Horizon scanning is a foresight methodology designed to uncover and assess emerging issues before they become full-blown emergencies. In the plant health context, it functions as the initial step in a surveillance pipeline, gathering signals from thousands of sources and distilling them into actionable intelligence for risk managers. The EU system, run by EFSA in collaboration with the European Commission’s Joint Research Centre (JRC) and the French Agency for Food, Environmental and Occupational Health and Safety (ANSES), produces monthly reports known as newsletters. As of 2024, this activity has generated 130 such reports, covering both pests already regulated under EU law and, crucially, unknown or unlisted species that no legislation yet anticipates.

The technological engine behind the monitoring phase is MedISys, the Medical Information System developed by the JRC. Originally built for public health surveillance, MedISys is a fully automatic, event-based system that in 2024 was scanning more than 3,200 scientific and media sources worldwide every day. Manually curated categories, keyed to the scientific and multilingual common names of individual pests, allow the platform to retrieve articles about specific organisms, while generic keywords such as pest, plant disease, crop and yield losses cast a wider net for the discovery of entirely new threats. Articles flagged by the machine are then reviewed and validated by a team of experts before being selected for the monthly newsletter, which is published in the EFSA Journal and presented to the Plant Health section of the EU Commission’s PAFF Committee.

The second stage of the pipeline is a rapid screening procedure called PeMoScoring, short for pest and monitoring scoring. Eligible pests must be capable of damaging plant species of economic relevance to the EU, must be able to enter the territory through at least one import commodity that is not banned or by natural spread from outside the EU, and must not be newly described species. PeMoScoring evaluates 15 criteria grouped into five items: host range, entry, establishment, spread and impact. Each criterion is posed as a question with predefined response options whose scores were calibrated against reference pests, ensuring consistency across assessments. The answers combine into a final net phi value ranging from minus one to plus one, with a threshold separating negative results, meaning insufficient evidence of risk, from positive results that signal a potential threat and can trigger further action by risk managers.

To evaluate how well this machinery has performed, the team compiled every article published in the newsletters between 2017 and 2024 into a database, recording the pest’s biological category, scientific name, regulatory status, article topic, source, PeMoScoring outcome and, where applicable, the country of detection and host plant. The dataset revealed that the 130 reports covered 1,153 pests, of which 836 were not listed in EU regulation or EPPO lists and 317 were regulated. Insects and mites accounted for the highest number of regulated species cited, while fungi and oomycetes stood out as the category in which unlisted species were more frequently reported than regulated ones.

The source analysis yielded instructive patterns. Although the majority of unique sources were media outlets, with 617 compared to 301 scientific sources, the bulk of the articles actually came from the scientific literature: 1,978 articles from journals versus 1,380 from media. For regulated pests, media coverage dominated in the insects and mites category, whereas viruses, viroids and phytoplasmas appeared more often in peer-reviewed publications. For unlisted pests, the scientific literature was generally the richer source, particularly for fungi and oomycetes and for viruses. Notably, a large share of articles came from sources consulted only once, suggesting that casting a wide net across many outlets, even shallowly, increases the volume of relevant intelligence without sacrificing quality.

Topic classification showed that new observations, encompassing first findings, interceptions, new host plants, new pests and new vectors, were by far the most common group, followed by management and then assessment, the latter two being associated predominantly with regulated pests. Among the subtopics, first findings were mainly reported in scientific publications, while new findings, defined as detections of a pest in a different region within the same country, were primarily reported in the media. For regulated insects and mites, the pattern was striking: 97 percent of articles reporting new findings originated from media sources. Of the 392 new pests reported during the exercise, only 27 were mentioned again in subsequent articles, and crops accounted for 67.5 percent of host plant citations, with forests, ornamentals and wild plants making up the remainder.

The PeMoScoring results demonstrate the system’s real filtering power. From 2017 to 2024, 233 pests were scored, with 119 receiving positive results and 114 negative ones. Twenty-seven of the positive pests were proposed for pest categorisation, EFSA’s more elaborate assessment of whether regulatory measures may be required, and 20 of those 27 categorisations returned a positive outcome, with three negative and four inconclusive. On average, it took about three years for a PeMo-positive case to undergo pest categorisation. The proportion of positive scores has declined since 2021, which the authors suggest may reflect improving protective measures, since diagnosis and control options feed into the scoring evaluation.

Perhaps the most vivid illustration of the system’s value is the case of Fusarium oxysporum f. sp. cubense tropical race 4, the soil-borne fungus behind Panama disease in bananas. TR4 scored positive in PeMoScoring as early as July 2017, when its known distribution was limited to eastern Asia and the Middle East, even though bananas are not widely cultivated in the EU. A pest categorisation followed in 2022, concluding that the pathogen met the criteria for consideration as a potential Union quarantine pest. Since then, TR4 has spread to Türkiye, Oceania and, significantly, Latin America, with detections in Colombia in 2019, Peru in 2021 and Venezuela in 2023. In 2024 the pathogen was added to the EPPO A2 list, and the International Plant Protection Convention issued new prevention guidelines covering e-commerce pathways.

Media signals have also proven decisive for other invaders. Reports of the invasive fire ant Solenopsis invicta in Sicily surfaced well before the formal scientific publication, underscoring the potential of integrating citizen science and social media into surveillance. The February 2022 newsletter, meanwhile, carried the first observation of the fungus Neofusicoccum mediterraneum in Italy, a new report of Citrus concave gum-associated virus in the United States, and evidence that barley serves as a host for Chinese wheat mosaic virus. None of these were listed in the EU Plant Health Regulation, yet the procedure triggered a request from the Swedish Board of Agriculture for tailored risk assessments, showing how scanning outputs translate directly into preparedness actions.

Looking forward, the authors argue that the vast dataset accumulated over seven years could support a shift from consequence-to-event tracking toward an event-to-consequence approach, and ultimately toward artificial-intelligence systems capable of pinpointing critical information automatically. A preliminary driver-based study of newsletters published between June 2021 and September 2022 confirmed that global trade is the main driver of pest emergence, and other research points to traits such as polyphagy, heat tolerance, high dispersal capacity and genetic diversity as markers of future risk. Workshops are now being organised to build a plant health horizon scanning community across institutions, and the newsletter data are available through an interactive EFSA dashboard. Limitations remain, including delays caused by constraints in data publication and web scraping, but the overall verdict is clear: seven years of systematic horizon scanning have measurably strengthened Europe’s capacity to anticipate, rather than merely react to, the next plant health threat.

Subject of Research: Horizon scanning methodology for early detection of emerging plant pest threats in the European Union

Article Title: Unravelling potential plant health threats for the European Union: application of horizon scanning methodology

Article References: Ribaya, M., López-Mercadal, J., Campese, C., Sarakatsani, E., & Tramontini, S. (2025). Unravelling potential plant health threats for the European Union: application of horizon scanning methodology. Web Ecology, 25(2), 189-200. https://doi.org/10.5194/we-25-189-2025

Image Credits: AI Generated

DOI: 10.5194/we-25-189-2025

Keywords: horizon scanning, plant health, EFSA, invasive pests, PeMoScoring, MedISys, biosecurity, plant pathogens, risk management, Fusarium TR4, early warning systems, European Union

The motivation is rooted in painful history. The catastrophic Central European flood of 2002, which caused an estimated 22.6 billion euros in direct damages, remains the most expensive natural hazard event ever recorded in Germany, and it exposed the limits of traditional flood protection thinking. Subsequent floods in 2013, 2021, and 2024 reinforced the lesson: absolute safety against flooding is unattainable, dikes can fail even below their design standards, and local protection measures can shift hazard downstream rather than eliminate it. In response, the field has been moving from regulation-based design toward integrated flood risk management, in which the entire spectrum of possible flood events is assessed probabilistically and investments are weighed against the risk reduction they deliver.

The Munich framework addresses three persistent challenges at once. First, it adopts a system-based perspective at river scale, dynamically coupling the components of the flood process chain, including downstream flood wave propagation and possible dike failures, rather than treating them as static, independent layers. Second, it embeds the risk calculation in a benefit-cost decision framework, estimating flood risk both with and without a given mitigation measure. Third, and perhaps most distinctively, it separates aleatory uncertainty, the irreducible randomness of nature, from epistemic uncertainty, the gaps in knowledge and model imperfection that better data could in principle reduce. This separation is achieved through a two-level Monte Carlo architecture, a statistical design that allows decision-makers to see not just a single risk number but a distribution of plausible risks and the sensitivity of conclusions to each uncertain input.

