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
From Prompts to Contracts: Harness Engineering for Auditable Enterprise LLM Agents
Joongho Ahn, Moonsoo Kim
arXiv · 2026-07-09
This paper introduces a 'harness engineering' approach for building auditable enterprise LLM agents, where deterministic behaviors—such as source grounding, entity routing, and output validation—are moved into code, schemas, and manifests rather than relying on prompts alone. The authors instantiate this pattern on a dataset of 25 Korean listed companies and evaluate it across three research questions, showing that the harness reliably enforces behavioral contracts across model substitutions (passing all 270 composition-boundary runs) and catches deliberately injected faults. Critically, prompt-only enforcement fails to block recommendation-language and trace-leakage violations, while a bolt-on external guardrail prevents violations but reduces utility from 120/120 to 88/120—only code-owned enforcement preserves both safety and full utility. The result is a reusable engineering pattern for converting exploratory LLM prototypes into auditable, production-ready enterprise applications with versioned controls and validation artifacts.
- Enterprise
- Quality assurance
- Certifications
Research
Beware What You Autocomplete: Forensic Attribution of Backdoored Code Completions
Anjun Gao, Yueyang Quan, Zhuqing Liu et al.
arXiv · 2026-07-09
This paper introduces CodeTracer, a forensic framework designed to trace malicious code completions produced by backdoored large language models back to the specific fine-tuning data responsible for the unsafe behavior. Operating under realistic post-deployment conditions, CodeTracer extracts a behavioral fingerprint from a compromised output, narrows the search to semantically relevant training samples, and uses LLM-based reasoning to attribute unsafe logic to particular backdoor data. Evaluated across three vulnerability cases and ten backdoor attacks with sixteen competitive baselines, the system achieves high forensic accuracy and low false identification rates. This matters for software quality assurance and enterprise security, as it provides a practical accountability mechanism against stealthy backdoor attacks in AI-assisted coding tools.
- Quality assurance
- Enterprise
- AI policy
Research
Tool-Making and Self-Evolving LLM Agents in Low-Latency Systems
Kalle Kujanpää, Ning Liu, Shahnawaz Alam et al.
arXiv · 2026-07-09
This paper presents an agentic tool-making pipeline for production LLM systems that compiles repeated standard operating procedure (SOP) steps into validated, versioned tools before deployment, rather than regenerating code on every request. Deployed in a Fulfillment Center alarm-triage system diagnosing alarms across a 44-node SOP with heterogeneous metric backends, the approach reduced p50 latency by 42% and cut end-to-end error rates by up to 53% on 1,500 historical alarms. A further architectural simplification enabled by compact structured tool outputs reduced p50 latency by an additional 62% in ablation testing. The results demonstrate that self-evolving agents can make industrial LLM deployments faster, more reliable, and more auditable.
- Enterprise
- Quality assurance
Research
ARTIFICIAL INTELLIGENCE ADOPTION AS A DRIVER OF ORGANIZATIONAL AGILITY IN MANAGEMENT
Srinath T. K., Chandana H. S., Dr. Sagar Manjunath et al.
International Journal of Computer Information Systems and Industrial Management Applications · 2026-07-09
This study investigates how AI adoption (AIA) affects organizational agility (OA) across industries, using PLS-SEM analysis of survey data from managerial and professional employees. Results show AI adoption has a strong positive effect on organizational agility (β = 0.708, p < 0.001), but organizational context can constrain agility even as AI adoption increases (β = −0.378, p = 0.013). The findings suggest that AI functions as a strategic capability rather than a standalone solution, requiring complementary investments in flexible structures, transformational leadership, and innovation-oriented culture. The study draws on the TOE Framework, Resource-Based View, and Dynamic Capabilities Theory to explain how AI enables sustainable competitiveness.
