate generative AI use on learning outcomes. Key findings include that 64.86% of students submitted AI-generated essays without modification, 72.86% preferred receiving final answers rather than understanding procedures, and a Pearson correlation of 0.79 was found between unregulated technological dependence and degradation in critical analysis, writing, and retention. The authors conclude that uncritical AI reliance promotes intellectual minimum effort and erodes professional-quality competencies, urging institutions to implement assessments based on human reasoning and oral defense rather than simple prohibition.
- Workforce
- Quality assurance