Computational efficiency is the engine that makes all of this possible. Traditional large-scale flood risk models face a brutal trade-off between process fidelity, spatial extent, and uncertainty analysis: coupling high-resolution hydrodynamic simulations with thousands of Monte Carlo runs would take prohibitive amounts of time. The Munich team sidestepped this by replacing runtime coupling with pre-processed surrogates and lookup tables. The model comprises five modules, covering hydrological load, dike failure, hydrodynamics, inundation, and damage, and the central simulation runs in an average of just 0.76 seconds per sample on an ordinary desktop processor. That speed enabled a staggering 720,000 Monte Carlo simulation runs in the case study, something unthinkable with conventional coupled models.

The dike failure module illustrates the technical sophistication involved. Dikes can fail through hydraulic erosion during overtopping, geohydraulic uplift driven by piping in the foundation, or global static instability from internal seepage. The researchers built on limit state functions developed by Vorogushyn, comparing the stress acting on a dike with its resistance, both treated as random variables, and evaluated these through Monte Carlo simulation to produce fragility functions. In each simulation run, a breach resistance value is sampled for every dike segment by inverse transform sampling from the fragility function, and breach widths are drawn from a lognormal distribution. Because the Bavarian Danube was discretized into 1,784 river segments of 300 to 600 meters, the model can represent breaches at essentially any location along the river, capturing the so-called levee effect in which a failure near a densely populated area produces catastrophic damage.

Perhaps the most elegant innovation is the vector-based flood routing scheme. When a dike breaches or a detention basin activates, the resulting reduction in discharge propagates downstream, and ignoring this effect can badly overestimate downstream hazard. Instead of running a full 1D hydrodynamic model at every Monte Carlo step, the team derived translational and attenuation vectors from a set of pre-computed hydraulic simulations with artificial discharge reductions. The translational vector captures the mean travel time of a discharge reduction to each downstream segment, while the attenuation vector spreads the reduction over a symmetric temporal window that preserves the total reduction volume. When tested against an independent set of 40 hydraulic simulations, the vector-based approach showed low differences in peak discharge, timing, and water level compared with the full 1D model, especially when measured against the much larger discrepancy between 1D and 2D simulations.

The team demonstrated the framework on roughly 380 kilometers of the Bavarian Danube, a river whose hydrology is shaped by Alpine tributaries including the Iller, Lech, Isar, and Inn, and which flows past major industrial centers such as Ingolstadt and Regensburg. The test case evaluated a controlled detention basin near Riedensheim, which can store around 8 million cubic meters of floodwater with a maximum intake capacity of 170 cubic meters per second. The basin activates only when a 100-year flood is exceeded, diverting water into the polder to attenuate the flood wave and relieve downstream defenses. Flood scenarios were drawn from the ClimEx project, which produced 3,500 years of climate-based precipitation simulations under the RCP8.5 emission scenario, from which 72 flood events exceeding the 100-year threshold were identified and routed through the river network.

The results carry a message that should resonate far beyond Bavaria. The uncertainty in the individual loss exceedance curves, with and without the detention basin, is large, but the two curves are highly correlated because most uncertainties affect damages in both scenarios in the same way. As a consequence, the uncertainty in the risk reduction itself, the benefit of the measure, is far smaller than the raw curves would suggest, with a coefficient of variation of 25 percent. This correlation effect is a genuine insight for benefit-cost analysis: it means that even under deep uncertainty, the relative value of a mitigation measure can be estimated with usable confidence. The sensitivity analysis added another striking finding: uncertainties in the damage module, particularly for river sections rather than dike segments, dominate the variance in annual flood risk, echoing earlier work showing that the choice of damage model matters more than the parameters of the hydraulic models themselves.

The scenario analysis also probed the shadow of climate change. When the team linearly increased the flood occurrence rate by factors of two and three toward the year 2120, the expected benefit of the detention basin grew, since more frequent extreme events mean the basin is activated and useful more often, but the uncertainty surrounding that benefit widened as well. The authors are careful to note that their contribution is methodological rather than a definitive risk assessment for the Danube, and the preliminary case-study numbers remain confidential pending further work on climate inputs. They also acknowledge limitations: the model neglects flow velocity and inundation duration in damage estimation, cannot evaluate nature-based solutions that alter runoff generation, and does not yet explicitly model tributaries, though the modular design would allow high-resolution 2D simulations to be slotted in for locally acting measures.

What makes this work genuinely exciting is its scalability and its philosophy. Because runtime scales approximately linearly with the number of river segments, extending the model from the Bavarian Danube to the entire 2,850-kilometer river, which crosses ten countries, would increase computation by no more than a factor of ten, opening the door to the kind of cross-border, system-level flood management that decades of research have advocated but practice has rarely delivered. The main code has been released openly on GitHub and archived on Zenodo, inviting other catchments to adapt the workflow. As extreme floods intensify in a warming climate and public budgets tighten, a tool that tells decision-makers not only how much risk a measure removes but how confident we can be in that estimate, and which uncertainties are worth investing in to learn more, may prove as valuable as any concrete wall along a riverbank.

Subject of Research: Probabilistic fluvial flood risk modeling for quantifying the benefit of flood mitigation measures under aleatory and epistemic uncertainty

Article Title: A fluvial flood risk model for quantifying the benefit of mitigation measures under uncertainty

Article References: Ruf, M., Hoffmann, A., & Straub, D. (2026). A fluvial flood risk model for quantifying the benefit of mitigation measures under uncertainty. Natural Hazards and Earth System Sciences, 26(9), 4549-4568. https://doi.org/10.5194/nhess-26-4549-2026

Image Credits: AI Generated

DOI: 10.5194/nhess-26-4549-2026

Keywords: flood risk, fluvial flooding, Monte Carlo simulation, dike failure, detention basin, uncertainty quantification, benefit-cost analysis, Bavarian Danube, flood routing, fragility functions, flood damage assessment, risk management

At the heart of the volume lies a simple but consequential observation: rapid advances in artificial intelligence, blockchain, cloud computing, the Internet of Things and data analytics are reshaping the way financial institutions operate and deliver services. These technologies are not incremental improvements to legacy systems. They represent a structural shift in how value is stored, transferred, priced and protected. The book argues that alongside opportunities for greater efficiency, transparency and innovation, the same technologies are creating new challenges related to cybersecurity, regulation, ethics and digital trust, and that understanding both sides of the ledger is essential for anyone working in or studying modern finance.

Artificial intelligence and machine learning receive particular attention throughout the volume. In banking, these techniques now underpin credit scoring, algorithmic trading, customer service automation and, increasingly, the detection of financial fraud. Machine learning models can identify anomalous patterns across millions of transactions in real time, flagging suspicious behavior that would be invisible to manual review. Yet the contributors also confront the harder questions that come with algorithmic decision-making: how to ensure that automated credit decisions are fair, how to explain model outputs to regulators and customers, and how to maintain accountability when machines make choices that were once the preserve of human judgment. Ethical decision-making in AI-driven finance is treated as a first-order concern rather than an afterthought.

Blockchain and distributed ledger technologies form another pillar of the analysis. By allowing transactions to be recorded on tamper-evident, shared ledgers, blockchain promises reductions in settlement times, reconciliation costs and counterparty risk. The book situates these capabilities within the wider movement toward digital payments and tokenized assets, exploring how distributed architectures could reshape everything from cross-border remittances to trade finance. At the same time, the contributors acknowledge the regulatory ambiguity that still surrounds crypto-assets and smart contracts, and the technical hurdles, including scalability and interoperability, that stand between pilot projects and production-scale deployment.

Cloud computing and the Internet of Things extend the transformation beyond software into infrastructure. Cloud platforms give banks elastic computing power and enable the rapid deployment of new services, while connected devices generate streams of data that can inform insurance pricing, asset monitoring and hyper-personalized financial products. The volume’s treatment of hyper-personalized banking is especially timely: financial institutions can now tailor offers, advice and risk assessments to individual customers with a granularity that was unimaginable a decade ago. But personalization at this scale depends on vast data collection, which in turn raises questions about privacy, consent and the security of the cyber-physical systems that link physical devices to financial networks.

