Applications, Major Risks and Feasible Preventive Measures of AI Agents in the Financial Industry

… models like Graph Neural Networks (GNNs) can learn patterns that signify systemic risks[11] … of uncertainty in financial risk estimations[12].In credit risk management, Agentic AIpowered …

… regarding the ethical implications of deploying agentic AI in financial services. Due to increasing … Moreover, the potential for systemic financial risk warrants careful consideration. Co-…

… Proposed Agentic Gen AI Architecture This section outlines the proposed agentic Generative AI framework for enhancing financial risk modeling. The framework integrates traditional …

… by traditional model risk management. This paper develops a financial-sector framework for agentic AI governance built on four elements: a practical risk taxonomy, behavioral stress …

… risks agentic systems pose to stability and volatility in the market. Our discussion posits that while agentic AI … understanding of agentic AI and its multi faceted impact on financial markets. …

The evolution of agentic AI has fundamentally redefined the Banking, Financial Services, and Insurance (BFSI) sector’s approach to risk management. For decades, financial institutions have relied on deterministic risk assessment controllers governed by fixed models, static thresholds, and linear workflows. While effective in stable conditions, these legacy systems struggle to handle the dynamic, interconnected, and data-intensive nature of today’s financial ecosystems [1]. Modern BFSI operations generate massive streams of multimodal data, including structured financial metrics, unstructured text, behavioural signals, and real time market data that exceed the processing capacity of traditional risk engines. As a result, many risk controllers remain reactive, discovering threats only after exposure or regulatory breach. The emergence of Agentic AI systems addresses these limitations by introducing autonomy, adaptivity, and explainability into the risk control process. Unlike static models, these systems employ specialized AI agents that collaborate across domains—credit, liquidity, compliance, cybersecurity, and actuarial—using shared context and feedback loops. Each agent continuously perceives, reasons, and acts within its environment to maintain optimal control states. At the core of this evolution lies Reinforcement Learning (RL) and multi-agent orchestration, enabling continuous decision optimization under uncertainty [2]. RL agents learn from environmental feedback, dynamically adjusting thresholds and capital allocations in response to market, operational, or regulatory changes. This paper presents a technical framework detailing how agent based architectures, reinforced by machine reasoning and control theory, can autonomously mitigate risk across BFSI domains. It explores how these systems improve early warning capabilities, enhance model governance, and ensure regulatory compliance all while maintaining explainability and auditability in high stakes environments. In doing so, Agentic AI establishes the foundation for self adaptive risk ecosystems, capable of operating with human oversight yet independent in execution transforming risk management from a reactive function into a predictive and preventive intelligence layer for the modern financial enterprise.

The evolution of Large Language Models, Roboadvisors or chatbots, and Machine Learning Models have opened the possibilities for automation across sectors to innovate and introduce new capabilities as part of their business process. This review paper targets all the critical aspects of artificial intelligence and machine learning models and their impact on the successful deployment of the Agentic AI capability. Agents can not only participate but also assist, self-organize, and negotiate using parallel workstreams and cognitive architectures. They solve problems iteratively based on their cognitive level and domain, creating value in the financial services market, where opportunities are vast. This study further includes principles for Agentic AI, its lifecycle, various key components, impact on the financial industry, architectural considerations, data considerations, benchmarking, opportunities, future capabilities, challenges, and more.

Financial technology (FinTech) is significantly reshaped in the rapid evolution of AI, moving into agentic systems that combine autonomy, reasoning, memory, and compliance guardrails, instead of scripted Chabot. concurrently, the appearance of Emotionally Intelligent Management Systems (EIMS) highlights the importance of integrating emotional awareness into digital enterprise solutions. This paper presents a multiagent FinTech system that employs large language model (LLM) in digital employees’ performance. Five roles were followed to extend beyond automation to deliver adaptive, explainable, and emotionally attuned services: Fraud Detection Assistant with Emotional Sensitivity, a Compliance Companion with Trust-Aware Guardrails, a Customer Support Agent with Empathetic Dialogue, a Personal Finance Mentor with Well-being Focus, and an Investment Analyst Partner with Sentiment Awareness. To evaluate the idea, real and simulated data are combined, the UCI Credit Card Default dataset provides a benchmark for credit risk modelling, while the CFPB Consumer Complaints dataset grounds analysis in regulatory and customer service contexts. Synthesized adversarial obfuscations, simulated suspicious transaction flows is implemented to strengthen the work evaluation. Experiments prove the proposed agentic system reliably, where predictive accuracy using F1-score and AUROC improves compared to traditional baselines, and time to decide reduced by over 30% in customer support and compliance workflows. Moreover, the integration of emotional intelligence enables adaptive responses to customer prevention, transparent compliance explanations, and finance mentoring that accounts for stress signals. Employing FinTech automation with emotionally intelligent management in the proposed work contributes to technical efficiency, regulatory transparency, and human-centric digital transformation, leading to predictive deployments in enterprise, and investment will be strictly illustrative.

… , tool use over heterogeneous financial data, long-horizon memory … agentic systems is essential for next-generation financial … review major financial application domains (eg, trading, risk …

… AI and Agentic AI is reshaping finance, offering unprecedented efficiency in decision-making, risk … in financial markets—from high-frequency trading to credit risk assessment—while …

Financial services are rapidly advancing towards highly autonomous, intelligent, and personalized solutions by integrating agentic AI (Artificial Intelligence) systems. This paper presents a comprehensive architecture and implementation of autonomous agentic AI frameworks, specifically designed for financial services, and built upon a series of Amazon Web Services (AWS) cloud technologies. We propose a scalable and secure architecture for developing intelligent financial assistants that can manage and performing a wide range of multi-step financial tasks, such as personalized financial planning, portfolio rebalancing, and account management, and we review the entire end- to-end workflow to build and deploy such autonomous systems. In particular, this work focuses on how large language models (LLMs) can be orchestrated with backend systems, services such as AWS Lambda, Amazon Bedrock, Agent Core Runtime for orchestration, and Amazon DynamoDB for state management, to enable autonomous financial services. We also address critical concerns related to security, ethical standards, and auditability, which are essential for responsible adoption of these systems in financial institutions. This research aims to bridge technological innovation with customer- centric and regulatory priorities in the finance industry. By doing so, this paper showcases how agentic AI can power next generation financial service delivery to transform customer experience and drive institutional efficiency.

