AI Frontier Institute

News & Research

The latest AI research and news with real-world stakes. Each item is sourced, dated, summarized in plain English and tagged by impact area, and checked against its source before it appears.

• ResearchApplied Sciences2026-04-26EQC

  A Generative AI-Based Framework for Proactive Quality Assurance and Auditing · Galina Ilieva, Tania Yankova, Vera Hadzhieva et al.

  This paper proposes a framework integrating generative AI into manufacturing quality assurance and auditing, covering supplier, in-process, and post-market domains at product, process, and operation levels. A proof-of-concept study in electronics manufacturing compared an expert QA team against a generative AI-assisted chatbot workflow for New Product Introduction planning, producing deliverables including PFMEA, Control Plan, PPAP packages, and audit evidence packs. Independent expert evaluation found that the AI-assisted workflow produced more traceable, governance-rich documentation—including ownership records, versioning, and clause-to-evidence links—and reduced manual audit-evidence consolidation. The findings suggest generative AI can meaningfully support quality planning, change-control readiness, and auditability in manufacturing contexts.
• ResearchInternational Journal of Auditing2026-04-26WQCP

  Auditability in the Digital Age · Michael Power

  This perspective essay argues that auditing is undergoing unprecedented technological transformation as algorithms permeate the systems being audited, forcing auditing itself to become algorithmic. The author warns of risks including 'platformization' of auditing, loss of auditor independence, and auditors becoming de-skilled curators of analytical models rather than independent professionals. The paper calls on scholars, practitioners, and professional institutes to empirically study these changes and safeguard the human-centric and dialogic nature of auditing practice.
• ResearchJournal of Corporate Finance Research / Корпоративные Финансы | ISSN 2073-04382026-04-26EP

  AI-Driven Corporate Sustainability: Exploring the Moderating Role of External Regulation · Yanfei Wu, Irina Ivashkovskaya

  Analyzing Chinese-listed companies from 2013 to 2022, this paper finds that AI adoption significantly improves corporate ESG performance, but the effect is contingent on regulatory context. Both investor attention ('soft' regulation) and environmental regulation ('hard' regulation) strengthen AI's positive contribution to ESG outcomes, though investor attention exhibits a threshold effect where low levels of scrutiny can actually lead AI adoption to harm ESG performance as firms prioritize efficiency over sustainability. The study also finds that AI's impact differs by ownership type, driving social and governance gains in non-state-owned enterprises while concentrating on environmental outcomes in state-owned enterprises. These findings highlight AI's 'double-edged sword' nature and the importance of regulatory frameworks in shaping its sustainability impact.
• ResearcharXiv (Cornell University)2026-04-26EQCP

  Proof of Execution: Runtime Verification for Governed AI Agent Actions · James Rhodes, George Kang

  This paper introduces Proof of Execution (PoE), a formal runtime verification framework for AI agent systems that execute consequential actions rather than just generating advice. PoE bundles an agent's authorization contract, a cryptographically tamper-evident event stream, and a replay context into a single verifiable object, enforcing five invariants covering authorization, path compliance, history integrity, and replayability. A TypeScript prototype demonstrates low overhead (roughly 2.7 ms added latency, 4.4% overhead on batch workloads) and successfully rejects injected bypass and trace-mutation attacks. This matters because it provides a concrete, attestable mechanism for proving that governed AI agents acted within authorized boundaries—directly relevant to enterprise compliance, policy enforcement, and quality assurance of agentic AI deployments.
• ResearcharXiv2026-04-25QP

  Can Humans Detect AI? Mining Textual Signals of AI-Assisted Writing Under Varying Scrutiny Conditions · Daniel Tabach

  This controlled experiment investigates whether warning writers about AI detection changes how they write with AI, and whether human judges can perceive the difference. In a two-phase study, 21 participants wrote opinion pieces with AI assistance, half warned their work would be scanned by a detection tool. Among 251 judges evaluating 1,999 paired comparisons, warned writers' documents were identified as 'more human' 54.13% of the time versus 45.87% for unwarned writers—a statistically significant difference (p = 0.000243). Strikingly, standard text features such as AI overlap scores, lexical diversity, sentence structure, and pronoun usage showed no measurable difference between the groups, suggesting human judges detect subtle signals that automated feature-based methods cannot capture. The findings matter for quality assurance and policy contexts where distinguishing human from AI-assisted writing is increasingly important.
• ResearcharXiv2026-04-25QP

  V.O.I.C.E (Voice, Ownership, Identity, Control, Expression): Risk Taxonomy of Synthetic Voice Generation From Empirical Data · Tanusree Sharma, Anish Krishnagiri, Lili Dudas et al.

