Agentic AI, Financial Infrastructure and Governance - Emerging Challenges and Strategic Imperatives for Financial Institutions

Abstract

The emergence of agentic artificial intelligence represents a significant transition in the evolution of digital technologies within financial services. Unlike earlier generations of AI that primarily supported prediction, classification, and content generation, agentic systems are capable of planning, reasoning, coordinating actions, and executing decisions with increasing levels of autonomy. While much of the contemporary discourse surrounding these technologies focuses on productivity, efficiency, and competitive advantage, this paper argues that the most consequential implications of agentic AI are institutional and governance-related rather than purely technological.

Drawing upon the writings of Oliver Bussmann and integrating recent academic, regulatory, and industry research, the paper develops a governance-centred framework for understanding AI transformation in financial services. Five interconnected themes are examined: agentic commerce and payment infrastructure, AI governance beyond vendor management, accountability in autonomous systems, explainability and regulatory readiness, and AI infrastructure as a source of strategic advantage. Across these domains, a common pattern emerges: as autonomous systems assume greater responsibility for economic decision-making, governance capabilities become increasingly important determinants of institutional performance.

The paper argues that the defining challenge of the agentic era is an emerging accountability crisis. Traditional governance frameworks assume that responsibility can ultimately be traced to identifiable human decision-makers. Agentic systems challenge this assumption by executing complex sequences of actions independently while legal and regulatory accountability remains with financial institutions. Consequently, capabilities such as auditability, decision replayability, observability, and operational control become strategic requirements rather than technical enhancements.

The analysis suggests that AI adoption in financial services is best understood not as a technology implementation challenge but as a process of institutional redesign. Sustainable competitive advantage will derive less from access to advanced models and more from the ability to establish governance structures capable of supervising increasingly autonomous systems. The future of AI-enabled finance will therefore be shaped not only by technological capability, but by the capacity of institutions to reconcile autonomy with accountability.

Keywords: Agentic AI, Financial Services, AI Governance, Autonomous Systems, Regulatory Technology, Explainability, Auditability, Financial Infrastructure, Operational Resilience, Digital Transformation.

1. Introduction

Artificial intelligence is undergoing a significant transition from systems primarily designed for prediction, classification, and content generation towards systems capable of autonomous action. This new generation of technologies, commonly described as agentic AI, combines large language models, reasoning engines, planning capabilities, memory architectures, and tool-use mechanisms to pursue goals, execute multi-step tasks, and interact directly with external systems with varying degrees of independence from human operators (Park et al., 2024; Aldridge et al., 2026). Unlike conventional machine learning systems that generate recommendations requiring human approval, agentic systems possess the capacity to initiate, coordinate, and execute actions across complex operational environments, creating new opportunities and risks for regulated industries.

The implications for financial services are profound. Agentic AI is increasingly being deployed across payments, lending, fraud detection, anti-money laundering (AML) processes, treasury management, compliance monitoring, customer servicing, and capital markets operations. Industry research suggests that autonomous agents could significantly reduce operational costs, accelerate decision-making processes, and improve service scalability, while simultaneously increasing the speed, complexity, and opacity of financial decision-making. As financial institutions move from experimental deployments towards enterprise-wide implementation, questions surrounding accountability, explainability, governance, and regulatory compliance become increasingly consequential.

Within this context, Oliver Bussmann has argued that the strategic significance of agentic AI is frequently misunderstood. In a series of commentaries directed at financial-sector executives and board members, Bussmann contends that agentic AI should not be viewed merely as another technology upgrade or productivity-enhancing tool. Rather, he characterises it as an infrastructural transformation that fundamentally alters governance structures, operational control mechanisms, risk-management frameworks, and institutional accountability (Bussmann, 2026). His central thesis is that the primary challenge facing financial institutions is not access to increasingly capable AI models, but the development of governance architectures capable of controlling, monitoring, and auditing autonomous systems operating at scale.

Emerging academic research provides substantial support for this perspective. Agentic AI systems differ fundamentally from traditional algorithmic and machine-learning applications because they exhibit goal-directed autonomy, adaptive decision-making, tool orchestration, and, increasingly, multi-agent coordination capabilities (Aldridge et al., 2026). These characteristics challenge many of the assumptions embedded within existing financial governance and regulatory frameworks, which were developed around models that functioned primarily as decision-support tools rather than autonomous decision-makers. As Kurshan, Balch and Byrd (2025) observe, traditional model-risk management frameworks presuppose relatively stable systems operating within clearly defined decision boundaries, whereas agentic systems are dynamic, context-sensitive, and capable of generating novel courses of action in response to changing environmental conditions.

This distinction has significant implications for institutional governance. The governance challenge presented by agentic AI extends beyond familiar concerns relating to model accuracy, bias, and data quality. Instead, it raises more fundamental questions regarding accountability, liability, oversight, auditability, and human authority. As autonomous agents increasingly execute financial transactions, conduct compliance investigations, monitor fraud, or make operational decisions with limited human intervention, institutions must establish mechanisms capable of tracing, explaining, and governing machine behaviour in real time (Axelsen, Licht and Damsgaard, 2025). Consequently, the adoption of agentic AI represents not merely a technological transformation but a governance transformation, requiring boards, regulators, and senior executives to rethink the relationship between automation, institutional control, and accountability in the digital financial system.

The argument advanced in this paper is therefore consistent with Bussmann's broader proposition: the long-term competitive advantage created by agentic AI will not be determined solely by model sophistication or computational scale, but by an institution's ability to build trustworthy governance infrastructures capable of managing autonomous systems within increasingly complex regulatory environments.

2. Agentic Commerce and the Future of Payment Infrastructure

One of the most consequential implications of agentic artificial intelligence for financial services is the emergence of agentic commerce: an economic model in which autonomous software agents initiate, negotiate, authorize, and execute transactions on behalf of individuals, corporations, and potentially other autonomous systems. Unlike conventional e-commerce, where digital platforms facilitate transactions ultimately approved by human users, agentic commerce introduces a fundamentally different paradigm in which decision-making authority is partially delegated to intelligent agents capable of acting independently within predefined constraints.

Oliver Bussmann has consistently argued that the strategic significance of this development is being underestimated. Commenting on recent initiatives by Visa and Mastercard, Bussmann (2026c) notes that industry discussions have focused predominantly on the prospect of increased transaction volumes resulting from autonomous procurement, subscription management, recurring purchases, and machine-to-machine economic interactions. While such volume growth may generate substantial revenue opportunities for payment networks, Bussmann contends that the more important issue concerns the readiness of the underlying infrastructure to support autonomous economic activity at scale. In his view, the future competitiveness of payment providers will depend less on their ability to process a larger number of transactions and more on their ability to establish trusted governance frameworks capable of managing machine-initiated transactions safely, transparently, and compliantly.