In February 2025, OpenAI announced ‘Operator’, an agent that includes a computerusing agent (CUA) model, which means that it does not need an application programming interface (API) to access services but can use buttons and navigate menus just as people do. Currently, an Operator requires human supervision to complete certain tasks, so a consumer needs to take control, for example, to enter payment information. However, with the advent of CUAs, the practical evolution of full-blown agentic commerce in strategic timeframes is well underway. Agents will go online to obtain services, look for an agent or an agentic API and then, if no such access methods are found, simply access the web pages as a human customer does. Agents will need more than API or web access to execute financial transactions on behalf of individuals or organisations; however, they will need authorisation. This means that agents need a fundamental property to deliver for their users: identity. This paper demonstrates that extending digital identity to agents is a fundamental enabler for bots to access financial services on behalf of individuals or organisations (or, indeed, themselves) to create a new financial world that can admit nonhuman customers to generate better outcomes for consumers and businesses. To deliver this vision of financial health for all, the paper posits that companies must first have a digital identity infrastructure that can provide identification, authentication and authorisation for not only financial institutions and their customers, but also for the customers’ agents. Know-your-customer (KYC) is necessary, as is know-your-business (KYB) and knowyour- employee (KYE). But without know-your-agent (KYA), which is much more complicated than KYC, KYB and KYE, there will be no progress. This article is also included in The Business & Management Collection, which can be accessed at https://hstalks.com/business/.

… To systematically understand the emergent diversity within the autonomous AI agent ecosystem, this section introduces a typology that categorizes autonomous agents according to …

Agentic AI strengthens financial technology by letting autonomous financial advisories use flexible systems to block and control their procedures. The new financial products deliver specific customer solutions in real-time, which make users' financial decisions better informed. Our study examines today's beneficial uses, development methods, and essential results of agentic AI technology in financial advising. Research studies look at current systems plus conduct real-world studies to show how agentic AI fares compared to usual technology. The research examines all the main ethical, security, legal, and system challenges the industry faces today. Our study demonstrates that agentic AI offers efficient ways to deliver financial advice while making services scalable and ethical. The findings demonstrate how agentic AI systems will help make money basics easier to understand and reach a wider set of people within the financial system. This essential research output builds the base for more development of agentic AI systems and recommends useful future study directions.

… In the context of this study, ‘autonomous AI agents’ are defined as advanced software … financial services, and participating in marketplaces, with minimal to no human intervention. These …

Recent advances in large language models, tool-using agents, and financial machine learning are shifting financial automation from isolated prediction tasks to integrated decision systems that can perceive information, reason over objectives, and generate or execute actions. The paper develops an integrative framework for analysing agentic finance: financial market environments in which autonomous or semi-autonomous AI systems participate in information processing, decision support, monitoring, and execution workflows. The analysis proceeds in three steps. First, the paper proposes a four-layer architecture of financial AI agents covering data perception, reasoning engines, strategy generation, and execution with control. Second, it introduces the Agentic Financial Market Model (AFMM), a stylised agent-based representation linking agent design parameters such as autonomy depth, heterogeneity, execution coupling, infrastructure concentration, and supervisory observability to market-level outcomes including efficiency, liquidity resilience, volatility, and systemic risk. Third, it presents an illustrative empirical application based on event studies of AI-agent capability disclosures and heterogeneous market repricing. It argues that the systemic implications of AI in finance depend less on model intelligence alone than on how agent architectures are distributed, coupled, and governed across institutions. The empirical application is intentionally exploratory: it does not validate the full AFMM but shows how one observable expectations channel can be studied using public data. In the near term, the most plausible equilibrium is bounded autonomy, in which AI agents operate as supervised co-pilots, monitoring systems, and constrained execution modules embedded within human decision processes.

The emergence of Agentic Artificial Intelligence (Agentic AI)—autonomous systems capable of independent reasoning, planning, and executing actions across enterprise environments—presents a defining strategic paradox for Managing Directors and C-Suite executives in financial services. These leaders face an unprecedented challenge: aggressively deploying autonomous AI systems to drive operational efficiency and competitive advantage while maintaining unyielding regulatory compliance, operational stability, and risk management. This paper examines this paradox through a dual lens of practitioner job descriptions and scholarly literature. Practitioner sources reveal that executives are expected to simultaneously architect cloud-native AI platforms, operationalize Model Risk Management (MRM) frameworks for autonomous agents, define portfolio prioritization rubrics for agentic systems, establish human-in-the-loop and human-on-the-loop control mechanisms, and deliver quantifiable ROI—often without established playbooks or precedents. These roles demand deep fluency across a complex technical stack, including multi-agent orchestration frameworks, tool-use architectures, MLOps for agent monitoring, and governance systems for autonomous decision-making. Scholarly literature offers foundational insights through transdisciplinary research models, temporal perspectives on the academic-practitioner gap, and critical pragmatism as a bridging philosophy. However, a critical gap exists at the intersection of agentic AI implementation and executive strategic leadership: there is no empirical understanding of how senior leaders actually navigate the organizational, regulatory, and technical complexities of scaling autonomous AI systems in highly regulated environments. Adopting a Scholar-Practitioner approach, this proposed research will investigate how Managing Directors, CIOs, CTOs, and CDOs in banking and insurance navigate these challenges. The study will employ a qualitative multiple-case study design, integrating adaptive leadership theory with AI governance constructs to explore how executives balance innovation imperatives with control requirements in the age of autonomous AI. Findings will contribute actionable frameworks for executive decision-making, organizational design, and risk governance, bridging the gap between academic theory and practitioner need.