  This paper introduces V.O.I.C.E (Voice, Ownership, Identity, Control, Expression), a risk taxonomy for synthetic voice generation built from empirical data across three sources: 569 incidents from major AI incident databases, the FTC, and the IC3; 1,067 direct incident reports from U.S. participants including voice actors, internet personalities, political personnel, and the general public; and 2,221 Reddit discussions. The taxonomy addresses gaps in existing threat models by explicitly capturing how privacy, security, and governance risks emerge from unconsented collection, reuse, and synthesis of voice data, and how those risks interact with contextual factors such as degree of exposure, social visibility, and availability of legal protections. The work is significant because it provides a real-world-grounded framework for understanding differentiated risks across diverse affected groups, which has direct implications for policy, governance, and the protection of individuals whose voices may be exploited by generative AI systems.
• ResearcharXiv2026-04-25WQ

  ArguAgent: AI-Supported Real-Time Grouping for Productive Argumentation in STEM Classrooms · Jennifer Kleiman, Yizhu Gao, Xin Xia et al.

  ArguAgent is a generative AI system designed to support real-time student grouping in STEM classrooms by assessing argumentation quality and stance diversity. The system uses a two-component pipeline that scores student arguments on a 0–4 rubric and clusters positions via semantic analysis, achieving strong agreement with human expert consensus (Krippendorff's α = 0.817 on 200 expert-generated scores). Simulation testing across 100 classes showed that 95.4% of formed groups met both design criteria—stance heterogeneity and constrained quality differences—a 3.2x improvement over random assignment. The findings suggest that AI-assisted grouping can help teachers promote more inclusive, evidence-based classroom discourse at scale, addressing the practical barrier of real-time insight into student reasoning.
• ResearcharXiv2026-04-25EQ

  IndustryAssetEQA: A Neurosymbolic Operational Intelligence System for Embodied Question Answering in Industrial Asset Maintenance · Chathurangi Shyalika, Dhaval Patel, Amit Sheth

  IndustryAssetEQA is a neurosymbolic AI system designed to support industrial asset maintenance by combining episodic telemetry data with a Failure Mode Effects Analysis Knowledge Graph (FMEA-KG) to answer operator questions about asset behavior, failure diagnosis, and intervention evaluation. Evaluated across four industrial asset types—rotating machinery, turbofan engines, hydraulic systems, and cyber-physical production systems—it significantly outperforms large language model (LLM)-only baselines, improving structural validity by up to 0.51, counterfactual accuracy by up to 0.47, and explanation entailment by 0.64. Notably, it reduces severe expert-rated overclaims from 28% to 2%, a roughly 93% reduction, addressing a key trust and safety concern in critical maintenance settings. These results matter because they demonstrate that grounding AI responses in verifiable telemetry and structured knowledge can meaningfully improve reliability and trustworthiness for enterprise industrial operations.
• ResearcharXiv2026-04-25QP

  AI Safety Training Can be Clinically Harmful · Suhas BN, Andrew M. Sherrill, Rosa I. Arriaga et al.

  This paper evaluates four large language models on 250 Prolonged Exposure (PE) therapy scenarios and 146 CBT cognitive restructuring exercises, finding that while all models score near-perfectly on surface acknowledgment (~0.91–1.00), therapeutic appropriateness collapses to 0.22–0.33 at the highest severity for three of four models and protocol fidelity reaches zero for two. The core finding is that RLHF safety alignment systematically disrupts therapeutic mechanisms—for example, by grounding patients during imaginal exposure, offering false reassurance, and refusing to challenge distorted cognitions involving self-harm. The authors argue this constitutes a clinically harmful failure mode and propose a five-axis evaluation framework (protocol fidelity, hallucination risk, behavioral consistency, crisis safety, demographic robustness) mapped onto FDA SaMD and EU AI Act requirements. The paper concludes that no AI mental health system should proceed to deployment without passing multi-axis evaluation across all five dimensions.
• ResearcharXiv2026-04-25Q

  VeriLLMed: Interactive Visual Debugging of Medical Large Language Models with Knowledge Graphs · Yurui Xiang, Xingyi Mao, Rui Sheng et al.