This argument reflects broader developments in the economics of digital platforms and financial infrastructure. Historically, payment systems have been designed around the assumption that human actors initiate, verify, and remain accountable for transactions. Agentic commerce challenges this assumption by introducing autonomous decision-makers capable of operating continuously and at machine speed. Consequently, payment infrastructures must increasingly manage interactions between agents rather than between humans alone. As autonomous systems become capable of executing procurement decisions, treasury operations, inventory replenishment, and financial transfers without direct human intervention, traditional mechanisms of authentication, authorization, dispute resolution, and liability allocation become insufficient (Aldridge et al., 2026).

Recent industry developments lend support to Bussmann’s assessment. Visa and Mastercard have both announced initiatives aimed at enabling autonomous commerce through AI-enabled payment credentials and programmable transaction frameworks. Mastercard’s development of dedicated trust layers for agentic commerce is particularly significant because it acknowledges that identity verification, consent management, transaction explainability, and agent authentication may become as important as payment processing itself (Mastercard, 2026). Similarly, pilot programmes involving Mastercard, ING, Worldline, and other financial institutions demonstrate increasing recognition that scalable agentic commerce requires new approaches to authorization, delegated consent, and accountability in environments where human intervention is minimal or absent (Bussmann, 2026).

From an academic perspective, agentic commerce introduces several categories of systemic risk. First, autonomous purchasing systems increase transaction velocity, potentially amplifying operational failures, fraudulent activity, and market disruptions before human oversight mechanisms can intervene (Aldridge et al., 2026). Second, the delegation of purchasing authority to intelligent agents creates novel questions concerning legal responsibility and liability. Existing payment regulations generally presume the existence of a clearly identifiable human decision-maker; however, autonomous agents blur the distinction between user intent and machine execution. Third, machine-to-machine commerce creates new cybersecurity vulnerabilities, as malicious actors may seek to manipulate agent behaviour, compromise delegated authority structures, or exploit weaknesses in agent coordination protocols (World Economic Forum, 2025).

These developments suggest that the future architecture of payment systems will differ fundamentally from current models. Rather than functioning primarily as transaction-processing networks, payment infrastructures are likely to evolve into comprehensive governance platforms that provide identity verification, policy enforcement, risk monitoring, auditability, dispute resolution, and accountability mechanisms for autonomous economic actors. In this respect, trust becomes an infrastructural capability rather than merely a legal or institutional characteristic. The institutions that successfully embed governance, explainability, and control mechanisms into agentic payment ecosystems may therefore occupy a strategic position analogous to that of payment networks in the early internet era.

Bussmann’s broader insight is that agentic commerce should not be understood merely as the next stage of digital payments evolution. Rather, it represents the emergence of a new economic infrastructure in which autonomous agents increasingly participate as operational actors within financial ecosystems. The critical competitive question is therefore not whether transaction volumes will increase—a proposition widely accepted across the industry—but which institutions will build the governance, trust, and accountability architectures necessary to support autonomous commerce at global scale. As machine-driven transactions become more prevalent, the ability to govern autonomous economic activity may become a more valuable source of competitive advantage than transaction-processing capability itself.

3. AI Governance Beyond the Vendor Model

A recurring theme throughout Bussmann’s commentary is the tendency of financial institutions to frame artificial intelligence governance primarily as a third-party vendor management issue. According to Bussmann (2026e), many banks continue to approach AI adoption through a procurement lens, assuming that technological risks can be mitigated through vendor selection, contractual safeguards, and external certifications. This perspective, however, understates a fundamental reality of financial regulation: while institutions may outsource technology, they cannot outsource accountability. Regardless of where an AI system is developed, trained, or hosted, regulatory responsibility for its operation ultimately remains with the financial institution deploying it.

This distinction is becoming increasingly important as financial institutions integrate advanced AI systems into customer-facing services, compliance functions, risk management processes, and operational workflows. Traditional outsourcing arrangements typically involve clearly defined services operating within predictable parameters. By contrast, contemporary AI systems—and particularly agentic AI architectures—exhibit adaptive behaviour, evolving decision pathways, and varying degrees of autonomy that complicate conventional approaches to third-party oversight (Kurshan, Balch and Byrd, 2025). As a result, governance challenges arise not only from the quality of the underlying technology but also from the institution’s ability to understand, supervise, and control AI-enabled decision-making processes after deployment.

Bussmann argues that the central governance challenge therefore concerns institutional capability rather than vendor capability. The critical question is not whether a technology provider claims its models are safe, explainable, or compliant; rather, it is whether the institution itself can independently audit, explain, monitor, and intervene in AI-driven processes operating within its environment. This distinction reflects a broader shift in governance thinking from procurement-based assurance towards operational accountability. In highly regulated sectors such as banking, evidence of control must be demonstrable, continuous, and institutionally owned rather than contractually delegated (Bussmann, 2026).

Emerging academic literature strongly supports this perspective. García-Llorente and Olmeda (2026), in their analysis of algorithmic governance within banking institutions, identify accountability structures, governance responsibilities, auditability requirements, and documentation standards as among the most significant unresolved challenges associated with AI adoption. Their findings suggest that effective governance depends not merely on technical performance but on the existence of organisational mechanisms capable of allocating responsibility and maintaining oversight throughout the lifecycle of AI systems.

Similarly, Strauss et al. (2025) observe a significant imbalance within contemporary AI governance research. While considerable attention has been devoted to model development, validation, fairness testing, and pre-deployment risk assessment, comparatively little research addresses the governance challenges associated with systems operating in production environments. This gap is particularly problematic in financial services, where regulatory expectations increasingly focus on ongoing monitoring, operational resilience, and the ability to demonstrate control over continuously evolving systems. In practice, institutions frequently possess sophisticated model validation frameworks yet lack equivalent capabilities for monitoring autonomous behaviour after deployment.

The significance of this challenge is amplified by recent regulatory developments. The European Union’s AI Act, the Digital Operational Resilience Act (DORA), supervisory guidance from the European Banking Authority (EBA), and emerging expectations from regulators such as the UK Financial Conduct Authority (FCA) and the U.S. Federal Reserve all reinforce the principle that accountability remains with the regulated entity regardless of whether AI capabilities are sourced internally or externally (European Union, 2024; EBA, 2025). These frameworks increasingly require institutions to demonstrate not only that AI systems function as intended, but also that they can identify, explain, and remediate failures when they occur.