Financial infrastructure management is facing unprecedented challenges as distributed architectures with hundreds of interdependent microservices impose complexity beyond the human cognitive limit for real-time monitoring and intervention. Conventional reactive monitoring systems bring intolerable latency between anomaly discovery and human perception, and the manual diagnostic processes take considerable time during which the services are degraded. Security alert proliferation overloads operations teams, with false positives above 90% in expert domains like anti-money laundering surveillance. Agentic AI frameworks overcome these inherent limitations with self-contained systems that combine perception, reasoning, and action capabilities within operational infrastructure cores. Multi-agent architectures provide specialist domain expertise in areas of network performance, database optimization, security threat response, and capacity management while retaining collaborative problem-solving abilities for sophisticated failure scenarios. Self-restoration mechanisms utilize predictive analysis to determine precursors to failure minutes to hours in advance of full service loss, allowing for preventive actions that do not impact customers at all. Automated threat identification and response condense incident containment windows from hours to seconds, significantly shrinking vulnerability windows that advanced attackers target. Immutable audit trails using blockchain technologies meet regulatory demands for operational visibility while smart contract execution ensures policy compliance. Explainability issues call for the creation of understandable decision models that can explain reasoning logic in a human-readable form. Trust calibration needs graduated autonomy models that move from advisory recommendations to supervised execution toward complete autonomy for routine situations. Directions for the future include federated learning that facilitates cross-institutional sharing of knowledge, sophisticated causal modeling to predict intervention cascades, and digital twin incorporation, offering safe test beds.

… workflow orchestration, and autonomous financial services delivery. This shift represents a … Using knowledge-based AI agents, they demonstrated how generative AI can identify …

Agentic AI, an emerging paradigm in artificial intelligence, refers to autonomous systems designed to pursue complex goals with minimal human intervention. Unlike traditional AI, which depends on structured instructions and close oversight, Agentic AI demonstrates adaptability, advanced decision-making capabilities and self-sufficiency, enabling it to operate dynamically in evolving environments. This survey thoroughly explores the foundational concepts, unique characteristics, and core methodologies driving the development of Agentic AI. We examine its current and potential applications across various fields, including healthcare, finance, and adaptive software systems, emphasizing the advantages of deploying agentic systems in real-world scenarios. The paper also addresses the ethical challenges posed by Agentic AI, proposing solutions for goal alignment, resource constraints, and environmental adaptability. We outline a framework for safely and effectively integrating Agentic AI into society, highlighting the need for further research on ethical considerations to ensure beneficial societal impacts. This survey serves as a comprehensive introduction to Agentic AI, guiding researchers, developers, and policymakers in engaging with its transformative potential responsibly and creatively.

… autonomous AI agents and their ability to enhance financial security through intelligent insider … the findings of the application of AI Driven Identity Governance Solutions and show how AI …

: Artificial intelligence (AI) is reshaping financial systems and services, as intelligent AI agents increasingly form the foundation of autonomous, goal-driven systems capable of reasoning, learning, and action. This review synthesizes recent research and developments in the application of AI agents across core financial domains. Specifically, it covers the deployment of agent-based AI in algorithmic trading, fraud detection, credit risk assessment, robo-advisory, and regulatory compliance (RegTech). The review focuses on advanced agent-based methodologies, including reinforcement learning, multi-agent systems, and autonomous decision-making frameworks, particularly those leveraging large language models (LLMs), contrasting these with traditional AI or purely statistical models. Our primary goals are to consolidate current knowledge, identify significant trends and architectural approaches, review the practical efficiency and impact of current applications, and delineate key challenges and promising future research directions. The increasing sophistication of AI agents offers unprecedented opportunities for innovation in finance, yet presents complex technical, ethical, and regulatory challenges that demand careful consideration and proactive strategies. This review aims to provide a comprehensive understanding of this rapidly evolving landscape, highlighting the role of agent-based AI in the ongoing transformation of the financial industry, and is intended to serve financial institutions, regulators, investors, analysts, researchers, and other key stakeholders in the financial ecosystem.

As cross-border financial transactions grow in scale and complexity, so too does the risk of fraud, regulatory noncompliance, and systemic vulnerabilities in global FinTech ecosystems.The heterogeneous regulatory environments, varying KYC/AML standards, and speed of digital finance innovation challenge traditional surveillance and compliance mechanisms.Conventional rule-based fraud detection systems often fall short in adapting to rapidly evolving threat patterns, particularly in high-volume, real-time cross-border contexts.This study introduces a novel framework based on multi-agent artificial intelligence (AI) systems designed to enhance fraud surveillance, increase transparency, and ensure regulatory compliance in global FinTech operations.The proposed architecture comprises autonomous, cooperative AI agents-each specialized in tasks such as behavioral profiling, transaction risk scoring, anomaly detection, and jurisdiction-specific regulation enforcement.These agents operate across decentralized data environments while maintaining privacy and interoperability through secure federated learning protocols.The paper explores how the multi-agent framework dynamically integrates data from diverse sources including digital wallets, blockchain ledgers, and SWIFT/ISO 20022 messaging formats.Agents leverage machine learning models for adaptive fraud pattern recognition and use explainable AI (XAI) to ensure decision traceability.Regulatory compliance agents monitor evolving legal requirements, generating automated audit trails to facilitate international supervisory reporting and minimize compliance latency.Case studies involving real-time remittance flows and digital asset transfers are used to evaluate the system's efficacy in mitigating fraud and false positives.The results demonstrate improved detection accuracy, faster resolution times, and enhanced trust between institutions and regulators.By deploying multi-agent AI, FinTech platforms can achieve secure, transparent, and compliant surveillance in the complex terrain of global financial exchange.

