Artificial intelligence (AI) has become one of the defining technologies shaping the future of financial services. Banks are embedding AI into customer service, credit assessment, payments, fraud detection, wealth management, software development and internal operations at an unprecedented pace. The rapid emergence of generative AI (Gen AI) and increasingly autonomous AI agents has further accelerated adoption, bringing forward productivity gains that many institutions had previously expected only later in the decade.

The same technologies, however, are simultaneously transforming the threat landscape. Cybercriminals are using AI to automate reconnaissance, personalise phishing attacks, generate convincing deepfakes and industrialise fraud at a scale that was previously impossible. Financial institutions therefore find themselves competing in an environment where both defenders and attackers increasingly rely on the same technological capabilities.

This convergence is changing the nature of cybersecurity itself. Security can no longer be treated as a specialised technology function operating alongside the business. Instead, it has become intertwined with AI governance, enterprise risk management, data governance, operational resilience and board oversight. Decisions about AI deployment increasingly have implications not only for efficiency and customer experience but also for regulatory compliance, institutional trust and financial stability.

Regulators worldwide have largely moved beyond asking whether financial institutions should adopt AI. Instead, the emphasis has shifted towards how institutions can deploy AI safely, transparently and accountably while maintaining public confidence in financial services. In this environment, digital trust increasingly becomes the outcome of disciplined governance rather than technological sophistication alone.

The new generation of intelligent threats

AI is altering the economics of cybercrime. Many attacks that previously required significant technical expertise, manual preparation and considerable time can now be executed more quickly, at greater scale and with far higher levels of personalisation. The challenge facing financial institutions is therefore no longer confined to defending against isolated attacks but responding to an increasingly adaptive threat ecosystem.

Industry evidence increasingly supports this shift. The World Economic Forum (WEF), International Monetary Fund (IMF) and leading cyber-security researchers have all warned that Gen AI is reducing the cost of fraud, increasing the speed of attack and enabling highly personalised social engineering campaigns. The result is that institutions now face adversaries capable of operating with levels of automation previously available only to large organisations.

On AI-driven fraud, Raad Khraishi, Head of AI Research and Development, NatWest Group, observed that AI is dramatically reducing the cost of personalisation for criminals. Publicly available information, breached data and AI-powered research tools allow attackers to construct highly targeted social engineering campaigns automatically. Combined with advances in voice synthesis, image generation and deepfake technologies, fraud can now exploit both digital identities and human trust with far greater sophistication than before.

The significance is not simply that attacks are becoming more sophisticated. Rather, AI is collapsing the traditional asymmetry between attack and defence. Defensive controls are still deployed through structured governance, procurement and regulatory oversight, while attackers can iterate continuously at machine speed. The challenge for financial institutions is therefore organisational as much as technological.

The implications extend beyond individual scams. Rather than relying on single exploits, attackers increasingly combine multiple vulnerabilities across identity systems, software supply chains and human behaviour. This evolution makes fraud detection considerably more difficult because institutions must identify coordinated attack patterns rather than isolated malicious events.

Deepfake technologies illustrate how rapidly this threat environment is evolving. Fraud no longer begins with the transaction itself. Instead, it often originates at the identity layer, where convincing synthetic voices, videos and documents are used to manipulate customers, employees or counterparties before financial transactions even occur. These attacks exploit confidence in familiar identities rather than weaknesses in banking systems alone.

AI is creating opportunities for defenders as well as attackers. Behavioural analytics, continuous authentication and AI-assisted monitoring can identify anomalies that traditional rules-based systems frequently miss. David Ng, Group Head of IT Security, Maybank pointed to behavioural biometrics as one area likely to become increasingly important. Rather than relying solely on passwords or physical biometrics, future authentication may increasingly consider behavioural characteristics such as typing rhythm, mouse movements and interaction patterns as additional indicators of identity.

Raad Khraishi, Head of AI Research and Development, NatWest Group

David Ng, Group Head of IT Security, Maybank

The wider cybersecurity ecosystem is already responding through greater investment in behavioural analytics, digital identity, AI-enabled monitoring and cross-sector collaboration. Rather than representing isolated technological threats, these developments reinforce the growing need for integrated cyber resilience, stronger identity assurance and closer collaboration across the financial ecosystem.

