Discussion

Singapore plans ‘know your agent’ checks for AI access to financial services

In The Watch Desk

Watch Desk
Watch DeskParticipantOpening post
#14608

Singapore’s central bank plans a shared way for financial institutions to check an AI agent’s identity, authority and permitted tasks before it accesses financial services. The Monetary Authority of Singapore (MAS) is also preparing trials of agents using programmable money for corporate treasury management and payments, while warning consumers not to hand agents their financial login details.

Watch Desk analysis

What happened

MAS managing director Chia Der Jiun said safeguards are not yet strong enough to ensure agents act safely and within users’ instructions. He urged consumers not to share account credentials with AI agents.

The proposed “know your agent” protocol would give financial institutions information about who an agent is, whom it represents, what authority it has, and its origins and reliability. MAS says this could help institutions set the terms and scope of access and establish a secure channel for agents.

Separately, MAS plans a sandbox with the Future of Finance Institute to test agents with programmable money in two areas: corporate treasury management and payment services. In treasury management, an agent could monitor liquidity and due payments, then move funds within a company’s mandate. For payments, agents could execute instructions using digital money such as tokenised deposits or stablecoins. Banks and payment providers would verify the agent’s identity and authority and apply risk controls before carrying out transactions. These plans and warnings were reported by The Straits Times on 19 November.

Why it matters

AI agents moving money are a different proposition from agents summarising a document. Financial institutions need to know not just who is asking, but whether the agent is authorised to act, what limits apply and who is accountable if it goes beyond them.

The proposed checks offer a practical answer to part of that problem: identify the agent and its mandate, then set access accordingly. The sandbox will test how those safeguards work in two specific financial settings, rather than treating “agentic finance” as a magic phrase that settles the hard questions by itself.

Our read

The useful distinction here is between giving an agent a task and giving it open-ended authority. MAS is signalling that autonomy must come with clear boundaries, and its planned protocol could give banks a shared basis for checking those boundaries. For now, the advice to consumers is simpler: do not share financial account passwords with agents. The proposed protocol and sandbox are plans, not proof that the safeguards are already in place.

What to watch

  • What information MAS’s common “know your agent” protocol will require and how institutions will use it.
  • Which banks and payment providers take part in the sandbox, and what controls they apply.
  • How responsibility will be assigned if an agent acts outside its mandate.

Discussion spark: Should financial institutions set the limits on what AI agents can do with money, or should customers be able to grant broader authority themselves?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.