JPMorgan and Hyundai Card complete simulated currency trade using AI agents
The bank says setting limits and assigning accountability are harder than building the technology. More complex, multi-agent testing is targeted for early 2027.
Loading page…
The bank says setting limits and assigning accountability are harder than building the technology. More complex, multi-agent testing is targeted for early 2027.
Listen to this story
JPMorgan and Hyundai Card’s trial shows that agentic finance depends less on proving an agent can submit a transaction than on defining enforceable limits and accountability. The completed proof-of-concept simulated a euro-liquidity foreign-exchange request for Hyundai Card’s Kinexys deposit account; it was not a live commercial service. The partners aim to test agent-to-agent negotiation and payments in the first half of 2027, but JPMorgan has announced no launch timeline.
Kinexys’s framework restricts agents to predefined transaction parameters and automatically rejects requests outside those limits.
Wee Kee Toh’s $5 million threshold was an illustration, not the simulation’s disclosed amount or a commercial product limit.
The planned 2027 phase will investigate agents negotiating agreements and making payments; the completed test did not demonstrate those capabilities.
AI agents carried out a simulated foreign-exchange transaction without direct human intervention in a completed JPMorgan and Hyundai Card trial, CNBC reported. But executing the transaction was not the hardest part: JPMorgan says establishing controls over what agents can do—and who is accountable—is the bigger obstacle to putting the technology into commercial use.
The proof-of-concept centered on a specific corporate cash-management task. A Hyundai Card AI agent requested a foreign-exchange transaction to maintain euro liquidity in its Kinexys blockchain deposit account. In plain terms, the request was intended to keep euros available in that account, rather than simply produce advice for a person to act on.
JPMorgan described the exercise as the first autonomous bank-agent-to-client-agent test for corporate treasury management on a private blockchain. That distinction is the bank’s claim. The completed transaction was simulated, making this a proof-of-concept rather than the start of a commercial service.
Wee Kee Toh, global head of Kinexys Insights at Kinexys by J.P. Morgan, told CNBC that building the technology was relatively straightforward. Establishing governance and internal controls proved harder. He pointed to accountability and the risks of letting agents execute financial transactions without direct human oversight.
Kinexys developed a framework that restricts agents to predefined transaction parameters and automatically rejects requests outside authorized limits. The approach gives the software authority to act within boundaries set in advance, rather than permission to execute any transaction it requests.
Toh illustrated the boundary with an agent authorized to transact up to $5 million: a request above that threshold would be blocked. The figure was an example of how the control would work, not a disclosed value for the Hyundai Card simulation or an announced limit for a commercial product.
You are going to draw a box around the agent. The agent will never be able to do anything out of the box.
Wee Kee Toh, global head of Kinexys Insights, speaking to CNBC
The companies are targeting the first half of 2027 for a more complex phase involving multiple AI agents, according to Toh. The intended step goes beyond executing a request: they want to explore agents negotiating, reaching agreement directly with one another and then making payments.
That is a planned expansion of the test’s scope, not a capability demonstrated by the completed currency simulation. Toh described the next step as examining how autonomous agents interact in an ecosystem, moving the focus from carrying out a transaction to how agents arrive at one together.
The partners are also exploring whether to extend their collaboration to other businesses within Hyundai Motor Group. Toh said the group’s broader operations were one reason JPMorgan chose Hyundai Card as a partner. No specific projects have been announced for that possible expansion.
JPMorgan has not set a commercialization timeline. The 2027 target is for further testing, not for customers to begin using an autonomous transaction service.
Loading discussion...
Join the conversation
Explain what you would still want a person to approve.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.