Building enterprise-wide digital trust

As AI becomes embedded across banking operations, governance is emerging as the primary determinant of whether institutions can deploy intelligent systems at scale without compromising trust. The discussion has therefore shifted beyond cybersecurity in its traditional sense towards the broader question of how banks govern AI throughout its lifecycle. This requires institutions to manage not only technology risks but also accountability, explainability, operational resilience and regulatory compliance within a single enterprise-wide framework.

The greatest risk today is not AI itself but the manner in which it is being introduced into organisations. Dee Lawler, Chief Technology Officer, Global Financial Services and Insurance Sector, International Markets, Atos, described AI as entering enterprise environments “by stealth”, through software upgrades, embedded copilots and vendor products that incorporate AI functionality without always being subject to the same governance disciplines applied to conventional technology deployments. In regulated industries such as banking, she argued that institutions would never deploy production systems without testing, validation, observability and security controls, yet AI capabilities are increasingly appearing across enterprise technology estates without equivalent scrutiny.

This observation reflects one of the industry’s emerging governance challenges. Traditional technology governance was designed around deterministic systems whose outputs could be predicted and validated. Gen AI fundamentally changes that assumption. Models evolve, generate probabilistic responses and increasingly interact autonomously with other systems through AI agents. Governance therefore becomes a continuous operational discipline rather than a project approval process conducted before deployment.

This represents a significant departure from traditional technology governance. Conventional governance assumes software behaves predictably once tested and deployed. AI systems, particularly foundation models and autonomous agents, require continuous supervision because their behaviour depends not only on software but also on evolving data, prompts, interactions and operating context. Governance therefore becomes an operational capability rather than a compliance checkpoint.

Lawler argued that organisations need to move beyond conventional governance mechanisms such as periodic steering committees and manual approvals towards what she described as “governance as code”, where automated supervisory controls continuously monitor AI behaviour, constrain agent activity and verify compliance in real time. Rather than governing individual applications, institutions increasingly need governance architectures capable of supervising machine decision-making across interconnected AI ecosystems.

Enterprise visibility is becoming increasingly important. As AI capabilities become embedded across software platforms, many organisations no longer have a complete inventory of where AI is operating or what decisions it influences. This creates governance blind spots that extend beyond cybersecurity into compliance, operational resilience and third-party risk management.

For Randall Duran, Senior Lecturer at Singapore Management University, the challenge is compounded by the pace of technological change. While institutions are attempting to define their long-term AI strategies, they cannot reliably predict what banking will look like by 2030. Rather than attempting to optimise for a single future, he argued that banks should build adaptable organisational capabilities that allow them to respond to multiple possible technology trajectories. Governance therefore becomes less about constraining innovation than enabling institutions to innovate safely despite uncertainty.

Dee Lawler, Chief Technology Officer, Global Financial Services and Insurance Sector, International Markets, Atos

Randall Duran, Senior Lecturer at Singapore Management University

There is a growing challenge to the assumption that every operational problem requires an AI solution. Lawler cautioned against deploying AI simply because it is available, arguing that some longstanding banking processes should instead be redesigned or eliminated altogether. Automation, robotic process automation, machine learning or business process re-engineering may in many cases deliver better outcomes than introducing Gen AI into unsuitable workflows. The objective should therefore be optimisation based on business value rather than technology novelty.

This pragmatic approach was reinforced by Balaji Rajagopalan, Chief Technology Officer, State Bank of India, who consistently framed AI adoption around measurable operational outcomes rather than technical sophistication. He cited examples where AI reduced document scrutiny processes for trade finance and guarantees from several hours to minutes by identifying exceptions, anomalies and rule breaches automatically. For him, value creation begins with removing repetitive work, improving productivity and allowing employees to focus on higher-value activities rather than replacing human judgement altogether. Customer-facing AI should similarly improve contextual service by providing continuity across digital channels while supporting faster decisions and more personalised interactions.

These operational discussions increasingly intersect with regulatory expectations. Financial regulators outlined broadly similar principles despite differences in national approaches. Rather than prescribing detailed technical requirements, each emphasised governance, accountability and proportionality as the foundations of responsible AI adoption.