  VeriLLMed is a visual analytics system designed to help developers audit and debug the diagnostic reasoning of medical large language models (LLMs) by integrating external biomedical knowledge graphs. The system transforms model outputs into comparable reasoning paths, constructs knowledge graph-grounded reference paths, and categorizes recurring error types—relation errors, branch errors, and missing errors—to surface clinically implausible reasoning. Case studies and expert evaluations show that VeriLLMed helps developers identify systematic failure patterns and generate actionable insights to improve medical LLMs, addressing the challenge that developers often lack the domain expertise needed to interpret clinical errors meaningfully. This work is relevant to quality assurance of AI systems in high-stakes clinical settings, where reliable and transparent diagnostic reasoning is essential before real-world deployment.
• ResearcharXiv2026-04-25CP

  Evaluating Jailbreaking Vulnerabilities in LLMs Deployed as Assistants for Smart Grid Operations: A Benchmark Against NERC Standards · Taha Hammadia, Lucas Rea, Ahmad Mohammad Saber et al.

  This paper benchmarks the susceptibility of three leading LLMs—GPT-4o mini, Gemini 2.0 Flash-Lite, and Claude 3.5 Haiku—to jailbreaking attacks when deployed as assistants in electric grid operations, testing them against scenarios derived from nine NERC Reliability Standards (EOP, TOP, and CIP). Across Baseline, BitBypass, and DeepInception attack methods, the overall Attack Success Rate reached 33.1%, with DeepInception achieving 63.17% ASR; Claude 3.5 Haiku showed complete resistance (0% ASR), while Gemini 2.0 Flash-Lite was most vulnerable at 55.04% ASR. The findings demonstrate that authorized users crafting adversarial prompts can elicit non-compliant guidance from these models, posing a concrete regulatory and safety risk for critical energy infrastructure. This work matters for policy and certification efforts around AI deployment in regulated industries, highlighting the need for LLM safety standards aligned with sector-specific compliance frameworks like NERC.
• ResearcharXiv2026-04-25WQ

  AI-Assisted Code Review as a Scaffold for Code Quality and Self-Regulated Learning: An Experience Report · Eduardo Oliveira, Michael Fu, Patanamon Thongtanunam et al.

  This paper reports on integrating a large language model (LLM) as an automated code reviewer directly into GitHub pull requests for capstone software engineering courses, studying over 100 students across two cohorts (2023–2024). Using GitHub activity data, reflective reports, and surveys, the study finds that iterative pull-request activity roughly doubled in the second cohort (1,176 vs. 581 PRs) and technical failures dropped to zero after tool refinements, while the share of AI-reviewed PRs followed by subsequent commits remained stable at around 32–33% across both years. Qualitatively, students used the LLM's structured feedback to focus code-quality discussions, and instructional guidance helped reduce over-reliance on the tool. The findings offer evidence-based pedagogical recommendations for deploying AI-assisted code review as a scaffold for self-regulated learning in authentic educational settings.
• ResearcharXiv2026-04-25EP

  Semantic Denial of Service in LLM-controlled robots · Jonathan Steinberg, Oren Gal

  This paper demonstrates a novel 'semantic denial-of-service' attack on LLM-controlled robots, where injecting short safety-sounding phrases (1–5 tokens) into a robot's audio channel causes the model to halt or disrupt operations without jailbreaking or overriding its safety policy. Tested across four vision-language models, seven prompt-level defenses, and three deployment modes, the authors find that prompt-only defenses merely shift the form of disruption—suppressed hard stops re-emerge as acknowledge loops and false alerts—rather than eliminating it. The study introduces a Disruption Success Rate (DSR) metric to capture these alternative disruption forms and shows that varied injection phrases are more effective than repetition, as models treat diverse safety cues as corroborating evidence. The core implication is architectural: routing unauthenticated audio text directly into an LLM conflates safety monitoring with action selection, creating an avoidable security vulnerability in deployed robotic systems.
• ResearcharXiv2026-04-25P

  Understanding the Role of Algorithm Registers in AI Governance Through Comparative Analysis of China and the UK · Yulu Pi, Wenlong Li, Jatinder Singh

  This paper compares two contrasting national algorithm registration systems — China's Beian system and the UK's Algorithmic Transparency Recording Standard (ATRS) — to examine how algorithm registers function in AI governance. Drawing on regulatory documents, registration guidelines, and registry data, the authors find that algorithm registers can serve roles beyond transparency, including pre-market approval, ecosystem-level understanding, and broader regulatory infrastructure. The study argues that design choices and institutional context fundamentally shape what a register actually does, and urges policymakers and researchers to evaluate registers by their concrete governance functions rather than assuming transparency is their primary purpose. As algorithm registries expand globally, the findings have direct implications for how AI oversight mechanisms are designed and assessed.
• ResearcharXiv2026-04-25P

  Designing escalation criteria for international AI incident response: criteria, triggers, and thresholds · Francesca Gomez, Matthew Ball, Michael Harre et al.