Cybersecurity and digital trust emerge as connecting threads across the entire book. As financial services migrate to digital channels, the attack surface for fraud and cybercrime expands in parallel. The contributors examine financial fraud detection as both a technical and an institutional problem, one that requires not only sophisticated analytics but also investor confidence, robust governance and coordinated regulatory oversight. The book’s chapters on risk management consider how institutions can build resilience into systems that are increasingly interconnected, automated and dependent on third-party cloud providers, and how regulators can keep pace with technologies that evolve faster than the rulebooks written for earlier eras.

Regulation itself is a major focus. The volume devotes sustained attention to fintech regulation, exploring the tension between fostering innovation and protecting consumers, markets and financial stability. Regulatory sandboxes, licensing regimes for digital-only banks and evolving standards for data protection all feature in the discussion. The contributors argue that the jurisdictions that strike the right balance between experimentation and oversight are likely to become hubs for financial innovation, while overly rigid frameworks risk pushing activity into less transparent channels. The book also connects regulation to ethics, emphasizing that trust is the fundamental asset of any financial system and that digital transformation succeeds only when customers believe their money and their data are safe.

Sustainability adds a further dimension to the analysis. Chapters on green banking examine how technology can support the transition to a lower-carbon economy, from digital processes that reduce paper and energy consumption to analytics that measure the environmental exposure of loan portfolios. The book positions green banking not as a niche marketing exercise but as a strategic imperative, driven by investor demand, regulatory pressure and the growing recognition that climate risk is financial risk. This perspective links the technological themes of the volume to the broader societal questions that will define the next decade of finance.

For portfolio management and investment services, the book surveys how data analytics and AI are changing the way assets are selected, allocated and monitored. Algorithmic strategies, alternative data sources and automated advisory platforms are democratizing tools that were once available only to large institutions, while also concentrating new forms of systemic risk when many market participants rely on similar models. The contributors consider what these developments mean for investor confidence, market efficiency and the responsibilities of asset managers in an era when machines execute a growing share of trading volume.

By bringing together conceptual analysis, contemporary research and practical perspectives, Reshaping Banking and Financial Services with Disruptive Technologies offers a comprehensive understanding of the digital transformation now underway in finance. The editors, Monica Gupta, Priya Jindal and Piyush Gupta, have assembled contributions spanning fintech, financial technology, digital transformation and the evolving banking landscape, making the volume relevant to researchers, financial-sector practitioners, policymakers and postgraduate students alike. The book is published by Bentham Science Publishers and carries the DOI 10.2174/97988988175341260101. As financial institutions worldwide grapple with the pace of technological change, the volume arrives as a timely map of both the promise and the peril of disruptive innovation in banking, a field in which the next disruption is always closer than it appears.

Subject of Research: The impact of disruptive technologies such as AI, blockchain and cloud computing on banking and financial services

Article Title: Reshaping banking and financial services with disruptive technologies

Article References: Reshaping banking and financial services with disruptive technologies. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: banking, fintech, artificial intelligence, blockchain, cloud computing, Internet of Things, data analytics, cybersecurity, financial regulation, digital payments, green banking, risk management

The study, authored by Di Teng of Harbin Finance University, addresses two stubborn problems that have limited earlier attempts to apply graph neural networks to financial compliance. The first is label sparsity. In a typical compliance dataset, confirmed instances of serious legal violations are so rare that a supervised model trained directly on them tends to overfit, memorizing the few known bad actors rather than learning generalizable patterns of risk. The second is a structural flaw in the standard Graph Attention Network, or GAT, one of the most widely used architectures for learning from networked data. In GAT, the attention weight assigned to a connection is computed in a way that is effectively static: for a given edge, the score depends only weakly on which node is asking the question. That rigidity makes it hard for the model to trace how legal risk actually propagates through a financial network, where the significance of a relationship changes dramatically depending on the perspective of the entity involved.

Teng’s framework unfolds in two stages. In the first, MoCo performs unsupervised pre-training on the unlabeled graph of financial entities. MoCo, originally developed for computer vision, works by maintaining two encoders: a query encoder that is updated by normal backpropagation, and a momentum encoder whose parameters drift slowly toward the query encoder according to the update rule in which the key parameters are replaced by a weighted blend of their previous values and the query parameters. The momentum encoder’s outputs are pushed into a large queue of negative samples, giving the model a vast dictionary of examples against which each new node can be compared. The training objective, an InfoNCE loss, pushes each node’s representation toward its positive counterpart and away from thousands of negatives, forcing the network to learn high-order semantic features without ever seeing a label. Crucially, the study adapts the queue mechanism to serve a compliance-specific purpose: buffering rare, high-risk legal event nodes so that their features are preserved rather than drowned out by the overwhelming majority of benign entities.

In the second stage, the pre-trained embeddings flow into GATv2, a refined attention architecture that fixes the static attention problem of its predecessor. GATv2 computes attention scores by applying a learnable linear transformation and a LeakyReLU non-linearity to the concatenation of two nodes’ features before taking the inner product with a learnable attention vector. This reordering of operations makes the attention mechanism strictly more expressive: the importance of a neighbor now genuinely depends on the query node, allowing the model to assign different weights to the same relationship depending on whose risk is being assessed. After normalization with a softmax function across each node’s neighborhood, the attention weights are used to aggregate neighbor features into an updated node representation, and multiple attention heads run in parallel to capture different facets of the risk landscape.

The full pipeline is trained with a composite loss function that reflects the realities of compliance work. The supervised component is Focal Loss, a formulation that down-weights easy, well-classified examples and concentrates the model’s gradient signal on the hard cases—precisely the meticulously disguised violations that fraud rings engineer to look ordinary. An auxiliary contrastive loss from the pre-training stage is retained during fine-tuning, weighted by a hyperparameter, to preserve the structure of the learned feature manifold, and an L2 regularization term guards against overfitting. When the model’s predicted probability for an entity exceeds a decision threshold calibrated by ROC analysis, the entity is flagged as high-risk, triggering a detailed legal compliance review such as an anti-money-laundering check, a contract compliance audit, or a litigation risk warning.

The framework was evaluated on two datasets. The first is the public Lending Club dataset of personal loans and defaults. The second is Fin-Law-CN, a proprietary heterogeneous graph built for the study containing roughly 20,000 entity nodes and 80,000 typed edges. Its nodes include borrowers, enterprises, guarantors, court-case records, and transaction events, while its edges capture borrower-enterprise links, ownership and control relations, guarantees, litigation involvement, and simulated supply-chain transactions. Legal-risk labels were mapped into three tiers—low risk, medium risk, and high risk—based on observable default, litigation, and compliance-warning outcomes, and the data were cleaned by merging duplicate entities and normalizing inconsistent identifiers before graph construction. The data were split 7:1:2 into training, validation, and test sets, and the model was benchmarked against GCN, GAT, GraphSAGE, RGCN, and the Heterogeneous Graph Transformer under a unified grid search over learning rates, dropout rates, and embedding dimensions.

The results were decisive. On Fin-Law-CN, where conventional models struggled with the intricate mix of relationship types, the proposed approach achieved an F1-Score roughly 4 to 6 percentage points higher than the baselines, with an AUC of 0.86. On the classification benchmarks, the MoCo-GATv2 model reached an AUC of 0.96, compared with 0.85 for standard GAT and 0.76 for GCN, and its precision-recall curve dominated across recall levels. Training curves showed the model converging to a loss of about 0.7 within roughly 100 epochs while validation accuracy climbed to approximately 0.89, against 0.85 for GAT and 0.79 for GCN—evidence that the combination is not only more accurate but also more stable during training.