Alan Lim, Director and Head, Digital and Infrastructure, AI Office, Monetary Authority of Singapore, argued that governance should not be viewed as an obstacle to innovation. Instead, clear guardrails provide institutions with the confidence to innovate more rapidly. He noted that while many governance disciplines, including model risk management and data governance, already existed before Gen AI, new challenges have emerged because large language models produce non-deterministic outputs. Unlike traditional rules-based systems, identical prompts may generate different responses, making consistency, validation and explainability significantly more complex within highly regulated financial environments.

Balaji Rajagopalan, Chief Technology Officer, State Bank of India

Alan Lim, Director and Head, Digital and Infrastructure, AI Office, Monetary Authority of Singapore

A similar balance between innovation and prudence emerged from Japan. Jutaro Kaneko, Deputy Commissioner for International Affairs, Financial Services Agency, Japan, explained that the regulator has deliberately refrained from introducing AI-specific financial regulation at this stage, preferring continued dialogue with industry while identifying emerging risks that may require future intervention. At the same time, the agency has sought to encourage adoption by supporting common AI capabilities for regional banks, reflecting the view that governance should enable innovation rather than discourage it.

From Indonesia, Muhammad Zikri, Director of Data Innovation and Digitalisation Department, Bank of Indonesia, described a framework centred on proportional risk management, lifecycle governance, accountability and human oversight. He argued that trustworthy AI depends fundamentally on trustworthy data, observing that explainability, fairness and model reliability cannot be achieved without disciplined data governance throughout the AI lifecycle. His emphasis reflected a broader regulatory trend in which AI governance increasingly begins with data governance rather than algorithms themselves.

Jutaro Kaneko, Deputy Commissioner for International Affairs, Financial Services Agency, Japan

Muhammad Zikri, Director of Data Innovation and Digitalisation Department, Bank of Indonesia

Taken together, these perspectives suggest that enterprise-wide digital trust cannot be achieved by strengthening cybersecurity alone. Instead, it requires institutions to integrate AI governance, data governance, operational resilience, cyber risk management and executive accountability into a unified operating model. As AI becomes embedded across banking, governance increasingly shifts from being a compliance function to becoming an enterprise capability that determines whether institutions can deploy intelligent technologies safely, responsibly and at scale.

Trustworthy AI as the next competitive advantage

For much of the past two decades, banks competed primarily on digital capability. Institutions invested heavily in mobile banking, cloud infrastructure, application programming interfaces (APIs), automation and data platforms to improve customer experience and operational efficiency. As AI becomes embedded across virtually every banking function, however, competitive differentiation is beginning to shift. The question is no longer which institution deploys the most AI, but which can demonstrate that its AI remains secure, explainable, resilient and trustworthy throughout its lifecycle.

This represents a significant change in the industry’s strategic priorities. During the first wave of digital transformation, technology investment was largely measured through adoption rates, speed of deployment and customer convenience. AI introduces an additional dimension. Institutions must now demonstrate that automated decisions remain consistent with regulatory expectations, organisational risk appetite and customer interests even as models evolve, learn and interact with increasingly autonomous systems. Trust therefore becomes an operational capability rather than simply a customer perception.

This distinction will become increasingly important as AI moves beyond productivity tools into decision-making processes. The early phase of AI adoption has focused largely on copilots, code generation, customer service assistants and workflow automation. These applications generally operate with relatively low levels of autonomy and retain significant human oversight. The next phase is expected to involve AI agents capable of initiating transactions, coordinating multiple business processes and interacting directly with customers and external systems. This progression substantially increases both the value and the risk associated with AI deployment.

Duran observed that banks should therefore avoid becoming overly focused on predicting a single technological destination. Instead, they should build organisational capabilities that allow them to adapt as AI continues to evolve. Whether future banking involves AI agents acting on behalf of customers, autonomous payment initiation or entirely new forms of customer interaction remains uncertain. What is more predictable is that institutions with stronger governance, better data quality and more adaptable operating models will be better positioned to respond as these developments materialise.

