Business•

Hyundai Card and JPMorgan Let AI Agents Compare Currency Quotes and Move Funds Between Subsidiaries in a Kinexys Trial

Hyundai Card completed a proof of concept with Kinexys by J.P. Morgan in which AI agents monitored balances, compared bank FX quotes and executed transfers within preset limits.

Written By :MINRK
Hyundai Card and JPMorgan Let AI Agents Compare Currency Quotes and Move Funds Between Subsidiaries in a Kinexys Trial

Photo by Katja Ano on Unsplash

Mentioned in This Article

ETH
ETH
Ethereum
$1,664.81
2.38%
AVAX
AVAX
Avalanche
$6.53
4.10%

1. Hyundai Card puts AI agents to work on the Kinexys blockchain

Hyundai Card, the credit card and financial services arm of Hyundai Motor Group, said on Wednesday, November 18, 2026, that it has completed a proof of concept with Kinexys by J.P. Morgan. The test examined whether artificial intelligence agents can automate cash management, foreign exchange and fund transfers on the Kinexys blockchain run by JPMorgan Chase, the largest bank in the United States.

The exercise used simulated transactions, not live money. Software agents acting for each company watched account balances, gathered currency quotes, and carried out conversions and transfers between hypothetical overseas subsidiaries, with human staff brought in only when a transaction fell outside the limits set in advance.

The trial belongs to a field often described as agentic commerce, in which AI systems do more than answer questions or draft documents and are instead trusted to initiate and complete financial actions on behalf of a business. For a corporate treasury, the attraction is straightforward: many of the tasks involved in moving money between group companies are repetitive, rule-bound and time-sensitive, which makes them natural candidates for automation.

2. What is Kinexys?

Kinexys is JPMorgan's blockchain business, built around a permissioned network based on Ethereum (ETH) technology. It was previously known as Onyx and has been in operation since 2020, making it one of the longer-running blockchain platforms operated by a major bank.

A permissioned blockchain differs from a public network in one central respect: participation is restricted. On a public chain, anyone can submit transactions or run the software that validates them. On a permissioned system, the operator decides who is admitted, and every participant is a known and vetted institution. Banks tend to prefer this model because it allows them to use shared ledgers and programmable transactions while keeping the customer checks and controls that regulation requires.

JPMorgan has been committed to digital assets for a long time, and the Kinexys platform is the main expression of that commitment. For the test with Hyundai Card, it supplied the environment in which the two sides' agents interacted, as well as the agents themselves.

3. The scenario: keeping a subsidiary's balance above a floor

Hyundai Card designed the business case that the technology was asked to solve. It imagined a group with overseas subsidiaries, one of which had to keep its account balance above a set threshold. That is a familiar treasury problem. A unit operating abroad needs enough local currency on hand to meet its obligations, and a shortfall can be costly or disruptive.

In the simulation, whenever the receiving subsidiary's balance dropped below the required level, the sending subsidiary was to make up the difference. Doing so involved two steps that normally require human attention: exchanging the right amount of one currency for another, and then transferring the proceeds to cover the gap.

The threshold is what turns the problem into something a machine can manage. A rule of the form "keep this account above a given level" is precise and measurable, and it produces an unambiguous trigger: either the balance is above the line or it is not. The size of the required transfer follows from the same rule, since the amount to be sent is whatever is needed to close the gap.

In a conventional setup, treasury staff would spot the shortfall, request quotes from banks, pick the best one, instruct the conversion, and then order the transfer, often across different time zones and working hours. The proof of concept set out to test whether that whole chain could be handed to software operating under fixed rules.

4. How did the AI agents carry out the transactions?

Two agents shared the work: one representing Hyundai Card, which decided what needed to happen, and one representing Kinexys, which executed it. The "Hyundai Card AI Agent" kept watch over the subsidiaries' balances and compared foreign exchange quotes offered by several banks. Once it had identified what was needed, it instructed the "Kinexys AI Agent" to perform the currency conversion and send the funds.

The division mirrors the relationship between a corporate customer and its bank. The customer's agent holds the knowledge of the company's own positions and needs, and it shops for a price. The bank's agent sits on the side of the institution that actually moves the money, and it acts only on instructions it is satisfied are genuine.

Although the scenario came from Hyundai Card, the agents for both companies were built by J.P. Morgan. Hyundai Card, for its part, describes itself as having "a ton of proprietary expertise" in AI-powered data science, and it contributed the design of the test case that the agents were asked to handle.

5. Identity checks, limits and human approval

Letting software move money raises an obvious question of control, and much of the proof of concept was devoted to it. J.P. Morgan added mechanisms for identity verification, authorization, control and validation, all intended to secure transactions that pass directly from one agent to another without a person in between.

