Markets

JPMorgan's Kinexys and Hyundai Card test AI agents that arrange a currency transfer on their own

The proof of concept used a simulated transaction between fictional overseas subsidiaries, not a live payment. The companies plan a test with several interacting agents in the first half of 2027.

Contents
  1. What the test involved
  2. Who did what
  3. The "first" claims
  4. What comes next
  5. What it could mean for corporate treasury and payments

Hyundai Card has completed a proof of concept with Kinexys, JPMorgan's blockchain business unit, in which AI agents from the two companies arranged a currency conversion and transfer between them, the South Korean card issuer said in a press release dated Wednesday, November 18, 2026. The test was a simulated transaction between fictional overseas subsidiaries, not a live payment, and neither company has given a date for commercial use.

What the test involved

According to Hyundai Card's release, which is in Korean, the scenario had a sending entity convert and transfer funds so that the receiving entity's account balance stayed above a set level. Both entities were fictional overseas subsidiaries, and the release calls the exercise a mock transaction.

In the test, the "Hyundai Card AI agent" checked the balances of both entities, compared foreign exchange (FX) quotes from different banks and then told the "Kinexys AI agent" to carry out the conversion and transfer. The Kinexys agent first confirmed the identity and permissions of the Hyundai Card agent. It executed transactions that fell within those permissions automatically and either asked a responsible employee for approval or rejected the rest, the release says.

CNBC, which reported the test on November 17 at 23:30 UTC, said the Hyundai Card agent requested the FX transaction to maintain euro liquidity in its Kinexys blockchain deposit account. The release does not name the currencies, the amount or the date of the test.

Who did what

Hyundai Card said it designed the scenario and built the permission and control framework for the agents. Kinexys developed both companies' AI agents on that basis and built the identity checks, authorization, controls and verification that protect the flow of transactions between them.

Wee Kee Toh, global head of Kinexys Insights, told CNBC that the framework limits agents to predefined transaction parameters. As an example, he said an agent authorized to transact up to $5 million would be blocked from going above that amount. He said governance and accountability, rather than the technology, are the main obstacle. "Technology-wise, it's really not the biggest hurdle," Toh said.

The "first" claims

Hyundai Card called the test the first case to verify that AI agents can exchange information and carry out financial transactions safely in a secured environment. It said earlier AI-based automation in finance mostly analyzed cash positions or executed trades on predefined conditions, while this test checked whether agents from different institutions could communicate and decide on the transactions needed.

JPMorgan described it to CNBC as the first private blockchain-based, autonomous bank-agent-to-client-agent test for corporate treasury management. We could not find a JPMorgan press release on the test, and neither claim has been independently verified.

What comes next

Toh told CNBC the companies aim to test more complex transactions involving multiple AI agents in the first half of 2027, with a focus on agents negotiating and paying each other. CNBC reported that no commercialization timeline has been set. A Hyundai Card official said in the release that the company plans to extend its cooperation with JPMorgan to new areas such as in-car payments.

The test comes days after Federal Reserve Governor Michael Barr warned that AI agents could speed up deposit outflows at banks under stress, and called for strict authentication, explicit consent and technical limits on what agents can do. Central banks have also been testing ledger-based settlement for currency trades, including a trial of currency trades settling across conventional and ledger-based systems by the UK, Singapore and Thai central banks.

What it could mean for corporate treasury and payments

In our assessment, the test shows how JPMorgan and Hyundai Card want AI agents in treasury work to operate: inside set limits, with identity checks between agents and with a human step for anything outside those limits. Because the transaction was simulated, it does not show how such agents perform with real money, real counterparties or market stress, and the companies have not said what the agents would be allowed to do in a live service. Who is accountable when an agent errs, the question Toh himself raised, remains open, as does how supervisors such as the Fed will treat agent-initiated transfers. Separately, Fed Governor Christopher Waller said on November 17 that central bank money must stay at the foundation of tokenized markets.

Sources

This article was researched and drafted with the help of AI tools, checked against its sources in a separate AI fact-check and edit, and approved for publication by Trading Metrics. Read our AI use policy.

This article is for information only and is not financial advice. Crypto assets and leveraged products are volatile and you can lose money. Do your own research and consider your circumstances before trading.

Trading Metrics · Trading journal and performance analytics platform

Trading Metrics (tradingmetrics.com) is a trading journal and performance analytics platform with the same owner as this publication. Articles under this byline are researched and drafted with AI assistance, checked against their sources in a separate fact-check and editorial review, and published only after the owner has approved them.