The implications are particularly significant for agentic AI systems. Traditional governance approaches assume relatively static models whose outputs can be periodically reviewed and validated. Agentic systems, by contrast, may dynamically orchestrate tools, interact with external systems, adapt their behaviour to changing circumstances, and generate novel decision pathways that were not explicitly anticipated by their designers (Aldridge et al., 2026). Consequently, governance becomes an operational discipline rather than a compliance exercise. Institutions must develop capabilities for continuous monitoring, behavioural observability, intervention management, decision replayability, and real-time auditability.

This shift aligns with emerging concepts in algorithmic accountability and operational resilience. Research increasingly suggests that trustworthy AI cannot be achieved solely through model transparency or explainability. Instead, institutions require governance architectures capable of producing verifiable evidence regarding what an AI system did, why it acted, which data informed its actions, and whether those actions remained within approved operational boundaries (Axelsen, Licht and Damsgaard, 2025). Such capabilities become particularly important when autonomous systems influence customer outcomes, financial transactions, regulatory reporting, or risk-management decisions.

The broader implication is that AI governance must evolve beyond the vendor model altogether. Vendor assurances, certifications, and compliance attestations may remain important components of risk management, but they are insufficient as primary governance mechanisms. Effective governance requires institutions to establish independent oversight capabilities that extend beyond contractual relationships and enable continuous supervision of AI systems throughout their operational lifecycle. In this sense, governance becomes a core institutional competency rather than a procurement function.

Bussmann’s central insight is therefore both simple and consequential: financial institutions may acquire AI from external providers, but they cannot transfer responsibility for its outcomes. As AI systems become increasingly autonomous, the strategic differentiator will not be access to advanced models alone, but the organisational capacity to govern, monitor, and control those models in ways that satisfy regulators, customers, and stakeholders. Institutions that fail to develop these capabilities risk discovering that their greatest AI vulnerability lies not in the technology they purchased, but in the governance mechanisms they failed to build.

4. The Emerging Accountability Crisis in Agentic AI

Perhaps Bussmann’s most significant contribution to contemporary discussions of artificial intelligence in financial services concerns the question of accountability. While much of the current discourse surrounding agentic AI focuses on productivity gains, automation opportunities, and competitive advantage, Bussmann repeatedly returns to a more fundamental concern: who remains accountable when autonomous systems make decisions, execute actions, and generate outcomes without direct human intervention (Bussmann, 2026f). His argument is that the emergence of agentic AI challenges one of the foundational assumptions underpinning modern financial governance—that responsibility can ultimately be traced to identifiable human decision-makers.

Traditional financial institutions operate within governance frameworks built around clear chains of accountability. Whether approving a loan, executing a payment, reporting regulatory data, or managing financial risk, systems have historically been designed to ensure that a human individual retains ultimate authority over consequential decisions. Technology has served primarily as a decision-support mechanism, providing information, recommendations, and analytical capabilities while leaving final responsibility with designated personnel. Regulatory frameworks, legal doctrines, and organisational structures have evolved around this assumption, enabling accountability to be assigned when failures occur.

Agentic AI fundamentally alters this arrangement. Unlike conventional automation technologies, agentic systems are capable of independently planning, coordinating, and executing sequences of actions across multiple systems and decision environments. A single autonomous agent may evaluate information, initiate transactions, communicate with counterparties, update records, trigger subsequent actions, and adapt its behaviour in response to changing circumstances without requiring human approval at each stage. In doing so, agentic systems blur the distinction between decision support and decision execution, creating uncertainty regarding where accountability ultimately resides (Bussmann, 2026).

This challenge is particularly significant within financial services because accountability is not merely an operational concern but a legal and regulatory requirement. Banking, insurance, payments, and capital markets operate under governance regimes that assume identifiable responsibility for decisions affecting customers, counterparties, markets, and financial stability. Regulatory frameworks generally recognise institutions and individuals as accountable actors; they do not recognise software agents as legal entities capable of assuming liability. Consequently, when autonomous systems execute actions that produce undesirable outcomes, responsibility inevitably reverts to the institution regardless of the degree of machine autonomy involved.

Bussmann captures this dilemma succinctly when he observes that in agentic environments “the audit trail shows a machine, not a person” (Bussmann, 2026f). This observation highlights a growing disconnect between technological capability and institutional accountability. As autonomous systems assume greater responsibility for operational decisions, the traditional mechanisms through which organisations establish responsibility become increasingly difficult to apply. The challenge is not simply determining what decision was made, but understanding who authorised it, what reasoning process informed it, and which governance mechanisms were responsible for overseeing it.

Recent academic research increasingly reflects these concerns. Kurshan, Balch and Byrd (2025) argue that prevailing model risk management frameworks were designed for a fundamentally different generation of technologies. Existing governance approaches assume relatively static algorithms operating within stable environments, allowing institutions to conduct periodic validation, performance testing, and risk assessment. Agentic systems violate many of these assumptions. Their behaviour may evolve dynamically over time, they interact with changing environments, and they frequently engage in multi-step decision processes that cannot be fully anticipated during initial validation exercises. As a result, governance frameworks designed around static models become increasingly inadequate for supervising adaptive autonomous systems.

This observation aligns with a broader shift in the literature from model-centric governance towards system-centric governance. Traditional approaches to AI oversight focus primarily on evaluating the properties of individual models, including accuracy, fairness, robustness, and explainability. Agentic systems require a different perspective because risks emerge not only from individual models but from interactions between models, agents, external systems, and organisational processes (Bengio et al., 2024). Accountability therefore becomes a property of the entire socio-technical system rather than of any individual algorithm.

A further challenge arises from the limitations of explainability as a governance mechanism. While explainable AI has received significant attention within both academic research and regulatory policy, recent scholarship suggests that explainability alone may be insufficient for governing autonomous systems operating in complex environments. Aggarwal (2026) argues that institutions require what he terms “decision replayability”: the capacity to reconstruct, review, and analyse the complete sequence of decisions leading to a particular outcome. In highly autonomous environments, understanding a final decision is often impossible without visibility into the broader chain of intermediate actions, contextual interpretations, tool selections, and reasoning pathways that preceded it.

Decision replayability represents a significant evolution in governance thinking. Traditional audit mechanisms focus primarily on recording outcomes and key decision points. Agentic systems require much richer forms of evidence. Institutions must increasingly capture information regarding what data an agent accessed, how objectives were interpreted, which external systems were consulted, what alternative courses of action were considered, and why a particular decision pathway was ultimately selected. Such capabilities are essential not only for internal governance but also for responding to regulatory inquiries, customer disputes, operational failures, and legal challenges.

