JPMorgan and Hyundai Card have completed a proof of concept in which an AI agent requested and executed a simulated foreign-exchange transaction without direct human intervention, CNBC reported. The bank describes it as the first autonomous bank-agent-to-client-agent test for corporate treasury management on a private blockchain. The scenario: Hyundai Card's AI agent filed a foreign-exchange request to maintain euro liquidity in its Kinexys blockchain deposit account — a routine corporate cash-management task, executed without a person pressing the button.
The more telling commentary is what JPMorgan says was hard. Wee Kee Toh, global head of Kinexys Insights, told CNBC that building the technology was relatively straightforward; establishing governance and internal controls was the real obstacle. "You are going to draw a box around the agent. The agent will never be able to do anything out of the box," Toh said. Kinexys's framework restricts agents to predefined transaction parameters and automatically rejects any request beyond authorized limits — Toh illustrated it with an agent authorized up to $5 million, which would see anything above that threshold blocked. The figure is an illustration of the control mechanism, not a disclosed limit from the simulation.
The partners plan a more ambitious phase in the first half of 2027: multiple AI agents negotiating with each other, reaching agreement directly, and then making payments. That shifts the test from "can an agent execute an instruction" to "what happens when autonomous agents interact in an ecosystem" — the transition Toh described as the core goal of expanding the trial. The two companies are also exploring whether to extend the collaboration to other businesses within Hyundai Motor Group, whose breadth JPMorgan cited as a reason for choosing Hyundai Card; no specific projects have been announced.
Notably absent is a commercialization timeline. JPMorgan has not said when, or whether, agent-executed treasury transactions become a live product, and the completed test was simulated rather than a real money movement. The framing matters: the bank is publishing a controls architecture, not launching a service, and accountability — who is responsible when an autonomous agent errs — remains the unsolved question Toh flagged alongside the risks of financial execution without direct human oversight.
The experiment lands as banks confront the practical edge of agentic AI. Consumer banks have been busy building walls — Schwab has said it blocks dozens of AI agents a week — while JPMorgan is testing the opposite premise: that with the right governance box, agents can be given authority inside it. Kinexys, JPMorgan's blockchain division, has been building toward programmable money for years, and an agent-to-agent payments layer is a natural extension of that thesis.
If the 2027 multi-agent phase works, the implications reach beyond treasury desks. Agent-to-agent negotiation and settlement on a permissioned blockchain is the concrete, regulated-industry version of the agentic payments protocols consumer platforms have been sketching. The constraint is not the blockchain, and it is not the model — it is whether a bank can define, in advance, every transaction an agent should never be allowed to make. That is the problem JPMorgan is now openly building in public.
Comments (0)
Log in to join the discussion
Log InNo comments yet