Artificial Intelligence (AI) multi-agent frameworks are enabling autonomous decision-making, intelligent collaboration, and the automation of complex workflows. These frameworks leverage Large Language Models (LLMs) and distributed AI systems to optimize operations across diverse sectors, with finance emerging as one of the most impacted domains. AI agents are increasingly employed in risk assessment, regulatory compliance, algorithmic trading, fraud detection, and customer service, fundamentally altering how financial institutions operate and manage market dynamics. This paper presents a review of AI multi-agent frameworks, evaluating their architectures, applications, and deployment challenges within financial services. We conduct an in-depth comparative analysis of prominent frameworks, including LangChain, CrewAI, and OpenAI Swarm, assessing their strengths, limitations, and suitability for different financial applications. Furthermore, we examine how these frameworks integrate into financial ecosystems, facilitating automated decision-making, enhancing operational efficiency, and mitigating systemic risks. Despite the transformative potential of AI agents, their widespread adoption introduces critical challenges, such as data quality inconsistencies, lack of model explainability, regulatory concerns, and ethical dilemmas. This paper explores these issues, emphasizing the necessity for transparency, accountability, and robustness in AI-driven financial solutions. Additionally, we highlight the role of AI governance and risk mitigation strategies in ensuring regulatory compliance and alignment with financial industry standards. We also outline future research directions, advocating for the development of interpretable, scalable, and resilient AI agent frameworks. As financial automation continues to evolve, a deeper understanding of multi-agent AI systems is essential for leveraging their full potential while mitigating associated risks.

Artificial Intelligence (AI) agents are revolutionizing industries by enabling autonomous decision-making, task execution, and multi-agent collaboration. This paper provides a comprehensive review of AI agent frameworks, focusing on their architectures, applications, and challenges in financial services. We conduct a comparative analysis of leading frameworks, including LangGraph, CrewAI, and AutoGen, evaluating their strengths, limitations, and suitability for complex financial tasks such as trading, risk assessment, and investment analysis. The integration of AI agents in financial markets presents both opportunities and challenges, particularly in terms of regulatory compliance, ethical considerations, and model robustness. We examine agentic AI design patterns, multi-agent systems, and the deployment of AI agents advancing the proposal to use them for fraud detection and risk management. By synthesizing insights from academic research and industry practices, this review identifies key trends and future directions in AI agent development. This work contributes to the growing discourse on AI-driven automation by outlining technical considerations and open challenges in deploying AI agents at scale. We highlight the need for enhanced transparency, interpretability, and security in AI-driven Agentic systems. Our findings provide valuable insights for researchers and practitioners seeking to harness AI agents for more efficient and intelligent decision-making.

… for enabling secure, collaborative artificial intelligence in financial risk management. Through … [26] “Risk AI in action: How risk teams are building AI agents for real-world impact,” https://…

This paper surveys the landscape of AI agent frameworks, highlights their core features and differences, and explores their applications in financial services. We synthesize insights from recent industry reports, academic research, and technical blog posts, focusing on frameworks such as CrewAI, LangGraph, LlamaIndex, and others. We also discuss the challenges and opportunities of deploying agentic AI in production environments, with an emphasis on financial trading, investment analysis, and decision support. We analyze the rapidly evolving landscape of agentic AI systems, focusing on their architecture, capabilities, and practical implementations in banking, trading, and risk management. The study examines prominent frameworks including LangGraph for stateful agent orchestration, CrewAI for collaborative multi-agent workflows, and AutoGen for conversational agent systems, alongside industry platforms like IBM watsonx and NVIDIA NIM. The study examines both technical frameworks (LangGraph, CrewAI, AutoGen, etc.) and practical implementations in financial institutions. We highlight productivity gains (up to 80% time reduction in data tasks), risk management improvements, and workforce transformation challenges. The paper concludes with recommendations for financial institutions adopting agentic AI solutions. Our analysis reveals three key findings: (1) specialized agent frameworks achieve 50-80% productivity gains in financial data tasks compared to traditional approaches, (2) multi-agent systems demonstrate particular promise in complex domains like algorithmic trading and fraud detection, and (3) successful deployment requires addressing critical challenges in workforce upskilling, risk alignment, and regulatory compliance. The paper provides a theoretical foundation for agentic AI in finance, introducing formal models for agent design patterns, multimodal fusion, and market microfoundations. We further present a summary of several evaluation frameworks for assessing agent performance across financial use cases, including portfolio optimization and AML compliance. The study concludes with recommendations for financial institutions adopting agentic AI, emphasizing the need for standardized architectures, robust testing protocols, and hybrid human-AI workflows.

In recent years, the interplay between AI agents and blockchain has enabled secure and scalable collaboration among multi-agent systems, promoting unprecedented levels of autonomy and interoperability. AI agents play a vital role in facilitating complex decision making and improving operational efficiency in blockchain systems. This collaborative synergy is particularly evident in how multi-agent systems collectively tackle complex tasks to ensure seamless integration within these frameworks. While significant efforts have been made to integrate AI agents and blockchain, most studies overlook the broader potential of AI agents in addressing challenges such as interoperability, scalability, and privacy issues. In this paper, we bridge these gaps by illustrating the interplay between AI agents and blockchain. Specifically, we explore how AI agents enhance decentralized systems and examine blockchain’s role in enabling secure and scalable collaboration. Furthermore, we categorize practical applications across domains, such as Web3, decentralized finance (DeFi), asset management, and autonomous systems, providing practical insights and real-world use cases. Additionally, we identify key research challenges, including the complexities of multi-agent coordination, interoperability across diverse systems, and privacy maintenance in decentralized frameworks. Finally, we offer future directions in terms of governance, sovereignty, computation, and interpretability to promote a secure and responsible ecosystem.