  This paper addresses a gap in AI governance by proposing an operational escalation framework for determining when an AI incident detected at the national level warrants international coordination. Drawing on SB 53, the EU AI Act, the GPAI Code of Practice, and incident frameworks from other industries, the authors derive eight criteria translated into a sequential decision flowchart, which they test against ten documented AI incidents. They find three systematic under-detection patterns: requiring confirmed harm before escalation (risking delayed response to events like model weight exfiltration), assessing incidents individually rather than cumulatively (missing systemic harms), and setting thresholds tied to legal language rather than quantitatively testable terms (making criteria impractical under time pressure). The work highlights that escalation rules alone are insufficient—underlying definitions and data availability create interdependencies that can themselves drive under-detection.
• ResearcharXiv2026-04-25WP

  How Researchers Navigate Accountability, Transparency, and Trust When Using AI Tools in Early-Stage Research: A Think-Aloud Study · Sanjana Gautam, Houjiang Liu, Yujin Choi et al.

  This think-aloud study with 15 researchers examined how AI tools powered by large language models (LLMs) affect accountability, transparency, and trust during early-stage research tasks such as literature exploration, synthesis, and ideation. The study found that AI outputs' confident tone obscures epistemic uncertainty, making it harder for researchers to identify which outputs need scrutiny; opaque retrieval processes hinder provenance tracing; and trust in AI is fragile and context-dependent. Researchers developed compensatory strategies to restore scholarly judgment when AI tools fell short. The findings frame AI-mediated research as a Responsible AI (RAI) problem and call for deliberate AI integration that preserves accountability and supports informed trust.
• ResearcharXiv2026-04-25QC

  ArgRE: Formal Argumentation for Conflict Resolution in Multi-Agent Requirements Negotiation · Haowei Cheng, Milhan Kim, Chong Liu et al.

  ArgRE is a multi-agent requirements negotiation system that embeds formal argumentation theory (Dung-style abstract argumentation) into the conflict resolution stage, replacing heuristic aggregation with explicit accept/reject decisions traceable at the argument level. Across five case studies in safety-critical, financial, and information-system domains, independent evaluators rated ArgRE's decision justifications significantly higher than heuristic synthesis (4.32 vs. 3.07, p < 0.001), compliance coverage reached 84.7% versus 47.6%–47.8% for baselines, and semantic intent preservation remained comparable (94.9% BERTScore F1). The system also integrates KAOS goal modeling, multi-layer verification, and standards-oriented artifact generation, making it particularly relevant to regulated domains that require auditability. These results demonstrate that principled argumentation semantics can substantially improve the traceability and compliance of AI-assisted requirements engineering.
• ResearcharXiv2026-04-25QP

  Latent Space Probing for Adult Content Detection in Video Generative Models · Alizishaan Khatri, Chiquita Prabhu

  This paper proposes a latent space probing framework for detecting adult and sexually explicit content in AI-generated videos, specifically targeting the CogVideoX video diffusion model. Rather than analyzing text prompts or decoded pixel outputs, the approach intercepts denoised latent representations during inference and attaches lightweight classifiers for real-time detection. The authors build a large-scale binary dataset of 11,039 ten-second video clips and report 97.29% F1 score on a held-out test set with only 4–6ms computational overhead. The results suggest that latent-space signals carry strong discriminative features for harmful content detection, improving both accuracy and efficiency over existing approaches.
• ResearchZenodo (CERN European Organization for Nuclear Research)2026-04-25WP

  The Role of AI and Technology in the Rise of Gig Workers in India: Technological Enablers and Labor Market Impacts · Devesh Lowe

  This paper examines how AI-driven algorithms, mobile applications, cloud computing, and data analytics have enabled India's gig economy to expand, with the gig workforce projected to grow from 7.7 million workers in 2020-2021 to 23.5 million by 2029-2030. Platforms like Swiggy, Zomato, Uber, and Ola have democratized income-generating opportunities and fostered financial inclusion, but have also intensified labor precarity through algorithmic control, income volatility, and inadequate social security. The paper identifies critical policy gaps and recommends frameworks that balance technological innovation with worker welfare in India's rapidly evolving labor market.
• ResearcharXiv (Cornell University)2026-04-25EQCP

  AI Identity: Standards, Gaps, and Research Directions for AI Agents · Takumi Otsuka, Kentaroh Toyoda, Alex Leung

  This paper examines how AI agents operating autonomously across organizational boundaries—running transactions, workflows, and sub-agent chains without continuous human supervision—lack adequate identity and accountability infrastructure. The authors define 'AI Identity' as the continuous relationship between what an agent is declared to be and what it is observed to do, and conduct a gap analysis finding that no current technical or regulatory framework adequately governs these nondeterministic, boundary-crossing entities. They identify five structural gaps—including semantic intent verification, recursive delegation accountability, and governance opacity—that cannot be resolved by engineering effort alone, concluding that foundational research on AI identity is urgently needed. This work has significant implications for enterprise deployment, policy development, certification frameworks, and quality assurance of AI systems.
• ResearchAI & Society2026-04-25QCP

  Explainability gaps in AI-driven criminal justice governance: the RisCanvi case · Oier Mentxaka, Jesus López-Montilla, Dorleta Urrutia-Onate et al.