Rajagopalan similarly argued that AI investments should continue to be evaluated through measurable business outcomes rather than technological novelty. For banks, productivity gains, faster turnaround times, improved customer experiences, stronger fraud detection and better resource utilisation remain the primary indicators of success. AI should therefore be deployed where it demonstrably improves business performance rather than because it represents the latest technological trend. This emphasis on business value reflects an important shift away from technology-led transformation towards outcome-led transformation.

Trust extends beyond AI models themselves to the quality of the data supporting them. Data governance is therefore the foundation of responsible AI. This is particularly significant because many AI failures originate not from model architecture but from incomplete, biased or poorly governed data. Financial institutions have long invested in data quality to support regulatory reporting and risk management. AI raises the stakes further because unreliable data can directly affect automated recommendations, customer interactions and operational decisions.

Ironically, institutions may find that AI itself becomes one of the least differentiated aspects of their operating model. Foundation models are becoming increasingly accessible through cloud platforms and commercial software providers. What remains difficult to replicate is the institutional capability to govern those models consistently across thousands of employees, business processes and customer interactions. Competitive advantage may therefore shift away from algorithms themselves towards organisational discipline.

Regulators increasingly share this perspective. Lim noted that many of the governance disciplines required for Gen AI, including model risk management, data governance and operational controls, already existed before the arrival of large language models. Rather than replacing these disciplines, AI amplifies their importance. Gen AI introduces new characteristics such as non-deterministic outputs, but institutions that have established mature governance foundations are generally better positioned to manage these additional risks.

Zikri’s emphasis that “trusted AI requires trusted data” reinforces this broader principle. Explainability, fairness and reliability cannot be achieved if institutions cannot demonstrate the provenance, integrity and governance of the data underpinning AI systems. As banks deploy increasingly sophisticated models, data governance is therefore becoming inseparable from AI governance itself.

Another recurring theme was the continued importance of human judgement. Despite rapid advances in AI capability, neither regulators nor industry practitioners suggested that human oversight should disappear from banking. Instead, the discussion focused on identifying where human intervention remains essential and where greater automation may become appropriate. This represents a more nuanced position than the often-polarised debate surrounding AI replacing human decision-makers.

Kaneko raised an important consideration in this context by questioning how effective human oversight remains when people themselves are susceptible to automation bias and cognitive bias. If users naturally place excessive confidence in AI-generated recommendations, simply requiring a human to approve an AI decision may not provide the safeguard many institutions assume it does. This observation shifts attention from the presence of humans in governance processes towards the quality and independence of their judgement.

Similarly, Khraishi suggested that human review is likely to remain necessary for many higher-risk banking activities, including credit decision-making in jurisdictions where regulatory expectations require meaningful human involvement. At the same time, lower-risk internal applications are increasingly likely to become fully automated as institutions gain greater confidence in AI reliability and governance. The issue therefore is not whether humans remain involved, but how organisations determine the appropriate balance between automation and oversight across different use cases.

This balance increasingly influences customer confidence as well. Customers rarely evaluate AI models directly. Instead, they judge the quality of financial institutions through the consistency, fairness and reliability of the services they receive. A highly accurate model that occasionally produces unexplained or inconsistent outcomes may ultimately undermine confidence more than a slightly less sophisticated system operating within well-understood governance boundaries. Trust therefore becomes observable through predictable customer outcomes rather than technological complexity.

Looking ahead, this emphasis on trust is likely to influence investment priorities across the banking industry. While institutions will continue investing in Gen AI, agentic AI and intelligent automation, they are also likely to devote increasing resources to model validation, continuous assurance, behavioural monitoring, identity protection, cyber resilience and quantum-resistant cryptography. These investments may not always produce visible customer features, but they are becoming essential infrastructure for maintaining confidence in increasingly autonomous financial systems.

Trustworthy AI may become a competitive capability rather than merely a regulatory requirement. Institutions that can demonstrate disciplined governance, transparent decision-making and resilient AI operations are likely to enjoy greater confidence from regulators, customers and counterparties alike. As AI becomes ubiquitous across financial services, trust may prove harder to replicate than the technology itself, making it one of the industry’s most enduring sources of competitive advantage.