The emergence of autonomous AI agents—systems proficient in reasoning, planning, and executing intricate tasks with digital tools—signifies a pivotal transformation in automation. In contrast to conventional generative AI, these agents function with considerable autonomy, offering potential productivity enhancements in fields such as healthcare, finance, and education. Nevertheless, their autonomy presents new governance difficulties, encompassing liability, monitoring, and systemic hazards. This study delineates the capabilities of AI agents, assesses their societal ramifications, and proposes a dynamic, evidence-informed governance system. Policymakers should leverage the advantages of AI agents and mitigate dangers through regulatory sandboxes, transparency standards, and international coordination. Our plan prioritizes collaboration among governments, industry, and civil society to guarantee safe, equitable, and innovative deployment of agents.

… provides the governance logic for agentic systems specifically. … challenges posed by agentic systems—AI systems exhibiting all … While agentic systems present unique accountability and …

Agentic artificial intelligence (AI) — systems capable of autonomous goal pursuit, adaptive learning, multi-agent coordination, and emergent behaviour — is being deployed across financial markets faster than the ethical and regulatory frameworks designed to govern it. The five technical features that make agentic systems valuable to financial institutions are precisely the features that defeat the accountability, transparency, human-oversight, and systemic-safety norms on which existing governance is built. This paper presents the first systematic scoping review of agentic-AI governance frameworks in financial markets, analysing 30 documents published between 2022 and 2025 across seven jurisdictions through a structured 67-provision analytical instrument and Bayesian Beta–Binomial evidence synthesis. Sixty-one per cent of governance provisions are absent from the majority of the global corpus, characterising the field as pre-paradigmatic. Transparency is the empirically weakest dimension (estimated coverage 35%, 95% credible interval 30 to 40 per cent), generating what we term the transparency paradox: the ethical mechanism that makes every other governance norm legible, auditable, and enforceable is the mechanism most comprehensively missing. We then propose a three-axis normative framework — autonomy level by market function by regulatory regime — that maps the empirical gaps onto four ethical anchors and yields actionable design implications. We illustrate the framework through a case-study application to the Qatar Financial Centre; the framework, the empirical findings, and the methodological approach are intended to generalise.

… paper proposes FASTRAC (Financial Agent Safety, Trust, Risk, and Compliance), a control-… This paper has argued that effective governance of agentic financial systems requires a …

… tasks like research reports, financial markets require systems that adapt to continuously … financial institutions adopting Large Language Models (LLMs) and autonomous agentic systems…

… , an agentic system can face a free-zone autonomoussystems rule, a federal data-protection baseline, and a sectoral health or finance authority concurrently, each projecting the system …

… Integrating blockchain with multi-LLM agents and artificial … contracts and multi-LLM agents in fraud detection, which is a … In this study, the optimal multi-LLM agent fraud detection …

Effective financial fraud detection requires systems that can interpret complex transaction semantics while dynamically adapting to asymmetric operational costs. We propose a hybrid framework in which a large language model (LLM) serves as an encoder, transforming heterogeneous transaction data into a unified embedding space. These embeddings define the state representation for a reinforcement learning (RL) agent, which acts as a fraud classifier optimized with business-aligned rewards that heavily penalize false negatives while controlling false positives. We evaluate the approach on two benchmark datasets—European Credit Card Fraud and PaySim—demonstrating that policy-gradient methods, particularly A2C, achieve high recall without sacrificing precision. Critically, our ablation study reveals that this hybrid architecture yields substantial performance gains on semantically rich transaction logs, whereas the advantage diminishes on mathematically compressed, anonymized features. Our results highlight the potential of coupling LLM-driven representations with RL policies for cost-sensitive and adaptive fraud detection.

… Our main contributions can be summarized as follows: (1) We propose CogAgent, a novel self-evolving cognitive Agent framework that combines LLM-driven reasoning with continuous …

Voice phishing (vishing) attacks have become increasingly sophisticated, driven by large-scale call centers, social engineering scripts, and AI-generated synthetic voices, posing serious risks to individuals and organizations. Existing vishing detection approaches predominantly rely on monolithic machine learning or deep learning models that offer limited transparency and struggle to capture the multi-dimensional nature of fraudulent voice interactions. This paper presents a novel, web-based vishing detection platform (VishGuard) powered by a large language model (LLM)-based multi-agent system that provides accurate, explainable, and real-time detection across voice calls, emails, and text messages. The proposed architecture decomposes the detection task into six specialized agents, each focusing on a distinct dimension of fraudulent behavior, including product and service deception, linguistic threat intelligence, social engineering tactics, financial fraud indicators, and compliance and legal violations. A central evaluator aggregates evidence generated by these agents to produce a calibrated scam likelihood score and an interpretable rationale. The analytical agents are implemented using the OpenAI GPT-4o Mini model, while voice inputs are processed through automatic speech recognition (ASR) using the Google Gemini 2.5 Flash model. To evaluate the system, a balanced dataset of 3,000 records was constructed, consisting of safe and malicious voice messages and call transcripts. The dataset was primarily curated by domain experts, augmented with synthetically generated samples using GPT-5 and Claude Sonnet 4.5, and further expanded through paraphrasing, Synonym replacement, and audio noise injection. Experimental results demonstrate strong performance, achieving 97.25% accuracy, 99.48% precision, 95% recall, and 97.19% F1-score, while maintaining a low false-positive rate of 0.5%. These results indicate that the proposed multi-agent approach effectively balances detection sensitivity with usability and trust. Overall, this work advances vishing defense by introducing an explainable, modular, and deployment-ready AI system that addresses key limitations of existing ML-, DL-, and LLM-based detection methods.