The accountability challenge becomes even more complex when multiple autonomous agents interact. Emerging agentic architectures increasingly rely on collaborative systems in which specialised agents perform distinct functions, share information, coordinate activities, and jointly contribute to outcomes. In these environments, causality becomes difficult to establish because no single agent may be solely responsible for a particular decision. Responsibility becomes distributed across a network of interacting autonomous components, creating governance challenges analogous to those encountered in complex financial networks themselves (Axelsen, Licht and Damsgaard, 2025).

From a regulatory perspective, these developments raise fundamental questions concerning institutional oversight. Existing regulatory frameworks—including the European Union AI Act, DORA, supervisory guidance from the European Banking Authority, and emerging international standards on AI governance—continue to assume that regulated entities can demonstrate effective control over automated systems (European Union, 2024; EBA, 2025). The deployment of increasingly autonomous agents therefore creates a potential accountability gap between regulatory expectations and operational reality. Institutions may possess highly capable systems while lacking the mechanisms necessary to demonstrate meaningful oversight of those systems.

This gap is likely to become one of the defining governance challenges of the agentic era. Financial institutions face growing pressure to increase automation in order to improve efficiency, reduce costs, and remain competitive. Simultaneously, regulators continue to demand evidence of accountability, explainability, and human oversight. These objectives may increasingly come into tension as machine autonomy expands. Excessive controls can undermine the efficiency gains promised by agentic systems, while insufficient controls create unacceptable governance and compliance risks.

The institutions most likely to succeed will therefore be those capable of reconciling autonomy with accountability. This requires moving beyond traditional governance models based on periodic review and static validation towards dynamic governance architectures capable of continuous monitoring, behavioural observability, intervention management, and comprehensive auditability. Accountability in agentic systems cannot depend solely on identifying a human decision-maker after the fact. Instead, it must be embedded within the operational design of autonomous systems themselves.

Viewed in this context, Bussmann’s accountability thesis represents one of the most important governance insights emerging from contemporary discussions of AI in financial services. The central challenge is not whether autonomous systems can make decisions effectively. Increasing evidence suggests that they can. The challenge is whether institutions can establish governance mechanisms capable of preserving accountability when decision-making authority becomes increasingly distributed across autonomous systems. In highly regulated industries such as finance, the answer to that question may ultimately determine the pace, scale, and success of agentic AI adoption.

5. Explainability, Auditability and Regulatory Readiness

Throughout his writings on agentic AI, Bussmann consistently frames explainability, monitoring, and auditability not as technical features but as strategic capabilities. His central argument is that as financial institutions increasingly deploy autonomous systems into regulated environments, the primary governance challenge shifts from achieving operational performance to maintaining regulatory defensibility. In this context, the critical question is no longer whether an AI system can execute tasks effectively, but whether the institution can explain, reconstruct, and justify those actions when challenged by regulators, auditors, customers, or courts (Bussmann, 2026f).

This distinction reflects a broader evolution in AI governance. Early discussions of explainable AI focused primarily on transparency and model interpretability, seeking to provide insights into how algorithms generated particular outputs (Doshi-Velez and Kim, 2017). However, the emergence of agentic AI systems introduces a more complex challenge. Autonomous agents do not simply produce outputs; they reason, plan, coordinate actions, invoke external tools, access multiple data sources, and execute sequences of decisions over time. Consequently, explainability can no longer be understood solely as a property of individual model outputs. Instead, it must encompass the entire chain of reasoning, decision-making, and execution that leads to an observable outcome.

Recent research on agentic governance architectures reinforces this perspective. Axelsen, Licht and Damsgaard (2025), examining the use of agentic AI in financial crime compliance, argue that successful deployment requires four interdependent governance capabilities: explainability, traceability, bounded autonomy, and comprehensive audit logging. Their research suggests that autonomous systems operating in highly regulated environments must be capable of producing detailed evidence regarding what actions were taken, which information informed those actions, and whether operational decisions remained within approved governance boundaries. Importantly, these requirements extend beyond traditional model validation and move towards continuous operational oversight.

Similarly, Doyle-Spare (2026) proposes the Agentic 3Cs Framework—Context, Control, and Coordination—as a governance model for trustworthy autonomous systems. The framework identifies a critical governance challenge emerging at what the author describes as the "reasoning layer": the stage at which agents interpret contextual information, formulate objectives, and determine appropriate courses of action before execution. Unlike traditional software systems, where behaviour is largely deterministic and predefined, agentic systems exhibit adaptive decision-making characteristics that can produce novel behaviours in response to changing environmental conditions. This introduces governance challenges that cannot be addressed solely through conventional software assurance methods.

The significance of the reasoning layer has profound implications for financial institutions. Existing governance frameworks in banking have historically focused on inputs, outputs, and controls. Regulatory assessments typically examine whether data quality standards are maintained, whether approved models are used, and whether outcomes remain within acceptable risk thresholds. Agentic systems introduce an intermediate layer of autonomous reasoning that may substantially influence outcomes without being directly observable through conventional monitoring mechanisms (Kurshan, Balch and Byrd, 2025). Consequently, institutions increasingly require governance architectures capable of capturing and analysing the decision pathways through which agents arrive at particular actions.

This challenge is further amplified by evolving regulatory expectations. Legislative frameworks such as the European Union's AI Act, the Digital Operational Resilience Act (DORA), and emerging supervisory guidance from global financial regulators increasingly emphasize transparency, accountability, human oversight, and demonstrable control over automated decision-making systems (European Union, 2024; BIS, 2025). Regulatory compliance is therefore becoming less concerned with whether AI systems function effectively and more concerned with whether institutions can provide verifiable evidence of governance. In practical terms, explainability is evolving from a desirable characteristic into a regulatory requirement.

A growing body of scholarship argues that explainability alone may be insufficient to meet these expectations. Recent work on algorithmic accountability increasingly emphasizes concepts such as observability, replayability, and auditability as more robust foundations for governance (Aggarwal, 2026). Replayability refers to the ability to reconstruct a complete decision sequence after an event has occurred, enabling institutions to determine precisely how an autonomous system reached a particular conclusion. Observability extends beyond logging outputs to encompass visibility into internal decision processes, interactions between agents, external tool usage, and contextual influences on behaviour. Together, these capabilities create the evidentiary foundation necessary for regulatory review and institutional accountability.