The study’s ablation experiments dissected where the gains come from. Removing the MoCo pre-training module dropped accuracy, F1-Score, and AUC from roughly 98, 96, and 97 percent to about 90, 88, and 89 percent, confirming that self-supervised representation learning is critical when labels are scarce. Replacing GATv2 with standard GAT was even more damaging, sinking the metrics to around 85, 83, and 84 percent and underscoring the value of dynamic attention for tracing risk propagation. Dropping multi-head attention cost a few more points. Parameter sensitivity analyses showed performance peaking at embedding dimensions of 64 or 128, and the MoCo queue size experiments revealed a sweet spot: downstream F1 rose from about 0.80 at a queue of 256 to a peak of 0.932 near 16,384, then plateaued and slightly declined at larger sizes, suggesting diminishing returns and potential redundancy from excessive negative sampling. Attention-head analysis identified heads H1, H5, and H8 as the most influential in the shallow layers, while some heads contributed almost nothing, hinting at room for pruning.

Practical deployment considerations were addressed as well. On a single NVIDIA RTX 3090 GPU, the eight-head configuration required about 1.2 seconds per training epoch on Fin-Law-CN and sustained inference latency of roughly 15 milliseconds per subgraph—fast enough for real-time risk control. The architecture also incorporates a human-machine feedback loop: expert reviewers examine flagged high-risk events, their judgments re-label or update annotations in the graph database, and the model is retrained on the enriched data in a version-controlled cycle that continuously sharpens its accuracy.

The author is candid about the framework’s limits. Validation so far rests on one public and one proprietary dataset, both drawn from a single regulatory context, and cross-market testing in the United States, Europe, and beyond remains to be done. The experiments did not include dedicated adversarial attacks, deliberately disguised fraud-ring subsets, or tests of temporal propagation delays, and the model lacks a full explainable-AI module capable of producing legally auditable, case-level justifications for its flags—something regulators are likely to demand. Building and maintaining large heterogeneous financial-legal graphs also carries substantial deployment cost. Future work, the study notes, will incorporate temporal graph neural networks to capture how risk spreads across evolving enterprise networks, explainability methods such as GNNExplainer and SHAP-style attribution, federated learning for privacy-preserving collaboration across institutions, and robustness testing under structural adversarial attacks. If those extensions succeed, the combination of contrastive pre-training and dynamic graph attention could become a standard weapon in the fight against the hidden networks behind financial crime.

Subject of Research: Machine learning methods for legal prevention and control of financial credit risk

Article Title: Research on legal prevention and control of financial credit risk based on MoCo and GATv2 algorithm

Article References: Teng, D. (2026). Research on legal prevention and control of financial credit risk based on MoCo and GATv2 algorithm. Discover Artificial Intelligence, 6(1), Article 1356. https://doi.org/10.1007/s44163-026-02297-7

Image Credits: AI Generated

DOI: 10.1007/s44163-026-02297-7

Keywords: MoCo, GATv2, graph neural networks, financial credit risk, contrastive learning, legal risk mitigation, anti-money laundering, fraud detection, label sparsity, heterogeneous graphs, focal loss, risk management

To understand why this matters, it helps to look at what most algorithmic trading systems actually do. The majority rely on price-based indicators such as moving averages, momentum oscillators, and volatility measures, which summarize what has already happened in the market. Some more recent systems have bolted on sentiment signals, but the authors of the new study argue that these are often simplified, reducing rich textual information to a single positive-or-negative score that strips away the context in which the sentiment was expressed. A headline saying that a company beat earnings expectations carries very different weight depending on whether it appears in a major financial outlet during a broad market rally or in an obscure forum during a panic. The framework developed by the Indian team was designed specifically to preserve that context, extracting sentiment features that capture both market relevance and directional tone before feeding them to the learning agent.

The technical heart of the system is proximal policy optimization, an algorithm that has become one of the workhorses of modern reinforcement learning. In reinforcement learning, an agent learns by interacting with an environment, taking actions, and receiving rewards or penalties that shape its future behavior. In a trading setting, the actions are typically buy, sell, or hold decisions, and the reward is tied to portfolio performance. PPO belongs to a family of methods called policy-gradient algorithms, which directly adjust the parameters of the policy, the mapping from market states to actions, in the direction that increases expected reward. What distinguishes PPO is its use of a constrained update rule that prevents the policy from changing too drastically between iterations, a property that makes training far more stable in noisy environments. Financial markets, with their erratic feedback and non-stationary dynamics, are about as noisy as environments get, which makes this stability property especially valuable.

This stability advantage showed up clearly in the study’s comparisons. The researchers found that policy-gradient methods like PPO consistently outperformed value-based reinforcement learning models, a rival family of algorithms that learn to estimate the long-term value of being in a particular state before deriving a trading policy from those estimates. Value-based methods, such as those built on Q-learning principles, can struggle when the relationship between states and rewards shifts abruptly, as it does when markets move from calm to crisis. Policy-gradient approaches, by contrast, optimize the behavior itself and appeared to adapt more gracefully across changing market regimes. The authors report that the sentiment-aware PPO agent delivered better risk-adjusted performance, meaning it earned returns that were more favorable relative to the volatility and drawdowns it endured, a metric that professional traders tend to care about far more than raw profit alone.

A crucial part of the work lies in how realistically the trading environment was constructed. Many academic trading simulations are criticized for ignoring the frictions that erode real-world profits, effectively training agents in a fantasy market where trading is free and positions can be scaled without limit. The Junagadh team deliberately built those frictions in. Their environment imposes transaction costs on every trade, enforces position limits that cap how much exposure the agent can take, and applies cooldown periods that prevent the kind of rapid-fire churn that would be impractical or prohibited in live markets. Technical indicators are also incorporated into the state representation, giving the agent access to the same quantitative signals that human traders watch. By forcing the learning algorithm to operate under these constraints, the researchers aimed to produce policies that would translate more credibly from backtest to practice, rather than exploiting loopholes that exist only in idealized simulations.

The sentiment pipeline itself reflects the current moment in artificial intelligence, in which large generative language models trained on vast text corpora can extract nuanced meaning from unstructured language. The study situates its contribution within a rapidly growing body of research that connects language models to finance. Recent work has explored whether models like ChatGPT can forecast stock price movements, and technical reports such as FinRL-Llama have experimented with integrating LLM-based sentiment analysis directly into financial reinforcement learning pipelines. Surveys of large language models in equity markets document how quickly this field is moving. What the new study adds is a systematic evaluation of how context-aware sentiment, rather than crude sentiment scores, changes the behavior and performance of a reinforcement learning trader when everything else in the environment is held constant.

The evaluation spanned multiple assets across both equity and index markets, giving the authors a way to test whether the benefits of sentiment awareness generalize beyond a single ticker or market structure. Equity markets, where individual stocks respond to company-specific news, and index markets, which aggregate the behavior of many firms, present different information environments, and a framework that works in both is more convincing than one tuned to a single instrument. The consistent pattern in the results, the authors report, is that incorporating context-aware sentiment information improves the agent’s adaptability across market regimes. In practical terms, this means the agent appears to change its behavior appropriately when the market shifts from bullish to bearish conditions or from low to high volatility, rather than continuing to apply strategies that worked in the previous regime, a failure mode that has plagued many quantitative systems.

The findings arrive at a time when the intersection of generative AI and finance is attracting intense attention from both researchers and practitioners, and they carry implications for how trading systems of the future might be designed. If the study’s conclusions hold up under further scrutiny, the lesson is not simply that sentiment data is useful, which many traders already believe, but that the way sentiment is computed and integrated matters enormously. A single scalar sentiment score may wash out exactly the contextual signals that give language its predictive power, while richer, context-sensitive representations extracted by generative models can give a learning agent genuinely new information about the state of the market. Combined with the stability of policy-gradient optimization and a realistic simulation environment, this could point toward trading systems that are more robust than either pure price-based algorithms or earlier sentiment-driven attempts.

Caveats remain, as they do with any backtested trading research. The authors note that the data generated and analyzed in the study are available from the lead author upon reasonable request but are not yet publicly available due to ongoing research, which means independent verification will depend on access to the underlying datasets. On the other hand, the team has provided the source code as supplementary material with the published article for research and reproducibility purposes, a step that should help other groups examine and extend the framework. The authors, who received no external funding for the work and declare no competing interests, emphasize that the goal is adaptability and risk control rather than spectacular returns. As generative language models continue to improve at understanding the subtleties of financial discourse, studies like this one suggest that the next generation of trading algorithms may be distinguished not by how fast they process prices, but by how well they read the story the market is telling about itself.