… -agent framework designed to address fraud detection in a scalable, interpretable, and adaptable way. Each agent … input analysis and phishing entity detection to source validation and …

This paper presents a novel approach to e-commerce payment fraud detection by integrating reinforcement learning (RL) with Large Language Models (LLMs). By framing transaction risk as a multi-step Markov Decision Process (MDP), RL optimizes risk detection across multiple payment stages. Crafting effective reward functions, essential for RL model success, typically requires significant human expertise due to the complexity and variability in design. LLMs, with their advanced reasoning and coding capabilities, are well-suited to refine these functions, offering improvements over traditional methods. Our approach leverages LLMs to iteratively enhance reward functions, achieving better fraud detection accuracy and demonstrating zero-shot capability. Experiments with real-world data confirm the effectiveness, robustness, and resilience of our LLM-enhanced RL framework through long-term evaluations, underscoring the potential of LLMs in advancing industrial RL applications.

… Accordingly, a multi-agent deep reinforcement learning … , each of which is an expert trader on a specific timeframe. The … the agents trading at higher timeframes to the agents trading at …

… Action Traders in deep multi-agent reinforcement learning to study the emergence of trading behavior, significantly increasing social efficiency compared to agents without action trading…

Algorithmic trading is now the most common form of trading in financial markets, and it has been estimated that it accounts for 60-75% of the total trading volume in major markets. However, algorithmic trading is still accompanied by cognitive and algorithmic biases such as overconfidence, confirmation bias, and anchoring effects that can result in suboptimal decisions and higher levels of risk. These biases are due to the excess reliance on certain kinds of data, historical overfitting, and the absence of mechanisms to adapt to changing market environments. We propose in this paper, the use of multi-agent AI systems (MAIS) to tackle these biases through collaboration, role differentiation, and learning. In this manner, MAIS design various agents that perform specific tasks, for instance, fundamental analysts, sentiment analysts, and technical analysts to ensure that the analysis is holistic yet without concentrating on a single kind of data. Thus, debate protocols and risk management teams ensure that the generation and evaluation of trading ideas are properly structured and that overconfidence and groupthink are avoided. Furthermore, there are market observer agents and reflective agents that provide online learning of model drift and offline learning of historical performance, respectively. Our architecture framework was tested in a simulated environment in which MAIS traded against human traders and rule-based algorithms using historical market data. The results showed that there were great quantitative improvements in the Sharpe ratios and drawdowns, which show that the system is good at improving riskadjusted returns and decreasing volatility. The last section of the paper contains a conclusion and the suggestions for future research.

… To address this problem, a novel approach called Multi-Agent … risk under the multi-agent reinforcement learning framework … Besides, the multi-agent model demonstrates its advantage …

The paper presents issues related to developing methods for fundamental analysis used to expand capabilities of multi-agent trading system, to better predict the financial market. The fundamental analysis indicators can be used as confirmation of decisions generated by other strategies of the system. The first part of the article discusses briefly the fundamental analysis issues in relation to the online trading on FOREX market. The statistical analysis of correlations of the different time series indicators and algorithms of fundamental analysis agents are examined. The final part discusses the results of the performance evaluation of selected investment strategies, including fundamental-based agents.

… The Common Agent Request Broker Architecture or CARBA has been suggested as a dynamic architecture for multi-agent systems that is similar to CORBA [II]. It is based on the …

… For this reason, our trading agents will be using data mining models for price prediction, but … of trades and the capital requirements, we will integrate these agents in a multi-agent system…

… algorithmic trading—especially in fast-paced, high-frequency … , from solo agents to multi-agent systems, applying DRL methods … in both single-agent and multi-agent setups, achieving a …

As automated trading gains traction in the financial market, algorithmic investment strategies are increasingly prominent. While Large Language Models (LLMs) and Agent-based models exhibit promising potential in real-time market analysis and trading decisions, they still experience a significant -20% loss when confronted with rapid declines or frequent fluctuations, impeding their practical application. Hence, there is an imperative to explore a more robust and resilient framework. This paper introduces an innovative multi-agent system, HedgeAgents, aimed at bolstering system robustness via ''hedging'' strategies. In this well-balanced system, an array of hedging agents has been tailored, where HedgeAgents consist of a central fund manager and multiple hedging experts specializing in various financial asset classes. These agents leverage LLMs' cognitive capabilities to make decisions and coordinate through three types of conferences. Benefiting from the powerful understanding of LLMs, our HedgeAgents attained a 70% annualized return and a 400% total return over a period of 3 years. Moreover, we have observed with delight that HedgeAgents can even formulate investment experience comparable to those of human experts (https://hedgeagents.github.io/).

… , highlighting the potential of multi-agent cooperation for robust high-frequency trading. … of more robust trading strategies. Furthermore, single-agent and multi-agent systems will be …

As financial markets are complex and often unpredictable, trading systems that rely on a single approach do not hold up well when significant unexpected events occur. Common algorithmic trading strategies use a single statistical model or a very fixed set of rules and examine only a limited number of technical signals, failing to adjust to changing conditions or provide clear, risk-aware guidance at decision time. This paper introduces a Hybrid Multi-Agent Decision Support System (HMADSS) designed for momentum-based trading in the Indian stock market. The proposed framework features a modular, multi-level architecture that integrates rule-based market filtering, advanced predictive analytics, confidence-weighted decision fusion, and reinforcement-learning-based trade management. Processing all 2,129 NSE stocks and more than 366,000 historical OHLCV records with 49 engineered technical features, the system achieves 88% precision in identifying genuine upward momentum breakouts using an ensemble of Extreme Gradient Boosting (XGBoost) and Long Short-Term Memory (LSTM) networks (weights: XGBoost 0.7, LSTM 0.3). To address the black-box nature of AI in finance, four autonomous agents (Market Trend, Technical Structure, Predictive Analytics, and Portfolio Risk) collaboratively compute an explainable Composite Confidence Score (CCS). A separate Reinforcement Learning agent dynamically manages the post-entry trade lifecycle within strict risk limits. Validated through 34 live-market executions, HMADSS successfully mitigates asymmetric risk, limiting maximum single-trade drawdowns to $-9.11 \%$ while capturing momentum swings up to $\mathbf{+ 1 6. 0 9 \%}$, delivering a live Profit Factor of 1.26. These results demonstrate that combining probabilistic deep learning with structured risk agents effectively bridges the gap between academic algorithmic research and practical swing-trading deployment.