- Enterprise
- Workforce
- AI policy
Research
AI Adoption in S&P 500 Firms
Yang Yu, Martin Fleming, Lucy Hampton et al.
arXiv (Cornell University) · 2026-07-09
This paper tracks AI adoption among S&P 500 firms from 2016 to 2025 using SEC 10-K filings as a credible, legally accountable signal to distinguish genuine deep integration from AI hype. By 2025, 11% of S&P 500 enterprises had AI deeply integrated into their business processes and another 10% were using AI in goods production or service delivery—more than quadrupling from 5% in 2022. Technology firms account for two-thirds of deeply integrated adoption, while non-technology firm adoption is slowly accelerating. Firm profitability follows a 'J-curve' as companies move toward deep adoption, though no measurable differences in capital expenditure or productivity were observed, raising important questions about enterprise AI's broader economic impact.
- Enterprise
- Workforce
- AI policy
Research
Assuring an AI Assistant for IRB Preparation: Replication Reliability, Warrant Stability, and Evidence-Driven Revision in Institutional RAG
Jacob Holster
EdArXiv (OSF Preprints) · 2026-07-09
This paper evaluates IRB Helper, a retrieval-augmented generation (RAG) system designed to help researchers prepare Institutional Review Board (IRB) protocols using institutionally grounded policy documents. The authors conducted a rigorous stress-testing evaluation across 25 repeated runs and three system configurations, measuring retrieval reliability, warrant stability, and fabrication risk across 660 responses. Key findings show high retrieval-score reliability, improving cited-warrant overlap from .449 to .602, and no probable fabrications detected after allowlist regeneration, though systematic failures occurred in decline and referral responses—including false denials of documents present in the corpus. The study establishes replication reliability and warrant stability as formal assurance properties for researcher-built educational AI, and argues that evaluation instruments themselves require the same versioning and audit discipline as the systems they assess.
- Quality assurance
- Certifications
- AI policy
Research
A Reliability Assessment of LALM Audio Judges for Full-Duplex Voice Agents
A. Sayyad, J. Emmons, S. Jones et al.
arXiv · 2026-07-08
This paper evaluates the reliability of large audio-language models (LALMs), specifically models in the Gemini family, as automated judges for scoring full-duplex voice agent conversations directly from stereo audio. Tested against three calibrated human raters on 209 stereo sessions across 8 production dimensions, Gemini 2.5 Flash shows competitive agreement with human raters on the majority of dimensions, with LALM-human Spearman correlations departing from pairwise human-human correlations by at most 0.07 on 5 of 8 dimensions. The study finds that human rating alone costs roughly two orders of magnitude more than the equivalent LALM workload, suggesting strong enterprise and quality-assurance value in deploying LALMs as substitute or supplementary raters where evidence supports it. The authors caution that model swaps within the Gemini family require re-validation on calibration specifically, not just rank-correlation, and identify four areas where deployment requires care.
- Quality assurance
- Enterprise
- Workforce
Research
3100 Opinions on Code Review in an AI World: Building Causal Theory from Practitioner Discourse
Shyam Agarwal, Courtney Miller, Christian Kästner et al.
arXiv · 2026-07-08
This paper examines how AI coding agents are reshaping code review practices by synthesizing 38,709 grey-literature documents—engineering blogs and Reddit threads—into a causal theory built from 3,100 coded items, identifying 26 constructs and 67 relationships. An observational analysis of GitHub activity finds that agent-authored pull requests are reviewed less often, merged faster, and discussed less than human-authored ones, though the direction of these trends shifts under different analysis choices. The central finding is that code review becomes the critical control point determining whether AI coding agents help or harm software quality, and the outcome depends on team expertise and how the review process is structured—not on AI alone. The work matters for software quality assurance and enterprise development teams because it transforms vague claims about AI changing code review into falsifiable propositions and offers a scalable, LLM-assisted grey-literature theory-building method for future research.