Subject of Research: Sentiment-aware reinforcement learning using generative language models for algorithmic trading

Article Title: Sentiment-aware reinforcement learning for algorithmic trading

Article References: Fuletra, J., Patel, H., & Panchal, S. (2026). Sentiment-aware reinforcement learning for algorithmic trading. SN Social Sciences, 6(10), Article 463. https://doi.org/10.1007/s43545-026-01682-4

Image Credits: AI Generated

DOI: 10.1007/s43545-026-01682-4

Keywords: reinforcement learning, algorithmic trading, proximal policy optimization, sentiment analysis, generative AI, large language models, quantitative finance, financial markets, machine learning, risk management, behavioral finance, trading algorithms

The conference was hosted by the Shidler College of Business at the University of Hawaii at Manoa and co-organized by China Finance Review International (CFRI), Modern Risk Society (MRS), and the Dishui Lake Advanced Institute of Finance at Shanghai University of Finance and Economics. This transpacific partnership reflects a broader trend in academic finance: the recognition that risk, in its many modern forms, does not respect national boundaries or disciplinary silos. The collaboration brought together journal editors, database providers, and research institutes from both sides of the Pacific, creating a forum in which methodological debates could unfold against the backdrop of a rapidly changing technological and environmental landscape. For many attendees, particularly early-career researchers, the event served as both a showcase and a matchmaking opportunity for future collaborations.

The scientific program opened with a pair of special panel discussions that confronted the practical challenges of applying artificial intelligence in financial research and practice. In the “Data, AI and Applications” roundtable, scholars Rui Dai of WRDS, Jian Chen of Fudan University, Mehmet Saglam of the University of Cincinnati, and Yuehua Tang of the University of Florida debated the boundaries of AI-enabled finance under the moderation of Xuanjuan Chen of the Dishui Lake Advanced Institute of Finance at Shanghai University of Finance and Economics. The discussion ranged from intelligent customer service systems to large-scale text mining, and the panelists converged on a point that resonates far beyond the conference hall: in core risk control areas, algorithms must still defer to professional judgment. The exchange captured a tension now running through the entire discipline, as machine learning tools demonstrate remarkable pattern-recognition capabilities while regulators and risk managers grapple with questions of accountability, interpretability, and model risk.

A second panel, titled “Let’s Talk AI,” shifted the spotlight to a younger generation of researchers. Jiakai Chen of the University of Hawaii, Eileen Zhang of Rutgers University, Bochen Li of Monmouth University, and Zhao Wang of Capital University of Economics and Business demonstrated the diverse possibilities of AI tools in cutting-edge financial research. Their presentations spanned the interpretation of Federal Reserve policy communications, the analysis of news narratives, and generative information extraction, illustrating how natural language processing and related techniques are transforming the raw material of financial economics. Where researchers once relied on coarse textual classifications or manually coded event studies, they can now parse central bank statements, earnings calls, and news flows at scale, extracting measures of sentiment, uncertainty, and narrative momentum that feed directly into asset pricing and risk models. The session offered a glimpse of a methodological shift that is reshaping empirical finance from the ground up.

The intellectual centerpiece of the opening day arrived with the keynote address. Hosted by Professor Qianqiu Liu, Chair of the Finance Department at the University of Hawaii, the keynote was delivered by Professor Torben G. Andersen of Northwestern University, one of the most influential figures in the empirical study of financial volatility. His talk, “Inference for Option-Implied Risk Measures,” tackled a deceptively simple problem with deep practical consequences: how to extract reliable risk measurements from option prices when those prices are contaminated by measurement error. Option-implied measures of risk are foundational to modern risk management, informing everything from value-at-risk calculations to stress testing and derivative pricing. Yet option markets are imperfect, and the errors embedded in observed prices can propagate through downstream estimates in ways that distort inference.

Andersen’s proposed solution is a new nonparametric estimation method that improves the handling of correlations across option pricing errors. The technical significance lies in what nonparametric approaches avoid: rather than imposing a rigid functional form on the relationship between option prices and underlying risk, the method lets the data speak while explicitly accounting for the error structure that contaminates observations. By modeling the cross-correlations of pricing errors more effectively, the approach yields risk measures with more reliable statistical properties, providing practitioners with a sturdier theoretical tool for risk management. For an audience steeped in the uncertainties of AI-driven markets and climate-exposed portfolios, the message was clear: even as finance embraces new technologies, the discipline’s progress still depends on rigorous measurement and careful inference at the most fundamental level.

The conference also served as a stage for honoring sustained excellence in risk research. At an awards ceremony hosted by Professor Wei Huang of the University of Hawaii, ten distinct honors were presented, including the Pacific-Basin Finance Journal Research Excellence Award, the Global Association of Risk Professionals Research Excellence Award, the China Finance Review International Research Excellence Award, the Asia-Pacific Journal of Risk and Insurance Research Excellence Award, the Sinofin CCER Database Research Excellence Award, and the Southwestern University of Finance and Economics Research Excellence Award. Modern Risk Society conferred its own slate of recognitions, among them the Edward Kane Memorial Research Excellence Award, the Contribution Award, the Distinguished Leadership Award, and the Lifetime Achievement Award. Professor Wenfeng Wu, Editor-in-Chief of CFRI and a faculty member at the Antai College of Economics and Management, Shanghai Jiao Tong University, presented the MRS Lifetime Achievement Award and the CFRI Research Excellence Award, underscoring the journal’s role in promoting interdisciplinary research on risk and finance.

Beyond the formal sessions, the conference cultivated the informal exchanges where collaborations are often born. An evening welcome reception at the Shidler College of Business set a relaxed tone in the Aloha atmosphere of Honolulu, and the following night the gathering moved to the Imin International Conference Center for the Presidents’ Reception. Amid the garden scenery of the University of Hawaii, attendees deepened their conversations over dinner, while a lucky draw featuring exclusive tickets designed and produced by CFRI added a moment of levity to the proceedings. These social rituals, easily dismissed as peripheral, perform a genuine scientific function: they lower the barriers between senior professors and doctoral students, between journal editors and authors, and between researchers from different national traditions, enabling the candid conversations that formal presentations rarely permit.

The parallel sessions of July 18 and 19 formed the analytical heart of the meeting. Across 64 sessions, the 199 presented papers ranged from subversive challenges to long-standing assumptions in financial economics to brave explorations of entirely new research paradigms. The heavy weighting of AI and climate topics was no accident. Artificial intelligence is simultaneously a subject of study, as markets adopt algorithmic trading and machine learning-based credit scoring, and a methodological instrument, as researchers deploy it to parse text, forecast volatility, and model complex dependencies. Climate finance, meanwhile, has moved from the margins to the mainstream as physical risks and transition risks reshape asset valuations, insurance markets, and regulatory frameworks. Many attendees remarked that the conference not only refreshed their thinking but also helped them find like-minded research partners, a testament to the organizing committee’s curation of the program.

The event also carried an institutional dimension. Throughout the three days, the CFRI journal team operated a booth at the venue, introducing the journal’s features, submission process, and recent key topics to participating scholars, distributing promotional materials, and inviting high-quality manuscripts. The team also visited the Shidler College of Business for a special meeting with Professor Qianqiu Liu, exchanging in-depth views on frontier research in finance, international academic cooperation, and journal co-development. CFRI, an English-language journal hosted by Antai College of Economics and Management at Shanghai Jiao Tong University and published by Emerald Publishing Group, has grown steadily since its founding in 2011 and is now indexed in more than twenty international databases, with a 2025 Impact Factor of 10.8. As the organizing committee closed the meeting with the message “Until next time, stay curious,” the gathering in Honolulu made a persuasive case that the study of risk, whether driven by algorithms or by a changing climate, has become one of the defining scientific projects of the decade.

Subject of Research: Artificial intelligence and climate risk in international financial research

Article Title: 2026 MRS International Risk Conference successfully concludes in Hawaii

Article References: 2026 MRS International Risk Conference successfully concludes in Hawaii. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: artificial intelligence, climate finance, climate risk, risk management, financial research, option-implied risk measures, machine learning, academic conference, China Finance Review International, University of Hawaii, text mining, volatility estimation

Researchers from Tsinghua University, Nanjing University of Aeronautics and Astronautics, and other institutions developed a closed-loop modeling and assessment framework based on multi-attribute decision-making. The study highlights that existing competency models, which were constructed around deterministic systems and process control, fail to explicitly address the unique risks associated with AI integration. These risks include black-box opacity, the difficulty of allocating accountability for AI-driven errors, and the need for traceable review of AI outputs. By failing to account for these factors, traditional models leave a critical void in the governance of modern aerospace projects.

