When automated systems inside a bank recommend a credit decision, flag a liquidity stress signal or initiate a transaction without direct human instruction, the question of where institutional judgement sits and who remains accountable for the outcome has become harder to answer. The governance frameworks designed for slower, human-led decisions were not built for a system in which models act on their own judgement. That concern shaped a closed-door CXO roundtable at The Asian Banker Summit 2026 in Kuala Lumpur, supported by Ant Digital Technologies, which brought together bank chief executives, payment leaders and technology executives from institutions including Affin Bank, Asian Institute of Chartered Bankers, Bank Islam, Cradle Fund, Deutsche Bank, Hong Leong Bank, Hong Leong Islamic Bank, MBSB Bank, MUFG Malaysia, NETS, OCBC Malaysia, PayNet, Prudential Assurance Malaysia, Standard Chartered Saadiq Malaysia and UOB Malaysia. Patrick McHenry, former Chair of the United States (US) House Financial Services Committee, joined as a principal speaker, together with Emmanuel Daniel, Founder of TAB Global and Gordian Gaeta of The Asian Banker Summit Advisory Council The discussion ranged across AI-native banking, the collapse of Silicon Valley Bank (SVB), cross-border payments, stablecoins, programmable money and agent-to-agent commerce. The strands shared a common analytical concern. The faster banks deploy AI for decisions and transactions, the more the governance, control and operating-model assumptions inside institutions become a key limitation, not the technology itself. AI-native banking is an operating-model question Leo Li, President of International Business at Ant Digital Technologies, said AI-native banking requires changes to organisational design, decision-making, product architecture and operating models, not a migration from conventional IT to AI infrastructure. He compared the shift to the electric vehicle. Replacing a petrol engine with a battery does not create a new category of vehicle until the entire system is redesigned around what the technology makes possible. Li set out three operating conditions for AI in regulated finance. Decisions must be explainable, the process from data to outcome must be auditable and the path from input to result must be traceable. AI systems cannot bear responsibility, but humans and institutions still do. Accountability for what AI decides must be assigned to specific humans and processes. Garry Sien, Chief Solutions Officer for International Business at Ant Digital Technologies, said financial firms’ objectives when applying AI had moved quickly. Productivity, the central use case in 2025, is now a baseline expectation, and board-level conversations have shifted to where AI can improve revenue, coverage and customer engagement. Banking remains fundamentally a risk management business, and the institutions that manage AI risk better than their peers will be the ones that win. How intelligent an AI system can be for a bank, Sien said, is determined by how much access the institution extends to it across core data, processes and organisational context. That access decision exposes whether the first, second and third lines of defence are configured for an environment in which AI participates in decisions, and whether the institution's risk appetite can accommodate that participation within its existing control framework. Sien also noted that the AI transformation is running on a three-to-five-year cycle compared with the fifteen years of the mobile internet, leaving banks materially less time to adjust. One participant placed AI risk on the same footing as capital and liquidity risk, arguing that a technology evolving this fast requires institutional treatment of equivalent weight. Daniel pointed to the treasury function as one area where banks could already redesign internal workflows without waiting for new regulation. AI can make the relationship between treasury and its internal counterparties in asset-liability management and product funding more dynamic and immediately responsive. With full information visibility, banks can surface internal liquidity that would otherwise sit idle. When risk moves faster than the institution can respond The collapse of SVB in 2023 anchored the discussion of how fast institutional systems can respond when deposits move at digital speed. McHenry, who chaired the US House Financial Services Committee at the time of SVB’s collapse, framed the failure as a central bank speed problem before it was a balance sheet one. A central bank’s core function in moments of stress is to lend against quality assets, with particular attention to duration risk. Supervisors did not require the bank to address a publicly disclosed balance sheet gap, and SVB experienced a larger single-day deposit outflow than any institution in American banking history, despite ranking outside the top fifty US banks. Central banks, McHenry said, must provide liquidity at the speed of the internet, with no human intermediary standing between the central bank and a bank in a moment of crisis. One participant pushed back on McHenry's framing of the failure as primarily a central bank speed problem. The