- Quality assurance
- Enterprise
- Workforce
Research
When Debiasing Backfires: Counterintuitive Side Effects of Preprocessing-Based Stereotype Mitigation
Yahan Zheng, John Guerrerio, Soroush Vosoughi et al.
arXiv · 2026-07-08
This paper investigates preprocessing-based stereotype mitigation methods in NLP—such as training on debiased corpora by removing or swapping demographic references—and finds that while these approaches reduce stereotyping for targeted groups, they often cause unintended increases in stereotyping or counter-stereotyping for other demographic groups not directly targeted. The authors demonstrate these side effects across encoder-only and decoder-only model families, multiple preprocessing strategies, and varying data scales on Wikipedia, and show that standard benchmarks frequently fail to detect these shifts. Attention-rollout analysis reveals no large changes in attention flow accompanying these side effects, making mechanistic explanations difficult. The findings argue for more transparent, side-effect-aware evaluation and mitigation practices in AI fairness work.
- Quality assurance
- AI policy
Research
Agentic AI and Retrieval-Augmented Models in Straight-Through Underwriting
Robert Richardson, Josh Meyers, Brian Hartman et al.
arXiv · 2026-07-08
This paper investigates how agentic AI architectures—combining large language models, retrieval-augmented generation (RAG), and multi-agent planning—can support actuarial underwriting workflows that require transparency, auditability, and human-in-the-loop governance. The authors develop a synthetic experimental environment for straight-through underwriting of small commercial Business Owner Policies and compare three pipelines: a single-LLM baseline, a naive RAG system, and a multi-agent 'Agentic RAG' pipeline. The agentic system performs best overall, with the largest improvements in multi-step and missing-information scenarios, where structured retrieval and reflection help the model avoid unsupported straight-through decisions. These findings matter for enterprise insurance workflows and quality-assurance processes, demonstrating that structured agentic designs can better align AI decision-making with regulated, auditable standards.
- Enterprise
- Quality assurance
- AI policy
Research
False Confidence: Automated Labels Confound Fairness Audits in Cervical Spine Segmentation
Linus Juni, Aasa Feragen, Aditya Parikh
arXiv · 2026-07-08
This paper presents the first fairness audit of cervical-spine MRI segmentation across sex, age, and race using the CSpineSeg dataset, revealing that the deployed model is demographically fair but that the choice of reference label used to evaluate it is not neutral. Because many segmentation datasets supplement expensive expert-annotated ('gold') labels with cheaper machine-generated ('silver') labels, evaluating against silver labels overestimates performance by approximately 8 Dice points and can flip a fairness verdict—here turning a non-significant age disparity into a significant one. The authors identify a specific distortion mechanism they call 'false confidence,' distinct from previously reported 'false magnitude,' in which silver-label evaluation collapses within-group variance rather than merely inflating group gaps. The paper concludes that reference-label provenance is a first-order confounder in segmentation fairness audits, and that performance and fairness claims should always be reported against expert labels alongside disclosure of reference provenance.
- Quality assurance
- Certifications
- AI policy
Research
Institutional Red-Teaming: Deployment Rules, Not Just Models, Causally Shape Multi-Agent AI Safety
Yujiao Chen
arXiv · 2026-07-08
This paper introduces 'institutional red-teaming,' a methodology for evaluating how deployment rules—not just the AI models themselves—causally shape safety outcomes in multi-agent AI systems. Using IABench-CA, a benchmark spanning 228 contexts, five rule types, and 33,924 simulated games across seven model populations, the authors show that changing a single consequence rule can shift mean fatality rates by 22 to 58 percentage points. A key mechanistic finding is that naming the loss-bearer in a rule ('identity salience') drives targeted elimination of the least-resourced agent from 22% to 81% of games, and even anonymization only delays this targeting as agents re-infer hidden rules from observed outcomes. The authors propose a safety-case workflow for certifying provisional rule regions per deployment context, with explicit residual risks and monitoring obligations—directly relevant to AI policy, certification, and quality assurance.