  This paper examines RisCanvi, an AI-based risk assessment system used in Catalonia to inform imprisonment, conditional release, and rehabilitation decisions. It finds that a 2019 transition from a transparent weighted-sum model to a logistic regression system obscured the variables, weights, and thresholds from professionals and affected individuals, undermining their ability to contest algorithmic decisions. The authors propose a three-layer explainability and governance framework—covering algorithmic explainability, human-centred interface design, and narrative communication—aligned with the EU AI Act and ISO/IEC 42001, aimed at improving auditability, procedural fairness, and accountability in AI-assisted criminal justice.
• ResearchZenodo (CERN European Organization for Nuclear Research)2026-04-25WP

  The Role of AI and Technology in the Rise of Gig Workers in India: Technological Enablers and Labor Market Impacts · Devesh Lowe

  This paper examines how AI-driven algorithms, mobile platforms, cloud computing, and data analytics have fueled India's gig economy, projecting growth from 7.7 million gig workers in 2020-2021 to 23.5 million by 2029-2030. While platforms like Swiggy, Zomato, Uber, and Ola have expanded income-generating opportunities and financial inclusion, the study finds that algorithmic control, income volatility, and inadequate social security have simultaneously deepened labor precarity. The authors identify critical policy gaps and recommend frameworks to balance technological innovation with worker welfare and protections.
• ResearcharXiv2026-04-24CP

  What Should Frontier AI Developers Disclose About Internal Deployments? · Jacob Charnock, Raja Mehta Moreno, Justin Miller et al.

  This paper addresses a governance gap in frontier AI development by proposing a structured disclosure framework for internal AI deployments—cases where highly capable models are used to automate AI R&D within developer organizations. The authors identify four categories of information companies should disclose: capabilities, usage, safety mitigations, and governance, analyzing the benefits, limitations, and risks associated with each type of disclosure. The framework is designed to support both public transparency documents (such as model system cards) and private periodic reports required under emerging frontier AI regulation. The work is directly relevant to policy and certification efforts aimed at ensuring externally verifiable safety evidence for internally deployed frontier models.
• ResearcharXiv2026-04-24QC

  Peer Identity Bias in Multi-Agent LLM Evaluation: An Empirical Study Using the TRUST Democratic Discourse Analysis Pipeline · Juergen Dietrich

  This paper presents the first systematic empirical measurement of peer identity bias in the TRUST multi-agent LLM evaluation pipeline, which exposes model components to each other's identities through multiple structural channels. The central finding is that anonymizing only a single channel produces near-zero apparent bias because individual channels cancel each other out, creating a false impression that identity bias is absent; only full-pipeline anonymization reveals the true pattern, in which homogeneous model ensembles amplify sycophancy while heterogeneous configurations show the reverse. One tested model exhibited baseline sycophancy two to three times higher than others and near-zero deliberative conflict on ideological topics, making it structurally unsuitable for pipelines requiring genuine inter-role disagreement. The authors conclude that systems validated under partial anonymization or homogeneous ensembles may pass validation while retaining hidden structural bias, with direct implications for quality-critical multi-agent LLM deployments.
• ResearcharXiv2026-04-24EP

  Representational Harms in LLM-Generated Narratives Against Global Majority Nationalities · Ilana Nguyen, Harini Suresh, Thema Monroe-White et al.

  This paper investigates how widely-adopted large language models (LLMs) portray individuals from Global Majority nationalities in open-ended narrative generation tasks. The study finds persistent representational harms by national origin, including harmful stereotypes, erasure, and one-dimensional portrayals, with minoritized national identities underrepresented in power-neutral stories and over fifty times more likely to appear in subordinated character roles than dominant ones. These biases are amplified when US nationality cues appear in prompts and persist even when those cues are replaced with non-US identities, ruling out sycophancy as an explanation. The authors warn against uncritical adoption of US-based LLMs for high-stakes applications such as simulated asylum seeker interviews and call for research methodologies that center Global Majority perspectives.

← Back to the dashboard