underlying problem, they argued, was a more conventional liability management failure. SVB had taken in deposit growth from a concentrated customer pool of technology companies and start-up founders, and channelled it into long-duration Treasuries. Better scenario planning around deposit concentration would have surfaced the run risk before it materialised. Another participant raised AI as a possible response. Banks are now asking how to use AI to get information faster, make decisions faster and forecast scenarios as deposits move at digital speed. Liability concentration cannot be assessed only by deposit volume. The composition of the deposit base matters too, and judgements about composition are exactly what AI-assisted analysis is now being deployed to support. A permissive US posture, and a more cautious Asian one McHenry said the current US regulatory environment is unusually permissive by historical standards. Financial regulators have been directed to be innovation-forward, with AI framed as both an economic competitiveness imperative and a national security priority. Institutions are expected to deploy first, with the regulatory framework following. The durability of that posture across the political cycle is a key uncertainty. McHenry pointed to bipartisan work underway on Capitol Hill intended to establish a consistent basis for AI deployment that can survive an electoral cycle, rather than leaving the current posture vulnerable to a policy reversal under a future administration. Sien said his conversations with regulators across the ASEAN region suggested a more cautious environment, though not obstructive. Security and governance came up repeatedly as core requirements. Regulators are asking whether institutions have updated their control frameworks to reflect the specific risks AI introduces, including hallucination risk in decisions with material consequences. Sandbox mechanisms in Singapore and Hong Kong allow limited production deployment under supervisory observation, giving regulators visibility into how institutions assess and contain risks such as hallucination as those risks surface in live use rather than on paper. One participant argued the scale of US capital expenditure on AI infrastructure now makes a policy reversal economically unlikely. Of greater concern was whether Asian regulators were keeping pace with the speed at which banks were deploying AI. If they were not, banks could find themselves deploying ahead of any framework that protects them. Another participant raised a related concern. Frontier AI systems may mature faster than cyber defences inside regulated institutions can adapt, exposing banks before the productivity and revenue gains have been captured. He compared the risk to Y2K and placed it alongside quantum computing as a category of operational threat the industry has not yet sized. Payments governance becomes AI governance On tokenised forms of money, McHenry said the US policy direction is away from a retail central bank digital currency (CBDC). The more significant development is the Clarity Act, which establishes equivalent regulatory treatment for bank-issued stablecoins, fintech-issued stablecoins and tokenised bank deposits. Banks are not capital-disadvantaged by offering stablecoins relative to fintech issuers. Daniel added that regulators in the Asian region are working through a related cluster of questions covering programmable money and digital assets. The accountability question becomes more complex when AI begins to act across payments infrastructure, transacting on behalf of users without direct human instruction at each step. Daniel described the emergence of agentic commerce, in which AI agents transact with each other at values that may be fractions of a cent, well below the economics of conventional payment pricing. The framework needed for that environment, a participant said, is know your agent, or KYA, a parallel discipline to know-your-customer and know-your-business protocols. The issue is authority: what rights an agent has been granted, what protections exist for the consumer and the institution processing the transaction and how those parameters are enforced at transaction speed. One example is a consumer instructing an agent to purchase a weekly grocery basket, with the agent selecting autonomously against defined criteria. What banks have not yet resolved Daniel framed the central question of the roundtable as one of locating decision authority. Where is the decision being made, who owns it, what judgement is being applied? A bank can deploy multiple AI tools, he argued, and still leave the institution's underlying decision-making unchanged. The configuration of the three lines of defence for an environment in which AI participates in decisions, the pace at which regional regulators are building their frameworks relative to deployment, and the economics of payment infrastructure under agent-to-agent commerce all remained unresolved across the discussion. The strategic question for banks is whether the institution is being redesigned around AI fast enough to keep accountability inside it as AI participates in decisions, monitors liquidity and transacts on behalf of customers