From a strategic perspective, Bussmann's argument is therefore highly significant. Institutions that deploy autonomous systems without comprehensive monitoring infrastructures may achieve short-term efficiency gains but simultaneously accumulate substantial governance risk. The absence of robust audit trails, behavioural monitoring, and decision traceability creates what may be described as a "compliance debt"—a growing gap between operational autonomy and regulatory accountability. This debt often remains invisible until a significant failure, regulatory inquiry, or customer dispute exposes deficiencies in governance arrangements.

The broader implication is that regulatory readiness in the age of agentic AI depends increasingly on an institution's ability to generate trustworthy evidence regarding machine behaviour. Explainability, auditability, and monitoring are therefore not ancillary technical capabilities but foundational elements of institutional governance. As autonomous systems assume greater responsibility for financial decision-making, the competitive advantage will accrue not merely to organizations capable of deploying agentic AI at scale, but to those capable of governing it in ways that remain transparent, defensible, and auditable under regulatory scrutiny.

Viewed through this lens, Bussmann's central insight becomes particularly salient: the greatest risk associated with agentic AI is not necessarily that autonomous systems will make incorrect decisions, but that institutions will be unable to explain, justify, or reconstruct those decisions when accountability is demanded. In highly regulated industries such as financial services, the ability to demonstrate control may ultimately become as important as the ability to automate.

6. AI Infrastructure as Strategic Competitive Advantage

A recurring theme throughout Bussmann’s analysis of financial services is that the competitive landscape of artificial intelligence is being fundamentally misunderstood. While public discourse frequently focuses on advances in model performance, benchmark scores, and the capabilities of individual AI systems, Bussmann (2026g) argues that sustainable competitive advantage will be determined by a different set of factors: infrastructure ownership, data architecture, governance capabilities, and regulatory positioning. In his view, the institutions most likely to succeed in the next phase of digital transformation will be those that integrate AI infrastructure, enterprise data architecture, and regulatory strategy into a coherent operating model rather than treating them as independent technological or compliance initiatives.

This perspective reflects a broader shift in how AI is understood within both academic and industry contexts. Early waves of AI adoption were often characterized as software innovation challenges, with competitive advantage deriving primarily from superior algorithms or model development capabilities. Increasingly, however, AI is being recognised as a general-purpose infrastructure technology analogous to electricity, telecommunications networks, or cloud computing (Brynjolfsson, Li and Raymond, 2023). As with previous infrastructure revolutions, the greatest economic value may not accrue to those who create individual applications but to those who control the foundational resources upon which entire ecosystems depend.

Recent developments within the global AI ecosystem provide substantial evidence for this shift. Governments, technology firms, and institutional investors are allocating unprecedented levels of capital towards sovereign AI infrastructure, hyperscale cloud capacity, advanced semiconductor supply chains, and strategic computing resources. Such investments reflect growing recognition that access to computational power, data assets, and governance frameworks increasingly constitutes a strategic asset rather than a technical prerequisite (McKinsey Global Institute, 2025). In this context, AI competitiveness is becoming multidimensional, encompassing not only model capabilities but also access to compute, proprietary datasets, trusted operating environments, and regulatory legitimacy.

For financial institutions, these developments are particularly significant because the value generated by AI depends heavily upon access to high-quality, proprietary, and context-rich data. Bussmann’s observation that future enterprise competition will increasingly depend upon ownership of the “data fabric” aligns closely with emerging scholarship on data-centric AI and digital platform competition (Davenport and Mittal, 2022). The concept of a data fabric extends beyond conventional data management architectures to encompass the integrated ecosystem of transactional data, customer interactions, operational records, governance metadata, and contextual information that enables AI systems to generate institution-specific insights and actions.

From a strategic perspective, this distinction is critical. Advances in foundation models have significantly reduced barriers to accessing sophisticated AI capabilities. Increasingly powerful models are available through commercial APIs, open-source ecosystems, and cloud-based platforms, reducing the likelihood that model access alone will constitute a durable competitive advantage. As a result, the source of differentiation shifts from model ownership to the quality of proprietary data, the effectiveness of integration architectures, and the institution's ability to operationalise AI within complex business environments (Nadella, 2025). In this sense, AI models are becoming increasingly commoditised, while data ecosystems remain highly differentiated.

This argument finds support in the resource-based view (RBV) of the firm, which suggests that sustainable competitive advantage arises from resources that are valuable, rare, difficult to imitate, and organisationally embedded (Barney, 1991). Applied to the AI economy, foundational models may be valuable but are increasingly accessible to competitors. By contrast, proprietary financial data, institutional trust, regulatory relationships, operational expertise, and governance infrastructures remain highly specific, difficult to replicate, and deeply embedded within organisational processes. These assets therefore represent more durable sources of competitive advantage than model performance alone.

The importance of governance infrastructure further reinforces Bussmann’s argument. As discussed previously, the deployment of agentic AI increasingly requires institutions to demonstrate explainability, accountability, auditability, and operational control. Consequently, governance capabilities themselves become strategic assets rather than compliance obligations. Institutions capable of governing AI effectively can deploy autonomous systems more rapidly, scale innovation with greater confidence, and respond more effectively to regulatory scrutiny. Conversely, organizations lacking governance maturity may find themselves constrained by compliance concerns regardless of their access to advanced technologies (European Banking Authority, 2025).

Regulatory positioning also emerges as a significant competitive factor. The accelerating development of AI-related regulation—including the European Union AI Act, DORA, and emerging supervisory frameworks across major jurisdictions—suggests that regulatory alignment will increasingly influence the pace and scope of AI adoption. Institutions that proactively incorporate regulatory requirements into technology architecture may benefit from reduced implementation friction and enhanced stakeholder trust. In contrast, firms treating compliance as a reactive exercise risk creating operational bottlenecks that undermine the value generated by AI investments.

The growing emphasis on sovereign AI infrastructure further highlights the strategic convergence of technology, governance, and geopolitics. Investments in national AI capabilities, data localisation requirements, and concerns regarding digital sovereignty indicate that access to computational resources and AI services is becoming intertwined with broader questions of economic resilience and strategic autonomy (World Economic Forum, 2025). Financial institutions operating across jurisdictions must therefore consider not only technological efficiency but also regulatory compatibility, data residency requirements, and geopolitical dependencies when designing AI strategies.

Viewed collectively, these developments support Bussmann’s central contention that AI infrastructure, data architecture, and regulatory positioning should be understood as components of a single strategic system. Treating these domains as separate workstreams risks creating fragmented capabilities that are difficult to scale and govern effectively. By contrast, institutions that align infrastructure investment, data strategy, governance frameworks, and regulatory engagement are likely to establish stronger foundations for long-term competitive advantage.

