Singapore's financial watchdog has drawn a clear line in an emerging and largely unregulated frontier: the Monetary Authority of Singapore is moving to develop a formal "know your agent" protocol that would allow banks to verify whether an artificial intelligence agent acting on a customer's behalf has been genuinely and explicitly authorised to do so — including authorised to move that customer's money. The initiative comes in direct response to a pattern of behaviour that the MAS chief has described as alarming: users willingly handing over their bank login credentials to AI agents, a practice the regulator has publicly advised against.

The warning and the accompanying regulatory proposal arrive at a moment when the deployment of autonomous AI agents inside personal finance is accelerating faster than the governance frameworks designed to contain it. AI agents — software systems capable of taking independent, multi-step actions on a user's behalf — have moved rapidly from productivity novelties into consequential financial territory. They are being tested and, in some cases, already used to execute payments, manage subscriptions, initiate transfers, and interface with banking platforms. The question of whether those actions truly reflect a customer's informed will is no longer theoretical.

The MAS chief's warning was unambiguous. Some users are giving AI agents their login credentials to financial accounts, the regulator noted, and that practice carries serious risks. The core concern, as framed by the authority, is not simply data security in the conventional sense — although that remains a live threat — but something more structurally troubling: there is currently insufficient assurance that an AI agent acting within a banking environment is doing so with the customer's genuine, traceable authorisation. The gap between a user loosely enabling an agent and that agent executing a financial transaction is not a gap the existing regulatory or technical architecture is equipped to close.

The "know your agent" protocol, as described by the MAS, is envisioned as a mechanism that would sit alongside existing customer verification frameworks — echoing the logic of know-your-customer (KYC) standards that banks have long been required to maintain. Where KYC establishes the identity and legitimacy of a human customer, the proposed KYA framework would extend that same principle of verified authorisation to the non-human actors increasingly operating within those customers' financial lives. Banks, under such a protocol, would be empowered to interrogate not just who a customer is, but whether the entity initiating a transaction or accessing an account on their behalf has been formally sanctioned to do so — and to what degree.

The MAS's decision to frame its role as helping to develop this protocol, rather than simply mandating one from above, reflects a pragmatic recognition that the technical architecture underpinning AI agent authentication does not yet exist in any standardised form. This is not a situation where regulators can reach into an established playbook. Financial institutions, technology developers, and standards bodies will need to collaborate on the underlying infrastructure — the credentialing systems, the delegation frameworks, the audit trails — before a binding supervisory regime can meaningfully operate on top of it. Singapore's approach here is calibrated: signal the concern, set the direction, and convene the right actors to build the foundation.

The stakes extend well beyond Singapore's borders. As one of Asia's most influential financial hubs, decisions made by the MAS carry significant weight across the region, and its regulatory posture on AI in banking is watched closely by peer authorities from Hong Kong to Sydney. If the MAS is able to develop a credible and workable KYA standard, it may well become a template that other jurisdictions adopt or adapt. The alternative — a fragmented, jurisdiction-by-jurisdiction patchwork of AI agent governance — creates precisely the kind of regulatory arbitrage that sophisticated bad actors exploit.

There is also a consumer protection dimension that should not be understated. When a user shares their banking credentials with an AI agent, they are effectively dissolving the authenticated boundary between their financial identity and a third-party software system. If that agent is compromised, manipulated, or simply misconfigured, the financial consequences can be severe and may be difficult to reverse. Existing fraud liability frameworks were written with human actors in mind; they do not map cleanly onto scenarios in which an AI initiates an unauthorised transaction using legitimately obtained credentials. The MAS's concern about insufficient assurance is, at its core, a concern about who bears responsibility when things go wrong in that grey zone.

What This Means for Banks, Fintechs, and Users

For financial institutions operating in Singapore, the MAS's announcement is both a warning and a signal to begin preparing. Banks that have not yet developed policies governing AI agent access to customer accounts — covering credential delegation, session scope, transaction limits, and audit logging — should treat this as an urgent compliance planning priority. For fintech developers building AI-powered financial tools, the message is equally pointed: the regulatory expectation is moving toward formal authorisation architecture, not informal user consent buried in terms of service. And for consumers, the guidance could not be more direct — do not give AI agents your bank login. The infrastructure to do so safely does not yet exist, and until it does, the risk is yours to bear.

Written by the editorial team — independent journalism powered by Codego Press.