AI agents could help decide what to buy and when, but payment authorisation and settlement may need to remain rules-based. That separation is one proposed way to let automated agents act without asking financial systems to treat a probabilistic decision as a final instruction.
Watch Desk analysis
What happened
An article by John Orchard for OMFIF examines the practical, systemic and geopolitical questions raised by AI in payments. One proposal, from International Monetary Fund researchers Sonja Davidovic and Hervé Tourpe, separates payment activity into three layers: an AI agent interprets intent and coordinates an action; a middle layer supervises and mediates; and deterministic rules govern authorisation and final settlement.
The article also describes existing industry work on “Verifiable Intent”, a Mastercard and Google standard intended to record a purchaser’s specific authorisations, and “Know Your Agent” approaches to identifying agents. Orchard notes that agent-initiated purchases may still prompt merchants to seek human confirmation, which can undo some of the promised convenience.
Why it matters
A chatbot can give a bad answer and be corrected. A payment agent can turn a mistaken interpretation into a transaction, while disputes over consent, liability and fraud are still waiting at the door. Keeping final authorisation and settlement rule-based could offer a way to use AI for recommendations and coordination without handing it the keys to the payment system.
There is a larger concern too: if many agents make similar decisions at machine speed, their actions could amplify market volatility. The article points to familiar financial risks, but argues that speed and scale could make them harder to manage.
Our read
The useful distinction is not “AI or no AI”. It is deciding which parts of a payment process can tolerate probabilistic judgement and which require clear authority, accountability and a final answer. That is a better starting point than asking an agent to shop around and hoping the checkout has a spare human attached.
The proposals are ideas discussed in the article, not a new regulatory rule. But they give regulators and payment providers a concrete design question: where should an agent’s freedom stop?
What to watch
- Whether Verifiable Intent and Know Your Agent approaches become widely adopted.
- How providers handle consent, disputes and liability when an agent initiates a transaction.
- Whether regulators address the risk of many agents acting in concert at machine speed.
Discussion spark: Should AI agents be allowed to choose and initiate payments if a separate rules-based layer still controls authorisation and settlement, or should a person approve every transaction?
Sources and evidence
- Source update (15 October 2026, 14:04 UTC)
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.