Hyundai Card’s AI agent compared exchange quotes from several banks, then told JPMorgan’s AI agent to convert and move the money — no human involved, the companies said Wednesday.
The simulated transaction capped a proof of concept with Kinexys by J.P. Morgan testing whether AI agents can run cash management, foreign exchange and fund transfers on their own.
Hyundai Card, the credit card and financial services arm of Hyundai Motor Group — the conglomerate behind Hyundai Motor Co. (KRX: 005380) — ran the test with Kinexys, the blockchain division of JPMorgan Chase (NYSE: JPM). Kinexys, which started operations in 2020 under the name Onyx, runs on an Ethereum-based, permissioned ledger, as CoinDesk reported.
Why does a simulated treasury exercise matter to anyone outside Seoul? Because until now, AI in finance has mostly analyzed positions or executed transactions under preset conditions. This proof of concept, or PoC, tested something harder: whether AI agents owned by two different institutions could exchange information, decide on their own that a transaction was needed, and request its execution — the pattern the payments industry calls agentic commerce.
“Through this PoC, we verified that AI agents can manage funds and conduct foreign exchange transactions without human intervention,” a Hyundai Card spokesperson said, as Chosun Biz’s English service reported.
The short read for treasury desks
- Hyundai Card completed a PoC with Kinexys by J.P. Morgan in which AI agents at two separate institutions executed currency conversion and fund transfers without direct human intervention.
- The control that made it bankable: only transactions inside predefined authorization limits were executed automatically; anything beyond that scope was escalated for human approval or rejected outright.
- Hyundai Card names in-vehicle payments as the next collaboration area with JPMorgan. No production timeline for the treasury use case has been disclosed.
Two AI agents, one authorization gate
The scenario was deliberately mundane, and that is the point. Hyundai said the test involved simulated transactions between hypothetical overseas subsidiaries, with the receiving entity’s account balance kept above a predetermined threshold. When a shortfall appeared, the sending subsidiary’s agent converted and transferred the necessary funds, CoinDesk reported.

The division of labor was clean. Hyundai Card designed the transaction scenario. J.P. Morgan developed the AI agents for both organizations and implemented the identity verification, authorization, control and validation mechanisms securing transactions between them.
During testing, the Hyundai Card AI agent monitored subsidiary account balances, compared FX quotes from multiple banks, and instructed the Kinexys AI agent to execute the required conversion and transfer. The Kinexys agent then verified the identity and authorization of its counterpart before acting.
But the test’s real test was the gate. Hyundai said that deals staying inside the authorized boundaries cleared without a human touch, whereas any request beyond those boundaries was sent up for a person’s sign-off or turned down altogether.
Chosun Biz described the same control from the Korean filing: for out-of-scope transactions, the Kinexys agent requested approval from the person in charge or rejected them. Two outlets, one mechanism.
Why this isn’t preset automation
Here is the distinction the sources keep drawing. According to Financial News, AI-driven financial offerings to date have centered on assessing cash positions or carrying out trades governed by pre-established rules. This project implemented a system in which agents from different institutions exchange information, assess whether a transaction is needed, and request execution — decision-making distributed across organizations, not scripted inside one.
“This PoC confirmed the potential of a next-generation financial transaction model in which AI agents manage companies’ funds and conduct foreign exchange transactions without human intervention.”
The wording differs slightly between the two outlets — both are translations from Korean — but the substance matches: agents transacting across institutions, unsupervised within their limits.

Avalanche remittances came first
This is not Hyundai’s first move onto blockchain rails, and the sequencing tells you where the group is heading. The Kinexys exploration builds on an earlier stablecoin-based cross-border remittance pilot, in which Hyundai moved an internal, stablecoin-denominated remittance system into production readiness on the Avalanche blockchain, CoinDesk reported.
Here is where the two pilots part ways.
| Attribute | Stablecoin remittance pilot | Kinexys AI-agent PoC |
|---|---|---|
| Blockchain | Avalanche | Kinexys (Ethereum-based, permissioned) |
| Function | Cross-border internal remittance | Cash management, FX, fund transfers via AI agents |
| Status | Production readiness | Proof of concept completed |
Read the two rows together and the strategy is legible: rails first, then autonomy. Hyundai proved it could move value across borders on a public-chain stablecoin rail; now it is testing whether software agents can decide when and how to move that value, inside a permissioned bank-run environment.
JPMorgan, for its part, has been building this infrastructure since launching the business unit in 2020 — the PoC gives Kinexys a marquee non-financial corporate client to show for it.
In-car payments are the prize
The treasury use case may be the proving ground, but it is not the destination. Both Korean outlets report the same forward plan from Hyundai Card: broaden the JPMorgan collaboration into payments made from inside vehicles. Chosun Biz quotes the spokesperson framing it as expansion “into new areas, including in-car payments.”
The logic is straightforward for a group that sells cars by the millions. An AI agent negotiating and executing a payment on a blockchain rail is the same architectural pattern whether the counterparty is a treasury desk or a toll road, a charging network or a parking garage. Hyundai Card also brings what CoinDesk describes as proprietary expertise in AI-powered data science to the payments business.
What remains unknown is the commercial path. Neither company disclosed transaction volumes, the currencies tested, or when — or whether — the PoC moves from simulated subsidiaries to production money. That gap between a completed validation and a deployed product is where most corporate blockchain projects have historically stalled.
Hyundai Card, at least, is signaling intent. “We plan to expand our cooperation with JPMorgan into new areas, such as in-vehicle payments,” the spokesperson told Financial News.
Disclaimer: The information provided in this article is for educational and informational purposes only and does not constitute financial advice.