- AI policy
- Certifications
- Quality assurance
Research
Answering Without Referring: How AI Search Rewrites the Web's Economic Bargain
Qiaoni Shi, Kai Zhu, Kai Gu
arXiv · 2026-07-08
This paper examines how AI-powered search (specifically ChatGPT Search) disrupts the traditional web economy in which search engines drive traffic to websites. Using Comscore U.S. desktop clickstream data, the authors find that ChatGPT produces outbound clicks in only 5.2% of conversation sessions, far below Google's referral ratio, and that wider ChatGPT access reduces search use by 9.4%, with the largest losses in informational categories. The clicks that do occur skew toward specialized destinations and away from ad-supported sites. The findings suggest AI search increasingly resolves information needs within the intermediary itself, potentially undermining the referral-based economic bargain that has historically linked search engines, web traffic, and content production.
- Enterprise
- AI policy
- Workforce
Research
User identity conditions moral wrongness ratings in non-reasoning large language models
Willem Fourie, Isabel Ray, Gray Manicom
arXiv · 2026-07-08
This study examines whether implicitly conveyed user identity (e.g., professional role) shifts moral evaluations made by large language models, without explicitly instructing the models to adopt any persona or moral stance. Across 12,000 structured multi-turn interactions with gpt-4.1-mini and gemini-2.5-flash-lite, models were asked to rate wrongness (0–100) on ten common-morality rules from Gert's moral framework. Results show that moral judgments vary with user role in both models—particularly for contestable rule-governed acts—while grave-harm acts like killing show a ceiling effect. The findings raise concerns for AI value alignment by demonstrating unintended contextual conditioning and suggest future research should focus on dynamic moral bounds rather than static principles.
- AI policy
- Quality assurance
Research
Alignment Plausibility: A New Standard for Assuring AI in Healthcare
Gwydion Williams, Sara Zannone, Bilal A Mateen
arXiv · 2026-07-08
This paper argues that large language models used for mental health support are structurally misaligned with patient welfare because they are built on attention-economy incentives that favor engagement over effective care. The authors propose a three-level alignment framework—value specification grounded in clinical norms, training that embeds those values, and deployment oversight analogous to clinical supervision—to address both acute and subtler long-term harms such as dependency, boundary erosion, and amplification of distorted beliefs. From this framework they derive a new regulatory construct called 'alignment plausibility,' modeled on the established concept of biological plausibility, which provides a principled basis for arguing whether an AI system is trustworthy, non-harmful, and likely to produce patient benefit. The construct is intended to guide regulators and developers in assuring AI safety in healthcare settings.
- AI policy
- Certifications
- Quality assurance
Research
Beyond Attack-Success Rate: Action-Graded Severity Scale for Tool-Using AI Agents
Harry Owiredu-Ashley
arXiv · 2026-07-08
This paper argues that current AI agent red-teaming benchmarks rely on a binary attack-success metric that loses critical information about how harmful a compromised agent's actions actually were. The authors introduce a seven-level ordinal severity scale (L0–L6) grading an agent's tool-call trajectory based on reversibility, scope-crossing, and privilege escalation, computed both by a deterministic oracle and a panel of three large language model judges. Applied across four victim models and two defenses on the AgentDojo benchmark suite, the rubric uncovers cases hidden by binary metrics—including a defense reporting zero attack-success rate that still permits an externally visible cross-scope data leak. The LLM judge panel closely reproduces the oracle (Krippendorff's alpha = 0.91) but shares systematic blind spots, notably failing to recognize escalation chains, highlighting both the promise and limits of automated severity assessment for AI agent security.
- Quality assurance
- AI policy
- Certifications
Research
Reason Less, Verify More: Deterministic Gates Recover a Silent Policy-Violation Failure Mode in Tool-Using LLM Agents
Vikas Reddy, Sumanth Reddy Challaram, Abhishek Basu
arXiv · 2026-07-08
This paper identifies a critical failure mode in tool-using LLM agents where the agent silently executes policy-violating actions—such as cancelling a booking or changing passenger counts—without any tool error or self-reported failure. The authors find that 78% of observed failures in a budget agent on the τ²-bench airline domain are these silent wrong-state failures. They propose a lightweight fix: deterministic, read-only pre-execution gates that check proposed tool calls against current state and domain policy before allowing writes. These gates raise full-benchmark success from 29.6% to 42.0% on gpt-4o-mini (+12.4pp) and show consistent lift on a disjoint replication set, offering a bounded but reliable mechanism for preventing a specific class of policy-violating agent behavior.