Securing confidence in an autonomous financial system

The financial industry has experienced several waves of technological transformation over the past three decades, from internet banking and mobile payments to cloud computing and open banking. Each has introduced new efficiencies while exposing new vulnerabilities. AI differs in one important respect: it is simultaneously changing how financial institutions operate, how customers interact with them and how attackers seek to exploit them. The same technology that enables greater productivity also lowers the barriers to increasingly sophisticated cybercrime, making trust the central strategic challenge of the next phase of banking.

This convergence requires a broader conception of resilience than the industry has traditionally adopted. Historically, cybersecurity focused on protecting networks, applications and infrastructure from external attack. Operational resilience expanded that scope to include business continuity, third-party risk and critical operations. AI now adds another layer, requiring institutions to govern intelligent systems whose behaviour may evolve over time, interact autonomously with other systems and influence decisions across virtually every business function. As a result, digital trust is no longer the responsibility of a single technology or security function but an enterprise-wide capability spanning governance, risk management, technology, operations and leadership.

Whether examining AI-enabled fraud, enterprise governance or board oversight, participants repeatedly returned to the need for integration rather than specialisation. Cybersecurity cannot operate independently of AI governance. AI governance cannot be separated from data governance. Operational resilience cannot be maintained without understanding how autonomous systems influence critical business processes. These disciplines increasingly reinforce one another, requiring institutions to develop operating models capable of managing technology as an interconnected ecosystem rather than as a collection of individual projects.

The role of leadership is also changing. Boards are no longer expected merely to approve technology investments; they must increasingly understand how AI affects institutional risk, customer trust and strategic resilience. Similarly, chief information security officers are evolving from technical specialists into enterprise risk leaders responsible for communicating cyber risk in business terms, prioritising investments and demonstrating how security enables sustainable innovation. Ng argued that cybersecurity should be viewed not simply as a cost but as a means of preserving a bank’s licence to operate, with stakeholder engagement and measurable delivery being essential to securing long-term organisational support.

Regulators, meanwhile, appear to be converging around broadly consistent principles despite differences in national implementation. Across Singapore, Japan and Indonesia, the emphasis was not on restricting AI adoption but on ensuring that innovation remains proportionate to risk, supported by robust governance and subject to meaningful accountability. Despite differences in regulatory philosophy, the three authorities converged on several common principles. None advocated slowing AI adoption. Instead, each emphasised proportional governance, strong data management, human accountability and continuous oversight as the foundations for responsible deployment.

While regulatory frameworks will undoubtedly continue to evolve, the discussions suggested that many of the underlying expectations, sound data governance, model validation, transparency, human accountability and operational resilience, are already becoming established supervisory norms rather than emerging concepts.

Looking ahead, the competitive landscape is also likely to evolve. AI capabilities themselves are becoming increasingly accessible through foundation models, cloud platforms and commercial software. Over time, many technical capabilities may become widely available across the industry, reducing their ability to differentiate one institution from another. What is likely to prove more difficult to replicate is the organisational capability to deploy AI consistently, securely and responsibly at scale. Institutions that can demonstrate disciplined governance, reliable operations and sustained customer confidence may therefore enjoy advantages that are considerably more durable than early technological adoption alone.

The next generation of banking technology investment is unlikely to be judged by how much AI institutions deploy, but by how confidently they can demonstrate that AI remains secure, explainable and resilient under real operating conditions. Banks will continue expanding AI across customer engagement, risk management, software engineering and operational automation, but these initiatives will increasingly be accompanied by investment in continuous assurance, behavioural analytics, identity protection, model monitoring, supply-chain security and quantum-resistant cryptography. Rather than viewing these as compliance costs, institutions may increasingly regard them as foundational infrastructure supporting the safe deployment of intelligent financial services.

Every major technology transformation in banking has eventually become mainstream. ATMs, internet banking, mobile banking, cloud computing and APIs all followed that trajectory. AI is unlikely to be different. Models will become more powerful, less expensive and increasingly accessible to every institution.

What will remain scarce is trust. Customers, regulators and counterparties will continue to judge financial institutions not by the sophistication of their algorithms but by the consistency, transparency and resilience of the decisions those algorithms produce. In that environment, digital trust becomes more than a security objective or regulatory requirement. It becomes an institutional capability that enables innovation at scale, making it one of the defining competitive advantages of banking in the AI economy.