Trading decisions often encounter risk and uncertainty complexities, significantly influencing their overall performance. Recognizing the intricacies of this challenge, computational models within information systems have become essential to support and augment trading decisions. The paper introduces the concepts of trading software agents, investment strategies, and evaluation functions that automate the selection of the most suitable strategy in near real-time, offering the potential to enhance trading effectiveness. This approach holds the promise of significantly increasing the effectiveness of investments. The research also seeks to discern how changing market conditions influence the performance of these strategies, emphasizing that no single agent or strategy universally outperforms the rest. In summary, the overarching objective of this research is to contribute to the realm of financial decision-making by introducing a pragmatic platform and strategies tailored for traders, investors, and market participants in the FOREX market. Ultimately, this endeavor aims to empower people with more informed and productive trading decisions. The contributions of this work extend beyond the theoretical realm, demonstrating a commitment to address the practical challenges faced by traders and investors in real-time decision-making within the financial markets. This multidimensional approach to financial decision support promises to enhance investment effectiveness and contribute to the broader field of algorithmic trading.

We introduce ABIDES, an open source Agent-Based Interactive Discrete Event Simulation environment. ABIDES is designed from the ground up to support agent-based research in market applications. While proprietary simulations are available within trading firms, there are no broadly available high-fidelity market simulation environments. ABIDES enables the simulation of tens of thousands of trading agents interacting with an exchange agent to facilitate transactions. It supports configurable pairwise noisy network latency between each individual agent as well as the exchange. Our simulator's message-based design is modeled after NASDAQ's published equity trading protocols ITCH and OUCH. We introduce the design of the simulator and illustrate its use and configuration with sample code, validating the environment with example trading scenarios. The utility of ABIDES for financial research is illustrated through experiments to develop a market impact model. The core of ABIDES is a general-purpose discrete event simulation, and we demonstrate its breadth of application with a non-finance work-in-progress simulating secure multiparty federated learning. We close with discussion of additional experimental problems it can be, or is being, used to explore, such as the development of machine learning trading algorithms. We hope that the availability of such a platform will facilitate research in this important area.

Agent-based modelling (ABM) approaches for high-frequency financial markets are difficult to calibrate and validate, partly due to the large parameter space created by defining fixed agent policies. Multi-agent reinforcement learning (MARL) enables more realistic agent behaviour and reduces the number of free parameters, but the heavy computational cost has so far limited research efforts. To address this, we introduce JaxMARL-HFT (JAX-based Multi-Agent Reinforcement Learning for High-Frequency Trading), the first GPU-accelerated open-source multi-agent reinforcement learning environment for high-frequency trading (HFT) on market-by-order (MBO) data. Extending the JaxMARL framework and building on the JAX-LOB implementation, JaxMARL-HFT is designed to handle a heterogeneous set of agents, enabling diverse observation/action spaces and reward functions. It is designed flexibly, so it can also be used for single-agent RL, or extended to act as an ABM with fixed-policy agents. Leveraging JAX enables up to a 240x reduction in end-to-end training time, compared with state-of-the-art reference implementations on the same hardware. This significant speed-up makes it feasible to exploit the large, granular datasets available in high-frequency trading, and to perform the extensive hyperparameter sweeps required for robust and efficient MARL research in trading. We demonstrate the use of JaxMARL-HFT with independent Proximal Policy Optimization (IPPO) for a two-player environment, with an order execution and a market making agent, using one year of LOB data (400 million orders), and show that these agents learn to outperform standard benchmarks. The code for the JaxMARL-HFT framework is available on GitHub 1.

… AI significantly enhances the efficiency of loan processing, reduces bias, and improves the precision of risk … 7B Loan Recommendation AI Agent, which assesses the loan application in …

The rapid evolution of artificial intelligence (AI), particularly in the form of agentic systems, has introduced unprecedented capabilities for autonomous decision making in various sectors, including financial services.1 These agentic AI systems, characterised by their ability to act independently and learn from interactions, promise to revolutionise credit decisioning by enhancing efficiency, accuracy, and risk assessment. The autonomous nature of agentic AI, however, necessitates a robust framework for managing inherent risks, including issues of accountability, bias, and compliance, especially within the highly regulated financial industry. This paper proposes the development and implementation of agentic AI Risk-Aware Credit Decisioning Sandbox (RA-CDS) as a critical mechanism to foster responsible innovation while mitigating potential systemic risks. This controlled environment would allow for the rigorous evaluation of AI agents’ performance and adherence to ethical guidelines within credit decisioning processes.2 This article is also included in The Business & Management Collection which can be accessed at https://hstalks.com/business/.

… This paper has argued that effective governance of agentic financial systems requires a paradigmatic shift: from passive, reactive oversight to active, embedded, and adaptive control. …

The investment management industry stands at the precipice of a transformative shift driven by the convergence of Generative AI (GenAI) and Agentic AI systems. This paper introduces and comprehensively analyzes the “Agentic Investment Firm” model—a paradigm where small Registered Investment Advisors (RIAs) and boutique investment teams can leverage autonomous AI agents to manage substantial assets with institutional-grade capabilities. We present a holistic framework encompassing architectural design, governance, operational implementation, regulatory compliance, and economic viability specifically tailored for resource-constrained teams. Our contribution is threefold: First, we propose a scalable, layered system architecture with specialized AI agents for due diligence, macro intelligence, compliance automation, and real-time portfolio management. Second, we develop a pragmatic implementation roadmap with a phased 16-week deployment strategy that reduces operating costs by 50-70% while enhancing analytical depth and client personalization. Third, we provide a critical integration of regulatory frameworks—including detailed mappings of the NIST AI Risk Management Framework (AI RMF) to small-team contexts and comprehensive analysis of securities regulations under the Investment Advisers Act of 1940 and state Blue Sky Laws—ensuring compliance and risk mitigation. Through technical implementation frameworks, economic cost-benefit analysis, and case studies for 3-person RIAs, we demonstrate how agentic AI systems act as force multipliers, decoupling analytical bandwidth from human headcount. This enables small firms to automate document-intensive due diligence for private markets, deploy real-time macro intelligence rivaling hedge funds, achieve near-total operational automation, and deliver hyper-personalized portfolio management. The synthesis indicates that small, agentic firms can not only compete with but potentially outperform larger institutions through superior agility, deeper personalization, and enhanced compliance robustness, fundamentally reshaping the competitive landscape of investment management.

