IMF sets out three-layer framework for agentic AI in payments
International Monetary Fund (IMF) HeadquartersOrhan Cam / Shutterstock.com in Washington D.C.

The International Monetary Fund published a note, dated 22 April, examining how agentic artificial intelligence could reshape payment systems, setting out a three-layer framework that keeps adaptive AI reasoning separate from the deterministic processes that authorise and settle transactions.

The note, IMF Notes No 2026/004 "How Agentic AI Will Reshape Payments", by Sonja Davidovic and Hervé Tourpe, describes agentic systems as able to interpret objectives, break them into tasks and interact with digital services with limited human input. In payments, this could shift transactions from human-initiated instructions to agent-mediated decisions. The note examines the effects on authorisation, liquidity, settlement, compliance and resilience, and identifies a key tension between probabilistic AI behaviour and the deterministic requirements of payment infrastructures.

INTENT, AUTHORISATION, SETTLEMENT

In the first layer, intent and orchestration, agentic systems translate high-level user objectives into structured, machine-readable instructions without executing transactions themselves. The second layer, control and authorisation, enforces deterministic policy rules that decide whether an action proposed by an agent may proceed. The third layer comprises traditional settlement infrastructure, such as real-time gross settlement systems, instant payment networks and card clearing engines, which executes payments with legal finality. The approach concentrates probabilistic reasoning upstream while preserving deterministic authorisation and settlement.

The note reviews use cases and risks including traceability, opacity, systemic effects, cybersecurity and legal uncertainty. It also examines emerging technical standards for connecting agents to payment rails, such as the Universal Commerce Protocol and the Agent Payments Protocol, and says payment networks and technology providers will need trusted identity frameworks and interoperable standards for Know-Your-Agent verification.

LEGAL QUESTIONS

On liability, the note observes that most payment regimes require a payment order to be traceable to an authorised instruction from an account holder, whereas individual agent-initiated transactions may not map to explicit instructions. It calls for legally workable concepts of authorisation based on verifiable mandates, scope limits and auditability. The IMF concludes that outcomes will depend on institutional design and governance as much as on technology.