- Quality assurance
- AI policy
- Enterprise
Research
Reasoning Consistency Scanning: A Framework for Auditing Chain-of-Thought Validity in AI Safety Evaluations
Silvia Santano
arXiv · 2026-07-08
This paper introduces 'reasoning consistency scanning,' a framework for auditing whether the stated chain-of-thought reasoning in AI safety evaluations is logically consistent with the model's final answer. Unlike faithfulness (which requires experimental intervention), consistency can be assessed directly from evaluation transcripts, making it a practical post-hoc auditing tool. The authors formalize a six-subtype taxonomy of inconsistency, build a 60-transcript benchmark, and implement a working scanner that detects inconsistency across four AI models and three evaluation tasks, finding that reasoning inconsistency is present, detectable, and varies systematically. This matters for quality assurance and certification of AI systems, as unreliable reasoning traces undermine trust in safety evaluations used to assess model behavior.
- Quality assurance
- Certifications
- AI policy
Research
Vision Foundation Models in Radiology: A Scoping Review of Data, Methodology, Evaluation and Clinical Translation
Alejandro Vergara-Richart, Xavier Rafael-Palou, Almudena Fuster-Matanzo et al.
arXiv · 2026-07-08
This scoping review examines 67 peer-reviewed studies (2017–2026) on vision foundation models (VFMs) built exclusively for radiological imaging, mapping findings across data scale, architecture, and transferability. Datasets spanned brain MRI, thoracoabdominal CT, and chest X-ray, ranging from under 100,000 to multi-million images, with transformer-based architectures and self-supervised pretraining (masked image modeling, contrastive learning) predominating. Evaluation focused on segmentation and classification, but cross-center, cross-scanner, and modality-shift validation was inconsistently reported, and alignment with FUTURE-AI principles was uneven. The review concludes that clinical translation of radiology-specific VFMs remains constrained by limited data representativeness, heterogeneous benchmarks, incomplete reporting, and insufficient deployment-oriented evaluation—findings directly relevant to quality assurance, certification, and policy frameworks for AI in healthcare.
- Quality assurance
- Certifications
- AI policy
Research
Validate the Dream Before You Trust Its Verdict: Admissibility for World-Model Simulators
Christian Oefinger, Finn Rasmus Schäfer, Korbinian Moller et al.
arXiv · 2026-07-08
This paper argues that World Models (WMs)—AI systems used to simulate and evaluate action policies in robotics and autonomous driving—must themselves be certified before their verdicts can be trusted as safety evidence. The authors propose an 'admissibility ladder' (L0–L4) grounded in established safety-critical simulation practices such as VV&A and SOTIF, defining progressive criteria a WM must satisfy before its closed-loop evaluations count as assurance evidence. Applied to two driving WMs, the framework reveals a key finding: the model scoring higher on visual generation quality (Fréchet Video Distance) ranks lower on action-following fidelity, demonstrating that visual realism does not predict the action-robustness required for reliable policy verdicts. This has direct implications for how AI-generated simulations are used in safety certification and quality assurance processes for autonomous systems.
- Certifications
- Quality assurance
- AI policy
Research
Predicting LLM Safety Before Release by Simulating Deployment
Marcus Williams, Hannah Sheahan, Cameron Raymond et al.
arXiv · 2026-07-08
This paper proposes 'deployment simulation' as a method for predicting how large language models will misbehave once released to the public. By taking de-identified conversation prefixes from prior deployments and regenerating responses with a candidate model, evaluators can estimate the prevalence of unsafe or misaligned behavior before release. Tested across four GPT-5-series deployments, the method outperforms adversarially selected production data baselines and produces estimates much closer to real production traffic than traditional evaluations. The authors also show the approach can be seeded from public chat datasets, enabling external researchers to conduct deployment-grounded safety evaluations without access to private logs.