Before acting on any instruction, the Kinexys agent verified the identity of the Hyundai Card agent and confirmed that it was authorized to make the request. This step addresses a risk specific to automated systems: an agent on the receiving end must be able to establish that the counterpart issuing orders is the one it claims to be and that it has been given the authority to issue them.

Spending limits formed the second layer. Transactions that fell within predefined limits were executed automatically. Those that went beyond the permitted scope were not simply processed anyway. They were either routed to a human for approval or rejected outright.

This tiered arrangement resembles the approval rules that companies already apply to their own employees, where a junior member of staff can release small payments alone but larger ones need a second signature. The test applied the same principle to software, giving the agents freedom to act on routine transfers while keeping people responsible for anything unusual.

6. Building on a stablecoin remittance pilot

The work with JPMorgan is not Hyundai's first use of blockchain technology for moving money inside the group. It follows a stablecoin-based system for cross-border internal remittances, which the company brought to production readiness on the Avalanche (AVAX) blockchain.

Stablecoins are digital tokens designed to hold a steady value against a currency such as the US dollar. For internal remittances, their appeal lies in speed and availability: a transfer can settle at any hour, without waiting for banks in two countries to be open at the same time.

Taken together, the two projects address different parts of the same problem. The remittance system concerns the rail on which value travels between subsidiaries. The Kinexys test concerns the decision-making on top of that rail: who notices that a transfer is needed, who selects the exchange rate, and who gives the order. The earlier project automated the movement of funds, while the new one explores automating the judgment that precedes it.

The two efforts also rely on different kinds of infrastructure. The remittance system was developed on Avalanche, while the agent test ran on JPMorgan's permissioned, Ethereum-based platform, where access is limited to approved participants.

7. Why agent-to-agent payments matter for corporate treasuries

Cash management across borders is one of the more labor-intensive parts of running a multinational group. Balances must be monitored in several currencies, funding gaps have to be closed before they cause problems, and each conversion carries a cost that depends on the quote obtained. Comparing offers from multiple banks, as the Hyundai Card agent did in the test, is a task that rewards speed and consistency.

A blockchain platform adds something to that picture beyond the agents themselves. Because transactions on such a network are programmable, rules such as transfer limits and approval requirements can be enforced as part of the transaction process instead of being checked afterward. That is the context in which the identity and authorization controls built by J.P. Morgan operate.

Foreign exchange is a good illustration of why consistency matters. Banks can quote different rates for the same currency pair at the same moment, and on large or frequent transfers a small difference in the rate adds up. A person collecting quotes by hand may check only one or two providers when time is short. An agent can run the same comparison across multiple banks every time a transfer is triggered, and apply the same selection logic on each occasion.

Availability is a further consideration. A funding gap at a subsidiary does not necessarily appear during the head office's working day, and a process that depends on staff being present can leave a shortfall unaddressed for hours. Agents that monitor balances continuously, as the Hyundai Card agent did in the simulation, can respond when the threshold is breached instead of when someone next reviews the account.

The test also shows how banks and their corporate clients may come to interact. In the scenario, the company's agent did not log in to a banking portal or send a message to a relationship manager. It dealt with the bank's agent directly, and the bank's agent decided, on the basis of verified identity and preset limits, whether to act.

8. What the proof of concept does and does not show

The results should be read within the boundaries of the exercise. The subsidiaries were hypothetical and the transactions were simulated, so the trial demonstrated that the workflow can function from end to end, not that it is running in production with real funds.

Within those boundaries, the test covered the complete sequence: detecting a shortfall, sourcing and comparing foreign exchange quotes, converting currency, transferring funds, verifying the instructing agent, and escalating or refusing anything outside the agreed scope. Hyundai Card announced the completed proof of concept on Wednesday, placing a large Korean financial company and the largest US bank among the institutions now testing how far AI agents can be trusted with money.

Related Articles

NEWSLETTERS

Don't miss another story.

Subscribe to the MINRK Newsletter today.

Tick the newsletters you want. No selection? You'll get MINRK Headlines.

By signing up, you will receive emails about MINRK products and you agree to our terms of use and privacy policy.

MINRK Daybook

Market analysis for crypto traders and investors.

EVERY WEEKDAY

MINRK for Advisors

Defining crypto, digital assets and the future of finance for financial advisors.

EVERY THURSDAY

MINRK Protocol

Exploring the tech behind crypto one block at a time.

WEEKLY

MINRK Long & Short

A must read for institutions. Insights, news and analysis delivered weekly.

EVERY WEDNESDAY

MINRK Headlines

The biggest crypto news and ideas of the day.

EVERY WEEKDAY

MINRK Policy

Examining the intersection of cryptocurrency and government.

WEEKLY

MINRK Research

Join thousands of readers who rely on MINRK for data-driven insights on the latest digital asset trends.

MONTHLY