To construct the new model, the research team began with a qualitative exploration, conducting RepGrid interviews with thirty experienced aerospace engineering managers. This process allowed the researchers to identify the core characteristics and challenges that managers face when integrating AI into their workflows. Following the interviews, the team applied principal component analysis to an exploratory sample to identify underlying patterns in the data. This statistical approach revealed five distinct dimensions of competency that are essential for effective AI management in the aerospace sector.

The resulting competency model comprises five primary dimensions: AI risk control and management, lifecycle AI coordination, AI cognitive readiness, AI compliance and safety assurance, and aerospace AI scenario enablement. Within these dimensions, the researchers identified a total of twenty specific attributes that define a manager’s proficiency. To validate the structure of this model, the team performed confirmatory factor analysis on a separate validation sample. The results demonstrated that both the first-order five-factor model and the second-order hierarchical model achieved satisfactory fit indices, confirming the reliability and validity of the proposed framework.

Understanding how these competency dimensions interact is crucial for effective management, so the researchers used the DEMATEL method to map the influence network among the attributes. Fourteen experts with extensive experience in aerospace engineering management provided pairwise evaluations of the direct influence relationships. The analysis revealed that AI risk control and management acts as the upstream driving dimension, indicating that safety and risk boundaries must be established before AI scenarios can be fully deployed. This finding aligns with the high-reliability culture of the aerospace industry, where zero-failure requirements dictate a safety-first approach.

The lifecycle AI coordination dimension emerged as the central hub, connecting all other dimensions within the network. This centrality underscores the importance of integrating AI considerations throughout the entire project lifecycle, from initial design to final delivery. In contrast, the aerospace AI scenario enablement dimension was identified as an outcome-oriented downstream dimension, meaning it is influenced by the preceding governance and coordination efforts. The DANP method was then used to derive global weights for each attribute, highlighting the relative importance of specific skills. Attributes such as AI use-case identification, critical evaluation of AI outputs, and human-AI decision boundary management ranked as the most critical competencies.

To test the practical applicability of the framework, the researchers applied the improved VIKOR method to assess five real candidates for aerospace engineering management roles. Four independent experts evaluated these candidates across the twenty identified attributes. The assessment process demonstrated high inter-rater reliability, with an average intraclass correlation coefficient of 0.912, indicating strong consistency among the evaluators. The VIKOR method calculated the weighted total gap and maximum weighted regret for each candidate, providing a nuanced view of their strengths and weaknesses relative to the ideal competency profile.

The results of the candidate assessment showed that one candidate, designated H5, performed optimally on both the total gap and maximum regret metrics. However, the analysis also revealed that the gap between the top candidate and the third-ranked candidate did not reach the VIKOR acceptable advantage threshold. This finding suggests that both candidates should proceed to the final review stage, with targeted verification conducted based on the specific bottleneck characteristics identified in the assessment. This approach ensures that the selection process is not only data-driven but also robust against minor variations in decision preferences.

Sensitivity analysis further validated the stability of the ranking results. The researchers tested the model under different risk preference coefficients and found that the candidate rankings remained stable regardless of the specific decision preference settings. This robustness is a significant advantage for the framework, as it provides a traceable and interpretable decision-support tool that can be reliably used in high-stakes selection processes. The framework’s ability to identify specific bottlenecks in a candidate’s competency profile also offers valuable insights for training planning and career development.

This new competency model provides a structured approach to addressing the governance challenges of AI in aerospace engineering. By explicitly defining the interdependencies between risk control, lifecycle coordination, and safety assurance, the framework offers a clear roadmap for managers to navigate the complexities of AI integration. As deep-space missions continue to advance, the ability to assess and develop these specific competencies will be critical for ensuring the delivery quality, safety, and compliance of future exploration efforts. The study thus represents a significant step forward in the field of engineering management, providing a rigorous tool for the AI era.

Subject of Research: AI competency assessment framework for aerospace engineering managers

Article Title: AI competency model for aerospace engineering managers: a multi-attribute decision-making approach

Article References: AI competency model for aerospace engineering managers: a multi-attribute decision-making approach. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: aerospace engineering, artificial intelligence, competency model, project management, deep-space exploration, decision-making, safety compliance, risk management, human-AI interaction, space technology, competency, model

The core problem is well understood among quantitative finance practitioners. Markowitz’s framework asks investors to weigh expected returns against covariance, the statistical measure of how assets move together, and to select the allocation that delivers the best return per unit of risk. But the expected returns and covariances it consumes are typically estimated from historical data using backward-looking statistical assumptions. When markets undergo structural breaks, sudden regime changes, or volatility spikes of the kind seen during the COVID-19 pandemic, those estimates become stale almost overnight. Portfolios that looked optimal on paper can turn fragile, forcing panicked rebalancing at the worst possible moments.

Confidence-MPT addresses this fragility by inserting a Long Short-Term Memory neural network into the heart of the optimization pipeline. LSTM networks are a class of recurrent neural architectures specifically designed to learn long-range temporal dependencies in sequential data, making them well suited to financial time series where patterns can span weeks or months. In the proposed system, the LSTM is trained on historical price data to produce three simultaneous outputs: a forecast of future returns, an estimate of future volatility, and a novel quantity the authors call predictive confidence, which captures how certain the network is about its own predictions.

The third output is what distinguishes this approach from earlier attempts to marry machine learning with portfolio construction. Rather than blindly feeding neural network forecasts into the Markowitz optimizer, Confidence-MPT dynamically adjusts the expected return vector and the covariance matrix according to both the confidence level and the volatility forecasts. When the network is highly confident and predicted volatility is low, the optimizer can lean more aggressively into the forecasts. When confidence drops or volatility is expected to rise, the framework pulls back, effectively shrinking the influence of uncertain predictions on the final allocation. The result is a portfolio construction process that knows not only what it expects, but how much it trusts its own expectations.

Technically, the workflow proceeds in two stages. First, a scaling and sequencing routine converts raw price series into normalized input windows of fixed length, which are used to train the multi-output LSTM. Second, the trained model supplies confidence-adjusted estimates of the mean return vector and covariance matrix to a classical optimizer, which recomputes portfolio weights. The system then compares the new weights against the previous allocation and applies decision thresholds: if a weight change exceeds a buy threshold, the framework signals a purchase; if it falls below a sell threshold, it signals a sale; otherwise the position is held. This thresholding mechanism is a deliberate design choice aimed at controlling turnover, the costly churn of buying and selling that erodes returns through transaction costs.

To test whether this added intelligence actually pays off, the researchers benchmarked Confidence-MPT against a battery of established strategies using fifteen years of daily stock data from the Dow Jones Industrial Average, sourced from publicly available Yahoo Finance data. The comparison set included the Constant Rebalanced Portfolio, Buy and Hold, Exponentiated Gradient, Dynamic CRP, and the Universal Portfolio, a well-known family of online portfolio selection algorithms that have long served as reference points in the literature. The evaluation was reinforced with Monte Carlo simulations and efficient frontier visualizations, which map out the full landscape of risk-return trade-offs available to an investor under each approach.

The findings are notable for what they do and do not claim. Confidence-MPT did not produce the highest raw returns among the strategies tested, and the authors are candid about that. What the framework delivered instead was a compelling trade-off between performance, risk control, and turnover. Its allocations remained more stable over time than those of competing methods, and it exhibited improved drawdown characteristics during turbulent periods, meaning the peak-to-trough losses investors would have endured were comparatively contained. In practical terms, a portfolio that whipsaws between extreme allocations may win on paper in backtests, but the version that holds steady through a crisis is often the one that survives in the real world.

The Monte Carlo analysis and frontier plots add a further layer of insight by showing how incorporating predictive confidence reshapes the entire risk-return landscape rather than merely shifting a single point on it. By adjusting the inputs to the optimizer, confidence weighting changes which combinations of assets appear efficient, altering the position of the maximum Sharpe ratio portfolio and the minimum-risk portfolio alike. The appendices to the paper present these visualizations in detail, highlighting optimal portfolios under both baseline and alternative scenario assumptions, and demonstrating how the adjusted Sharpe distributions differ when LSTM-based predictions are used in place of purely historical estimates.