… -agent AI systems based on generative artificial intelligence … that SAFE agentic AI models can be effectively implemented. … safety and harm in agentic systems is the use of guardrails. …

Agentic artificial intelligence (AAI) represents a significant evolution in the field of AI, moving beyond traditional and generative systems toward models characterized by autonomy, adaptivity, proactiveness, and decision agency. Unlike earlier AI paradigms that were reactive or limited to narrow tasks, AAI integrates reasoning, memory, planning, and tool orchestration to pursue complex objectives with minimal human oversight. Using a systematic literature review method, this study synthesizes current knowledge on AAI by examining its conceptual foundations, practical applications, and emerging research directions. Conceptually, AAI is distinguished from automation, generative AI, and multi‐agent systems through its unique capacity to operate as a socio‐technical partner in organizational and societal contexts. In practice, AAI is being applied across sectors such as healthcare, finance, manufacturing, education, and sustainability, enabling organizations to enhance decision support, optimize processes, and improve resilience in global business contexts. However, these advancements present significant challenges, including governance, transparency, accountability, workforce transformation, and integration with legacy systems. On the research front, four major streams dominate current scholarship: human–AI collaboration and co‐agency; balancing AI autonomy with human control; governance and trust; and societal and ethical implications. To unify these insights, this paper develops an antecedent–mechanism–outcome framework linking technological, organizational, and societal enablers to the mechanisms and outcomes of AAI adoption. Building on this synthesis, a future research agenda is proposed that emphasizes conceptual refinement, responsible integration, methodological innovation, and interdisciplinary collaboration. Overall, the study contributes to both academic and managerial understanding in the global business context by highlighting AAI as both a driver of business strategy and a potential enabler of organizational excellence and sustainable development.

… Deployment with Guardrails The deployment of agentic AI systems … that can ensure financial stability, regulatory compliance, and … The governance of financial agentic AI addresses …

… and operational foundations of agentic AI and highlighting … ; and (3) introducing the “agentic AI tradeoff framework,” which outlines … , and sustainable pathways for agentic AI integration. …

… -agent AI systems based on generative artificial intelligence … that SAFE agentic AI models can be effectively implemented. … safety and harm in agentic systems is the use of guardrails. …

Agentic Artificial Intelligence systems, characterized by autonomous reasoning, memory augmentation, and adaptive planning, are rapidly reshaping technological landscapes. Unlike traditional AI or large language models, agentic AI integrates decision-making with persistent execution, enabling complex interactions across dynamic environments. However, this evolution introduces novel security risks, governance challenges, and ethical considerations that current frameworks inadequately address. This survey provides a cross-layer review of agentic AI, encompassing architectural paradigms, threat taxonomies, and governance strategies. It consolidates findings from adjacent domains such as cybersecurity, AI safety, multi-agent coordination, and ethics, offering a holistic understanding of vulnerabilities and mitigation approaches. We integrate insights from recent advances in defense architectures and governance innovations, highlighting the limitations of static policies in addressing dynamically evolving threats. Real-world deployments from industrial automation to military and policy applications reveal both successful integrations and notable failures, underscoring the urgency of resilient oversight mechanisms. Furthermore, we identify critical research gaps in benchmarking, memory integrity, adversarial defense, and normative embedding, emphasizing the need for interdisciplinary collaboration to develop adaptive, accountable, and transparent systems. This review serves as a narrative synthesis rather than a systematic literature review, aiming to bridge technical, governance, and ethical perspectives. By integrating cross-disciplinary findings, it lays the foundation for future research on securing, aligning, and governing agentic AI in real-world contexts. Ultimately, this work calls for cooperative innovation to ensure that agentic AI evolves as a trustworthy, accountable, and beneficial technology.

You’re in the middle of a critical product launch. Your agentic AI system has been managing customer feedback, routing high-priority issues to engineering, and even flagging potential PR risks before they escalate. Suddenly, your phone buzzes with a crisis alert. A customer complaint has gone viral, citing biased and unfair treatment, rooted in a decision made by your AI agent. Legal is demanding an immediate response. Your AI logs show no obvious errors. Your team scrambles to understand how the model reached this decision, only to discover gaps in documentation and oversight. You realize too late that your governance playbook was never fully implemented. Trust, both within your company and with customers, is at stake. And you’re left wondering: How did this happen?

As artificial intelligence shifts from static inference toward autonomous agency, traditional linear governance and risk frameworks are becoming increasingly inadequate. Agentic AI systems reason, plan, act, and learn within recursive decision loops, repeatedly ingesting the consequences of their own actions. This paper introduces the Circular Flow Model as a conceptual framework for understanding how risk emerges, compounds, and propagates across these autonomous cycles. By analysing the stages of Input, Model, Output, and Action as an interconnected control loop, the paper shows how small errors can accumulate into feedback-driven instability, leading to behavioural drift and operational loss of control. Building on this analysis, the paper proposes a four-domain governance framework-Structural, Execution, Memory, and Assurance guardrails-designed to bound autonomy, preserve system integrity, and maintain accountability in real-world deployments. The model provides a practical lens for organisations seeking to manage recursive risk and operational instability as AI systems transition from tools to autonomous actors.