The broader implication is that the next phase of competition in financial services may be determined less by who possesses the most advanced AI models and more by who controls the infrastructure through which those models create value. Models can be licensed, replicated, or replaced. Proprietary data ecosystems, trusted governance frameworks, regulatory credibility, and integrated digital infrastructures are significantly more difficult to imitate. In this environment, competitive advantage increasingly resides not at the application layer of AI, but at the infrastructural layer that enables autonomous systems to operate effectively, securely, and at scale.

7. Frontier AI Governance and Institutional Risk

Bussmann’s governance concerns extend beyond the deployment and operation of AI systems within financial institutions to encompass the broader evolution of frontier AI itself. In commenting on calls from leading AI developers for greater caution in the development of increasingly capable systems, Bussmann frames the issue not as a technical debate concerning model architectures or performance benchmarks, but as a strategic governance challenge for boards and senior executives (Bussmann, 2026). His central argument is that financial institutions are becoming increasingly dependent on a small number of external AI providers whose future technological trajectories may have significant implications for operational resilience, regulatory compliance, and institutional risk. Consequently, boards must consider not only the governance of deployed AI systems, but also the governance practices and developmental pathways of the organizations upon which those systems depend.

This perspective reflects an important evolution in the nature of technology risk. Historically, financial institutions assessed third-party technology providers primarily through operational, cybersecurity, and vendor-management frameworks. Contemporary frontier AI systems introduce a different category of dependency. Unlike traditional software vendors, frontier AI developers continuously modify and improve the capabilities of their models, often at a pace that exceeds the ability of customers, regulators, and governance frameworks to fully evaluate emerging risks (Bengio et al., 2024). As a result, institutions may find themselves relying upon technologies whose future capabilities, limitations, and behavioural characteristics remain uncertain.

The relevance of this concern has increased significantly as frontier AI developers have begun publicly expressing reservations regarding the pace of capability advancement. Anthropic, one of the world's leading AI laboratories, has argued that society may require mechanisms to slow or pause the development of increasingly powerful AI systems if alignment research, governance frameworks, and societal institutions fail to keep pace with technological progress. The company's concerns focus particularly on the possibility of increasingly autonomous systems, recursive capability improvement, and the growing gap between technological advancement and governance preparedness. While such scenarios remain contested within both academic and industry communities, they highlight a broader governance challenge: institutions are increasingly dependent on technologies whose future developmental trajectories may be difficult to predict or control.

From a financial services perspective, this challenge extends beyond speculative discussions concerning artificial general intelligence or existential risk. The more immediate issue concerns strategic dependency. A growing number of banks, asset managers, insurers, and payment providers rely on foundation models developed by a small group of technology companies. These models increasingly support customer service operations, fraud detection, compliance processes, software development, risk analysis, and decision-support functions. Consequently, changes in the capabilities, governance structures, ownership arrangements, regulatory status, or commercial strategies of frontier AI providers may have direct implications for financial institutions' operational and strategic positions.

Recent scholarship on systemic AI risk supports this interpretation. Bommasani et al. (2021) argue that foundation models create new forms of concentration risk because large numbers of downstream users become dependent upon common technological infrastructures. Similarly, Hendrycks et al. (2023) identify governance, accountability, and systemic dependency as critical challenges arising from the increasing centralisation of advanced AI capabilities. In financial services, where operational resilience and third-party risk management are already central regulatory concerns, such concentration effects may create vulnerabilities that extend beyond traditional vendor relationships.

Bussmann's analysis is particularly relevant because it reframes frontier AI governance as an enterprise risk management issue rather than a public policy debate. Traditional AI governance frameworks typically focus on deployed systems, examining questions of fairness, explainability, robustness, and compliance. However, frontier AI introduces an additional layer of risk associated with future capability development. Institutions must therefore evaluate not only how AI systems perform today, but how they may evolve tomorrow. This includes assessing whether providers possess adequate governance mechanisms, alignment research programmes, safety testing protocols, and escalation procedures capable of managing increasingly powerful technologies (Bussmann, 2026h).

The challenge is compounded by the possibility that future AI systems may exhibit capabilities that significantly exceed those assumed during initial procurement and deployment decisions. Regulatory frameworks such as the European Union AI Act increasingly recognise that risk profiles may change throughout the lifecycle of AI systems, requiring ongoing assessment and governance rather than one-time approval processes (European Union, 2024). For boards, this implies that AI oversight cannot be limited to implementation decisions alone. Governance must become a continuous activity that monitors technological developments, vendor governance practices, and emerging regulatory expectations.

An additional dimension of risk concerns strategic optionality. As competition among frontier AI providers intensifies, differences in governance philosophies, regulatory relationships, intellectual property strategies, and geographic positioning may significantly influence the future availability and suitability of particular AI systems. Bussmann frequently emphasizes that institutions should evaluate AI vendors not only on technical performance but also on their long-term sustainability, governance maturity, and ability to operate within increasingly complex regulatory environments. This reflects a growing recognition that the viability of AI providers may become as important as the capabilities of their models.

From a governance perspective, these developments imply an expansion of board responsibilities. Boards have traditionally been expected to oversee operational risk, financial risk, cyber risk, and third-party risk. The rise of frontier AI introduces a new category that might be described as capability trajectory risk: the possibility that the future evolution of foundational technologies materially alters the institution's risk profile, competitive position, or regulatory obligations. Managing such risks requires governance mechanisms capable of addressing uncertainty, monitoring external technological developments, and evaluating strategic dependencies that may extend far beyond conventional procurement relationships.

The broader implication is that frontier AI governance can no longer be treated as a matter solely for technology providers, policymakers, or AI researchers. As financial institutions become increasingly dependent upon a small number of advanced AI ecosystems, the governance practices of frontier model developers become a matter of direct institutional concern. Boards must therefore evaluate not only the risks associated with the systems they deploy, but also the developmental pathways, governance structures, and strategic resilience of the organizations upon which those systems depend.

Viewed through this lens, Bussmann's central insight is that the most significant AI governance challenge may not be the behaviour of today's deployed systems, but the uncertainty surrounding the capabilities and governance of tomorrow's foundational models. In an environment characterised by rapid technological advancement, governance increasingly requires institutions to manage not only operational risks, but also the risks arising from dependence on technologies whose future evolution remains uncertain.

8. Discussion

The analysis presented in this paper suggests that Oliver Bussmann’s contributions extend beyond industry commentary and constitute a coherent governance framework for understanding the implications of agentic AI in financial services. Across his writings, Bussmann consistently advances a governance-first interpretation of technological transformation, arguing that the long-term consequences of agentic AI are likely to be determined less by advances in model capability than by the ability of institutions to govern increasingly autonomous systems. While many discussions of artificial intelligence focus on computational performance, productivity gains, or competitive differentiation through technology adoption, Bussmann repeatedly directs attention toward the institutional structures required to manage autonomy at scale.