The study arrives amid a broader wave of research applying machine learning to investment management, from deep reinforcement learning agents that learn trading policies directly from market data, to hybrid systems that use support vector machines for stock pre-selection before optimization, to knowledge-distilled architectures that connect Markowitz optimization with Bellman equations. A systematic literature review published in 2025 catalogued the rapid expansion of deep learning applications in portfolio management, and recent work has explored robustifying Markowitz under model ambiguity and non-Gaussian return distributions. What sets Confidence-MPT apart within this crowded field is its explicit focus on predictive certainty as a first-class input to optimization, rather than treating the neural network as a black-box oracle whose forecasts are accepted at face value.

For investors and asset managers, the appeal of the framework lies in its realism and scalability. It relies on freely available price data, standard neural network tooling, and a classical optimization core that is well understood by practitioners, making it a plausible candidate for real-world deployment rather than a laboratory curiosity. The authors, Ehsan Ameri, Majid Mirzaee Ghazani, and Donya Rahmani, position the method as a tool for robust portfolio construction in dynamic and uncertain market environments, and their contribution underscores a growing consensus in data-driven finance: the future of portfolio optimization may lie not in replacing the classical frameworks, but in teaching them when to trust the machines that advise them.

Subject of Research: Machine learning-enhanced portfolio optimization integrating LSTM predictions into Modern Portfolio Theory

Article Title: Confidence-MPT: a machine learning-enhanced framework for robust portfolio optimization

Article References: Ameri, E., Ghazani, M. M., & Rahmani, D. (2026). Confidence-MPT: a machine learning-enhanced framework for robust portfolio optimization. Neural Computing and Applications, 38(19), Article 763. https://doi.org/10.1007/s00521-026-12515-z

Image Credits: AI Generated

DOI: 10.1007/s00521-026-12515-z

Keywords: portfolio optimization, Modern Portfolio Theory, LSTM neural networks, machine learning, quantitative finance, risk management, deep learning, covariance estimation, drawdown, Monte Carlo simulation, Dow Jones Industrial Average, financial engineering

The study, authored by Irlenys Josefina Tersek Rodríguez and published open access in October 2026, develops a construct called AI-supported decision structures: the formal organizational arrangements through which AI-generated cybersecurity inputs become admissible, reviewable, and actionable within decision processes. The framing deliberately shifts attention away from the AI tool itself and toward the decision architecture that surrounds it. In cybersecurity, decisions about escalating an alert, isolating a host, deferring a patch, accepting residual risk, or invoking crisis procedures are made under time pressure and uncertainty, and they carry legal, operational, reputational, and continuity consequences. Cybersecurity decision-making, the paper insists, is therefore not merely technical problem solving but a governance process in which evidence, authority, accountability, and coordination must be aligned.

To build the framework, the study draws together several research streams that have rarely been integrated. Information systems governance explains how organizations allocate decision rights and design structural, procedural, and relational mechanisms such as committees, policies, and escalation procedures. Cybersecurity governance research emphasizes board oversight, risk ownership, and incident escalation. Responsible AI scholarship contributes principles of transparency, accountability, explainability, and human oversight, now reinforced by regulation such as the European Union’s AI Act, which ties high-risk AI systems to risk management, documentation, record keeping, and monitoring. Research on AI and organizational decision-making shows that AI reshapes search, speed, scale, explainability, and delegation, while resilience research describes how organizations anticipate, cope with, and adapt to adversity. What remains underdeveloped, the paper argues, is an account of how AI-generated analytical inputs move from machine output to legitimate organizational evidence.

The gap is sharpest in cybersecurity itself, which the author describes as a particularly demanding context for AI-supported governance for five reasons. Decisions are time-sensitive, since delayed action may allow an intrusion to expand while premature action may disrupt operations. Evidence is probabilistic and incomplete, requiring interpretation of alerts, logs, and threat indicators. Decisions cut across security operations, infrastructure, legal, compliance, business continuity, and senior management. Actions must be retrospectively defensible to auditors, regulators, boards, customers, or courts. And the domain is surveillance-intensive, depending on continuous monitoring of users, endpoints, networks, identities, and behavior. In this setting, AI may increase detection and prioritization capacity, but it can also produce false positives, false negatives, automation bias, alert fatigue, opaque evidence trails, and unclear responsibility.

The proposed construct comprises five interrelated dimensions, which the author presents as a governance configuration rather than a checklist. Validation refers to routines that filter AI-generated inputs for reliability and contextual relevance before they influence authorization, including checks on data quality, model suitability, confidence thresholds, and review by qualified personnel. Control refers to rules governing how inputs may be used, challenged, escalated, or overridden, including permissions, review gates, segregation of duties, and exception procedures. Institutional embedding incorporates AI outputs into recurring routines such as vulnerability review meetings, security operations center triage forums, risk committees, and post-incident reviews. Accountability ensures that someone remains responsible for the final decision and that the influence of AI-generated inputs can later be reconstructed. Substantive human oversight, finally, preserves the meaningful capacity of qualified actors to interpret, contest, contextualize, and override machine outputs.

Each dimension operates through a distinct theoretical mechanism. Validation works through evidentiary filtration, reducing the probability that unreliable or decontextualized outputs enter formal decisions as credible evidence. Control works through procedural constraint, clarifying when AI inputs may influence decisions and when human or committee review is required. Institutional embedding works through routinized coordination, turning isolated analyst artifacts into shared signals that support collective prioritization and learning. Accountability works through responsibility preservation, preventing responsibility from being displaced onto the AI system or diffused across technical and managerial actors. Oversight works through judgmental correction, mitigating automation bias when outputs are incomplete, misleading, or organizationally inappropriate. The paper formalizes these mechanisms in five propositions linking the dimensions to decision quality, decision legitimacy, coordinated response, and organizational resilience.

Crucially, the framework introduces theoretical tension rather than a simple prescription that more governance is always better. Stronger validation may improve reliability while slowing response. Stronger control may improve legitimacy while reducing improvisational flexibility during a live incident. Institutional embedding may improve coordination while normalizing surveillance and unquestioned reliance on executive dashboards. Accountability may improve defensibility while encouraging defensive documentation. Human oversight may correct AI outputs while becoming purely symbolic when workload, hierarchy, or time pressure prevents meaningful challenge. The effect of each dimension is also conditional: validation matters most when threat conditions are ambiguous, control matters most in regulated or audit-intensive settings, and oversight matters most when reviewers possess genuine domain competence and actual authority to override the machine.

The study also connects cybersecurity to the broader politics of digital surveillance. Because organizations protect digital assets partly by observing systems, networks, identities, and behavior, AI-supported security tools intensify organizational visibility by classifying behavior, detecting deviations, and producing prioritized alerts. The author, drawing on recent work on digital surveillance governance, argues that this does not make security monitoring illegitimate, but it does mean governance must address the dual character of AI-generated inputs: they can support anticipation and rapid response while simultaneously expanding surveillance capacities, encoding classifications of risky behavior, and shaping how employees or events are treated. AI-supported decision structures, on this reading, must specify not only how AI outputs support security action but also how the surveillance capacities underlying those outputs remain transparent, bounded, reviewable, and accountable.

Organizational resilience serves as the framework’s core outcome, understood through the capability-based triad of anticipation, coping, and adaptation. AI-supported decision structures contribute to anticipation when validated inputs improve early warning, vulnerability prioritization, and risk visibility. They contribute to coping when control mechanisms and embedding connect signals to escalation paths, response authority, and coordinated action during disruption. They contribute to adaptation when accountability and documentation make it possible to reconstruct how AI inputs influenced decisions and to revise models, thresholds, procedures, and training after incidents. But the resilience claim is explicitly conditional: the same structures can weaken resilience when they produce overreliance, brittle procedures, excessive centralization, or surveillance practices that erode trust and reduce reporting behavior.