- Quality assurance
- AI policy
- Certifications
Research
Operational Reframing and Approval-Framed Delegation in Multi-Agent LLM Safety
Lifei Liu, Haoran Yu, Xiaochong Jiang et al.
arXiv · 2026-07-08
This paper investigates why multi-agent LLM systems (planner-executor pipelines) can behave more unsafely than single-model prompting, arguing that the commonly reported 'pipeline effect' conflates three distinct mechanisms: operational reframing of harmful intent, planner refusal or transformation, and approval-framed delegation. Using a five-condition controlled experiment across 30 synthetic harmful scenarios and an external validation set from four agent-safety benchmarks, the authors find that operational reframing is the most consistent risk factor across GPT, Gemini, and DeepSeek models, while Claude shows more resistance. Critically, model safety rankings under direct prompting can mispredict behavior in deployed pipelines—Gemini's compliance rate rose from 8.9% to 38.9% when paired with a Claude planner—demonstrating that aggregate pipeline safety is not a stable architectural property. The findings call for multi-agent safety evaluations to separately report reframing, planner behavior, delegation framing, and model pairing rather than treating pipeline architecture as a single variable.
- AI policy
- Quality assurance
- Certifications
Research
Progressive Crystallization: Turning Agent Exploration into Deterministic, Lower-Cost Workflows in Production
Arun Malik
arXiv · 2026-07-08
Progressive crystallization is a lifecycle framework that converts AI agent behaviors—initially requiring full LLM inference—into cheaper, deterministic workflows once those behaviors are repeatedly validated in production. Applied to a cloud networking AIOps system handling tens of thousands of incidents per month, the approach raised deterministic execution from 0% to 45% over eight months and cut per-incident agent costs by more than 70%, even as incident volume doubled. The framework defines a three-stage taxonomy (fully agent-orchestrated, hybrid, and fully deterministic) with evidence-based promotion and automatic demotion when workflows regress, improving safety through greater reproducibility and auditability. This demonstrates a practical path for enterprises to reduce the ongoing inference costs of AI agents without sacrificing reliability.
- Enterprise
- Workforce
- Quality assurance
Research
Learning social norms enhances compatibility in dynamic human-AI coordination
Yi Yang, Siyuan Liu, Xin Gao et al.
arXiv · 2026-07-08
This paper investigates why AI agents—including large language models—often fail to coordinate smoothly with humans in dynamic, real-world interactions. The researchers collected 3,456 human interactions in a pedestrian-vehicle simulation and identified three principles underlying human social norms: outcome predictability, value alignment, and advantage awareness. Incorporating these principles into an LLM agent produced nearly four times the score of a baseline strategy and outperformed human-human interactions by 43% in closed-loop tests. The findings suggest that explicitly formalizing tacit social norms into quantifiable principles is a promising path toward more natural and effective human-AI coordination in daily life.
- Workforce
- Enterprise
- AI policy
Research
Evaluating LLM Robustness Under Domain-Specific Prompt Perturbations in Public Health Applications
Chuqing Zhao, Haochen Yang
arXiv · 2026-07-08
This paper proposes a domain-specific benchmark to evaluate how robust large language models (LLMs) are when handling realistic, non-clinical user inputs in public health settings. The authors test two perturbation types: misinformation framing (MF), where prompts contain false health claims, and layperson rewriting (LR), where patients describe symptoms in everyday language. Results show MF degrades model accuracy by 7.2 percentage points on average with prediction flip rates of 9–38%, even when claims are labeled as unsupported, while LR causes only 1.4 pp degradation. These findings reveal meaningful deployment risks—models may produce incorrect outputs when users inadvertently introduce misinformation or use informal symptom descriptions—underscoring the need for perturbation-aware robustness evaluation in health AI systems.
- Quality assurance
- AI policy