Three interconnected propositions emerge from this analysis.

First, agentic AI should be understood as an infrastructural transformation rather than merely an operational enhancement. Throughout his commentary on autonomous payments, AI-native banking, digital assets, and financial market infrastructure, Bussmann argues that agentic systems are reshaping the underlying architecture through which economic activity is coordinated (Bussmann, 2026). This interpretation aligns with broader scholarship on general-purpose technologies, which suggests that transformative innovations generate their greatest economic impact not through isolated productivity improvements but through the reconfiguration of institutional and organisational systems (Brynjolfsson, Rock and Syverson, 2021). From this perspective, agentic AI represents a shift in how financial decisions are initiated, executed, monitored, and governed rather than simply a more efficient means of performing existing tasks.

Second, governance capabilities are likely to become more strategically important than model capabilities as AI adoption matures. Contemporary AI discourse frequently focuses on competition between models, benchmark performance, and computational scale. However, both the academic literature and the evidence examined in this paper suggest that model capabilities are increasingly becoming commoditised. Access to advanced foundation models is expanding through cloud providers, open-source ecosystems, and commercial partnerships, reducing the likelihood that model ownership alone will provide a sustainable source of competitive advantage (Bommasani et al., 2021). By contrast, governance capabilities—including explainability, auditability, observability, operational resilience, and regulatory compliance—remain institution-specific and difficult to replicate. As regulatory scrutiny intensifies and autonomous systems become more deeply embedded in critical business processes, these governance capabilities may emerge as the principal determinant of successful AI adoption.

Third, institutions that develop robust accountability, auditability, and operational control mechanisms are likely to capture disproportionate value from AI-enabled transformation. A recurring theme throughout Bussmann’s writings is that the principal challenge associated with agentic AI is not whether autonomous systems can perform effectively, but whether organisations can maintain meaningful oversight of those systems once deployed. This argument is increasingly reflected in academic research on agentic governance, algorithmic accountability, and financial regulation (Axelsen, Licht and Damsgaard, 2025; Kurshan, Balch and Byrd, 2025). Institutions capable of demonstrating control over autonomous systems may be able to deploy AI more aggressively, respond more effectively to regulatory expectations, and scale innovation with greater confidence. Conversely, organisations lacking governance maturity may find themselves constrained by compliance concerns, operational risks, and stakeholder distrust regardless of their technological sophistication.

Taken together, these propositions point towards a broader theoretical shift in how AI transformation should be understood. Much of the existing literature conceptualises AI as a technological capability that organisations acquire and deploy. Bussmann’s perspective suggests an alternative interpretation in which AI adoption represents a process of institutional redesign. The critical challenge is not merely integrating intelligent technologies into existing organisations but adapting governance structures, accountability frameworks, and operational processes to accommodate autonomous decision-making systems. This distinction is particularly important in financial services, where regulatory obligations, fiduciary responsibilities, and systemic stability considerations impose constraints that are less pronounced in many other industries.

The findings also suggest that prevailing approaches to AI governance may underestimate the significance of operational oversight. Traditional governance frameworks have focused heavily on model validation, fairness assessment, and explainability prior to deployment. While these considerations remain important, the emergence of agentic systems introduces new requirements relating to continuous monitoring, behavioural observability, decision traceability, and real-time intervention capabilities (Aggarwal, 2026). Governance therefore evolves from a periodic compliance activity into a continuous operational discipline. This observation supports emerging scholarship arguing that accountability in autonomous systems depends less on understanding isolated model outputs and more on maintaining visibility into ongoing system behaviour.

An additional contribution of Bussmann’s framework concerns the relationship between technology strategy and institutional resilience. Across discussions of frontier AI, digital sovereignty, payment infrastructure, and AI-native banking, he consistently emphasizes the importance of infrastructure, governance, and strategic dependency management. This perspective aligns with growing concerns regarding concentration risk within the AI ecosystem, where a relatively small number of providers increasingly supply foundational technologies used across entire industries (Financial Stability Board, 2025). The implication is that AI governance must extend beyond the management of internal systems to encompass oversight of external technological dependencies, infrastructure providers, and evolving regulatory environments.

The broader significance of these findings is that they challenge a dominant assumption underlying much of the current AI discourse: namely, that technological capability is the primary determinant of competitive success. The evidence examined in this paper suggests that as AI systems become more autonomous, institutional factors become increasingly important. Trust, accountability, governance, regulatory legitimacy, and operational control may ultimately prove more consequential than incremental improvements in model performance. In this sense, the competitive dynamics of AI adoption increasingly resemble those associated with financial infrastructure, risk management, and organisational capability development rather than conventional software innovation.

The emerging academic literature largely supports this interpretation. Research on agentic finance, algorithmic accountability, AI governance architectures, operational resilience, and systemic technology risk increasingly suggests that the greatest challenges associated with advanced AI systems arise not from capability expansion itself, but from deficiencies in institutional oversight and governance (Bengio et al., 2024; Axelsen, Licht and Damsgaard, 2025; Financial Stability Board, 2025). The central question is therefore not whether AI systems will become capable of performing increasingly complex tasks—a trend that appears likely to continue—but whether institutions can develop governance mechanisms capable of exercising meaningful control over those capabilities.

Viewed collectively, Bussmann’s writings point towards a fundamental proposition: the future of AI in financial services will be determined not primarily by the sophistication of autonomous systems, but by the sophistication of the institutions responsible for governing them. In the era of agentic AI, competitive advantage may ultimately depend less on what organisations automate and more on what they can prove, control, and account for once automation occurs.

9. Conclusion

This paper has examined the implications of agentic artificial intelligence in financial services through the governance-oriented perspective developed across the writings of Oliver Bussmann. While discussions of AI frequently emphasize advances in model performance, computational scale, and automation capabilities, the evidence reviewed throughout this study suggests that the most significant challenges associated with agentic AI are institutional rather than technological. As autonomous systems become capable of initiating, coordinating, and executing increasingly complex financial activities, the central question is no longer whether machines can perform these tasks effectively, but whether institutions can maintain meaningful oversight of systems that increasingly act on their behalf.