The practical implications are pointed. AI adoption, the paper warns, should not be confused with governance maturity: organizations may invest heavily in models, data pipelines, and AI-enabled platforms while underinvesting in the structures required to validate, challenge, document, and justify the outputs those systems produce. Managers should specify who may rely on AI-generated inputs, what review is required before action, how disagreement is handled, what must be documented, when escalation is mandatory, and where final responsibility remains. The framework is most applicable to governance-intensive, high-stakes settings such as cybersecurity, financial fraud detection, clinical decision support, credit-risk assessment, and critical infrastructure monitoring, and less relevant to low-stakes uses like routine chatbots. As a conceptual contribution, it does not empirically test its propositions, and the author acknowledges that the five dimensions may overlap in practice and that validated measurement instruments remain future work. Even so, the study offers a precise foundation for one of the defining governance questions of the AI era: how machine-generated judgment becomes accountable human decision.

Subject of Research: AI-supported cybersecurity governance and organizational resilience

Article Title: Toward an integrative theoretical model of AI-supported cybersecurity governance and organizational resilience

Article References: Rodríguez, I. J. T. (2026). Toward an integrative theoretical model of AI-supported cybersecurity governance and organizational resilience. Discover Artificial Intelligence, 6(1), Article 1322. https://doi.org/10.1007/s44163-026-02050-0

Image Credits: AI Generated

DOI: 10.1007/s44163-026-02050-0

Keywords: artificial intelligence, cybersecurity governance, AI governance, decision structures, human oversight, organizational resilience, digital surveillance, accountability, incident response, EU AI Act, automation bias, risk management

The core problem, the authors explain, lies in the unusual properties of modern machine learning systems. Traditional software assurance grew up around deterministic programs: given the same inputs, the same outputs follow, and engineers can trace behavior through explicit lines of code. AI systems break these assumptions. They are non-deterministic, meaning identical inputs may yield different outputs across runs or model versions. They are deeply data-dependent, so their behavior shifts with the distributions of the data they are trained and operated on. And they carry evolving vulnerabilities, from adversarial attacks to silent model degradation, that static, one-off certification approaches were never designed to catch. Assurance methods borrowed from safety-critical engineering, such as the certification regimes used in aviation, assume a stability that learning systems simply do not possess.

Khadka and Maple frame their review around a crucial conceptual move. Rather than treating safety, fairness, robustness, privacy, explainability, and accountability as separate silos, each with its own framework, checklist, and community, they argue these should be understood as assurance goals that only become meaningful when operationalised through five cross-cutting commitments. The first is testable claims: any assurance statement about an AI system must be phrased in a way that can, in principle, be verified or falsified. The second is evidence obligations: whoever makes the claim must be able to produce the artifacts, test results, documentation, and monitoring data that support it. The third is lifecycle validity, the recognition that evidence gathered before deployment decays as models, data, and operating contexts change. The fourth is accountability and recourse, ensuring that when systems fail, affected parties can identify responsible actors and obtain redress. The fifth is interoperability, so that evidence and assurance artifacts can move across jurisdictions, sectors, and toolchains without being rebuilt from scratch.

Using these five commitments as an analytical lens, the researchers examined the leading AI assurance initiatives now shaping global practice. Chief among them is the NIST AI Risk Management Framework, the United States benchmark that organises AI risk work into govern, map, measure, and manage functions. They also assessed international guidance including the toolkits developed under the Organisation for Economic Co-operation and Development and the G7, along with practitioner-facing resources and representative research-led approaches. The verdict is nuanced. These initiatives have been valuable in establishing shared vocabulary and elevating AI risk to boardroom and ministerial attention, but the review finds that they frequently stop short of specifying the deployment-facing practices that credible assurance in real-world settings demands. High-level principles abound; concrete, auditable procedures for a hospital, a bank, or a transport operator remain scarce.

This gap between regulatory intent and operational reality is the review’s central finding. A regulator may mandate that an AI system be fair, but fairness itself fractures into multiple incompatible mathematical definitions, demographic parity, equalised odds, calibration across groups, and choosing among them is a context-sensitive judgment, not a technical formality. Similarly, requirements for explainability or robustness often lack the specified metrics, evidence formats, and thresholds that would allow an auditor to determine compliance. The result is a landscape in which organisations can declare alignment with principles while the underlying assurance work remains shallow, inconsistent, or unverifiable. The authors warn that this under-specification is not a cosmetic problem; it undermines the very credibility that assurance regimes exist to provide.

To make these abstractions concrete, the paper develops illustrative assurance cases across three high-impact domains. One sketches how explainability might be assured for an AI-enabled skin cancer detection system, tracing the chain from clinical claims about model outputs to the evidence a clinician and a regulator would need. Another addresses fairness in an AI-driven financial decision-making system, where lending outcomes must be demonstrably free of unjustified bias across protected groups. A third considers safety for an AI-enabled transportation system, where the consequences of failure are measured in lives. These worked examples show how the five commitments interact in practice: a fairness claim is only as strong as the evidence behind it, that evidence is only valid across the system’s lifecycle if monitoring continues after deployment, and recourse mechanisms must exist for individuals harmed by erroneous decisions.

The review also surveys the growing ecosystem of toolkits, platforms, benchmarks, and evaluation suites that practitioners increasingly treat as assurance evidence. Catalogued in the paper’s appendices, these resources range from model documentation templates to adversarial robustness benchmarks. The authors’ assessment is cautious: such artifacts are useful building blocks, but benchmarks alone do not constitute assurance. A model that scores well on a static test set may still fail catastrophically under distribution shift, and a completed documentation template may describe a system that has since been retrained. Evidence must be tied to specific, testable claims and kept current, or it becomes assurance theater, paperwork that simulates accountability without delivering it.

What emerges from the analysis is a call for assurance strategies that are coherent, context-sensitive, and adaptive. Coherent means that safety, fairness, privacy, and the other goals are pursued within a single integrated structure rather than as competing compliance exercises. Context-sensitive means that the appropriate depth and rigor of assurance scales with the risk profile of the deployment: a recommendation engine and an autonomous vehicle should not face identical burdens. Adaptive means that assurance is treated as a continuous lifecycle activity, with monitoring, re-evaluation, and evidence refreshment built into operations rather than bolted on at certification time. This vision aligns with, but pushes beyond, the direction of travel in emerging regulation, including the European Union’s risk-based approach to AI systems.

The practical implications reach every stakeholder in the AI ecosystem. For policymakers, the message is that principles without operationalisation create a compliance culture that rewards paperwork over performance; standards bodies must invest in specifying measurable requirements and acceptable evidence formats. For developers, the five commitments offer a design discipline: build systems so that claims about them can be tested, evidence can be generated automatically, and lifecycle changes are tracked. For regulators and auditors, the review highlights the need for interoperable assurance artifacts so that evidence produced in one jurisdiction or sector can be evaluated elsewhere, reducing duplication and closing loopholes. The authors also point toward the insurance industry as an underappreciated actor, noting that research programs on assurance and insurance for AI suggest financial markets may soon demand the kind of verifiable evidence that assurance frameworks provide.

The Warwick study arrives at a decisive moment. Governments across the world are standing up AI safety institutes, regulators are drafting enforcement rules, and enterprises are racing to demonstrate trustworthiness to customers and courts. Yet as Khadka and Maple demonstrate, the connective tissue between ambition and verification is still missing. Their contribution is less a new framework than a rigorous audit of the frameworks we already have, measured against a simple standard: can a claim about an AI system be tested, evidenced, maintained across its lifecycle, contested through recourse, and trusted across boundaries? By that measure, most current initiatives fall short. The path forward they chart, treating assurance as a living, integrated engineering practice rather than a static compliance checkbox, offers the clearest route yet to AI systems that earn, rather than merely assert, the public’s trust.

Subject of Research: AI assurance frameworks, their gaps, and the operationalisation of trustworthy AI governance

Article Title: Artificial Intelligence (AI) assurance: challenges, gaps, and the path forward

Article References: Khadka, A., & Maple, C. (2026). Artificial Intelligence (AI) assurance: challenges, gaps, and the path forward. Artificial Intelligence Review. https://doi.org/10.1007/s10462-026-11719-y

Image Credits: AI Generated

DOI: 10.1007/s10462-026-11719-y

Keywords: AI assurance, trustworthy AI, NIST AI RMF, machine learning, AI regulation, assurance cases, AI safety, fairness, explainability, accountability, risk management, AI governance