The analysis identifies a consistent intellectual thread running throughout Bussmann’s work: the future of financial services will be determined less by the sophistication of artificial intelligence itself than by the sophistication of the governance frameworks surrounding it. Across discussions of agentic commerce, payment infrastructure, AI-native banking, regulatory oversight, explainability, digital sovereignty, and frontier AI development, Bussmann repeatedly argues that governance capabilities are emerging as the critical determinant of successful AI adoption.

Among the paper’s findings, the issue of accountability emerges as particularly significant. Traditional financial governance is founded upon the assumption that consequential decisions can ultimately be attributed to identifiable human actors. Agentic AI challenges this assumption by introducing systems capable of executing multiple interconnected decisions autonomously while legal, regulatory, and fiduciary responsibility remains firmly attached to institutions and individuals. This creates a structural accountability gap that existing governance frameworks were not designed to address.

The implications of this accountability gap extend throughout the financial system. Agentic commerce requires new approaches to consent, liability, and transaction governance. AI governance frameworks must evolve beyond vendor management towards continuous institutional oversight. Explainability must be supplemented by observability, auditability, and decision replayability. Infrastructure strategy increasingly depends upon governance capability as much as technological capability. Even frontier AI governance becomes a question of managing strategic dependency and institutional resilience rather than solely technological risk.

Taken together, these developments suggest a broader theoretical shift in how AI transformation should be understood. Much of the current literature treats artificial intelligence as a technology adoption challenge. The evidence presented in this paper suggests that agentic AI is more accurately understood as a governance challenge arising from the increasing delegation of decision-making authority to autonomous systems. In this context, organisational structures, accountability mechanisms, regulatory frameworks, and governance architectures become as important as the technologies themselves.

This perspective has important implications for both research and practice. For scholars, it suggests that future research should focus not only on AI capabilities but also on the institutional mechanisms required to govern those capabilities effectively. For practitioners, it implies that investments in models, data, and infrastructure must be accompanied by equivalent investments in accountability, observability, auditability, and operational control. The institutions most likely to capture long-term value from agentic AI will not necessarily be those deploying the most advanced systems, but those capable of governing those systems in ways that remain transparent, defensible, and trusted.

Ultimately, the defining challenge of the agentic era is not the emergence of machine autonomy itself. It is the creation of governance frameworks capable of ensuring that autonomy remains aligned with institutional accountability. As financial services move from AI-assisted decision-making to AI-executed decision-making, governance becomes the critical infrastructure upon which trust, compliance, and sustainable innovation depend. In the final analysis, the future of AI in finance will be shaped not by what autonomous systems can do, but by what institutions can govern.

10. References

Aggarwal, D. (2026) The Irrecoverable Institution: Why Replayability – Not Explainability – Is the Governance Standard for Agentic Banking. SSRN.

Aldridge, I. et al. (2026) Agentic Artificial Intelligence in Finance: A Comprehensive Survey. arXiv.

Axelsen, H., Licht, V. and Damsgaard, J. (2025) Agentic AI for Financial Crime Compliance. arXiv.

Bank for International Settlements (2025) Financial stability implications of artificial intelligence. Basel: BIS.

Barney, J. (1991) ‘Firm Resources and Sustained Competitive Advantage’, Journal of Management, 17(1), pp. 99–120.

Bengio, Y., Hinton, G., Russell, S. and others (2024) International Scientific Report on the Safety of Advanced AI. Montreal: International AI Safety Report Consortium.

Bommasani, R. et al. (2021) On the Opportunities and Risks of Foundation Models. Stanford Center for Research on Foundation Models.

Bussmann, O. (2026a–2026h) LinkedIn articles and posts on agentic AI, governance, digital assets, payments infrastructure and financial services transformation. Available via LinkedIn profile.

Brynjolfsson, E., Li, D. and Raymond, L.R. (2023) ‘Generative AI at Work’, National Bureau of Economic Research Working Paper Series.

Cockburn, I.M., Henderson, R. and Stern, S. (2019) ‘The Impact of Artificial Intelligence on Innovation’, in Agrawal, Gans and Goldfarb (eds.) The Economics of Artificial Intelligence. Chicago: University of Chicago Press.

Davenport, T.H. and Mittal, N. (2022) All-in on AI: How Smart Companies Win Big with Artificial Intelligence. Boston: Harvard Business Review Press

Doyle-Spare, M. (2026) The Agentic 3Cs Framework: A Reasoning-Layer Risk Governance Model for Agentic AI in Financial Services. SSRN.

Doshi-Velez, F. and Kim, B. (2017) Towards a Rigorous Science of Interpretable Machine Learning. arXiv.

European Banking Authority (2025) Rising application of AI in EU banking and payments sector. Paris: EBA.

European Union (2024) Regulation (EU) 2024/1689 Laying Down Harmonised Rules on Artificial Intelligence (AI Act). Official Journal of the European Union.

García-Llorente, C. and Olmeda, I. (2026) ‘Algorithmic governance in banking: a comparative analysis of risk-based and accountability-oriented oversight’, Journal of Banking Regulation, 27(19).

Hendrycks, D., Mazeika, M. and Woodside, T. (2023) ‘An Overview of Catastrophic AI Risks’, arXiv Working Paper.

Institute of International Finance (2025) IIF-EY Annual Survey Report on AI Use in Financial Services. Washington, DC: IIF.

Kurshan, E., Balch, T. and Byrd, D. (2025) The Agentic Regulator: Risks for AI in Finance and a Proposed Agent-Based Framework for Governance. arXiv.

McKinsey Global Institute (2025) The State of AI: Infrastructure, Compute and Competitive Advantage. New York: McKinsey.

Miller, T. (2019) ‘Explanation in Artificial Intelligence: Insights from the Social Sciences’, Artificial Intelligence, 267, pp. 1–38.

Park, J.S., O'Brien, J., Cai, C. et al. (2024) 'Generative Agents: Interactive Simulacra of Human Behavior', Proceedings of the ACM Symposium on User Interface Software and Technology, 37(4), pp. 1–22.

Rudin, C. (2019) ‘Stop Explaining Black Box Machine Learning Models for High Stakes Decisions and Use Interpretable Models Instead’, Nature Machine Intelligence, 1(5), pp. 206–215.

Russell, S. (2019) Human Compatible: Artificial Intelligence and the Problem of Control. New York: Viking.

Strauss, I., Moure, I., O’Reilly, T. and Rosenblat, S. (2025) Real-World Gaps in AI Governance Research. arXiv.

Teece, D.J. (2018) ‘Business Models and Dynamic Capabilities’, Long Range Planning, 51(1), pp. 40–49.

Contact

Reach out via email for inquiries.

Email

Subscribe to newsletter

info@grcadvisory.ch

© 2025. All rights reserved.