At the Singapore FinTech Festival (SFF2026) on Nov. 18, Ollie Harris (올리 해리스), global head of JPMorgan Kinexys, Nam Hee-jung (남희정), head of Hyundai Card's Payments Business Division, Jeong Sang-min (정상민), a department head in Hyundai Card's Payments Business Division, and Toh Wee Kee (토 위 키), global head of insights at JPMorgan Kinexys, discuss "How is trust established when AI agents transact?" (from left). [Photo by reporter Lee Ji-young]

SINGAPORE - Hyundai Card and JPMorgan showcased financial transaction technology in which AI agents compare foreign exchange rates and handle currency conversion and remittances. AI carries out transactions without direct human intervention, while deals that exceed preset authority are automatically blocked. The two companies stressed that the demonstration showed the potential of AI-based financial transactions and the importance of securing trust and control systems in autonomous processes.

Hyundai Card and JPMorgan on Nov. 18 demonstrated AI agent financial transaction technology they jointly developed at the Singapore FinTech Festival 2026 held at Singapore Expo.

The presentation, titled "When AI Agents Transact, How Is Trust Established?", was attended by Nam Hee-jung (남희정), head of Hyundai Card's Payments Business Division, and Ollie Harris (올리 해리스), global head of JPMorgan Kinexys, among others. They introduced the results of a proof of concept for AI agent-based financial transactions carried out by the two companies.

JPMorgan Kinexys is JPMorgan's blockchain-dedicated business unit. It provides services including cross-border fund transfers and real-time payments for corporate clients.

Harris said, "Kinexys has supported customers for more than 5 years so they can transact in near real time around the clock across multiple currencies and borders." He said the proof of concept began with the question of who would monitor and process transactions in a 24-hour payment environment, and whether AI agents could take on that role.

Nam said, "AI agents are always running, make independent judgments and can make decisions quickly." She said she expects it to become commercially realistic within a few years for AI agents to carry out payments and financial transactions in place of people.

She also stressed that the more autonomously AI carries out transactions, the more essential technology becomes for controlling related risks.

Harris said, "AI agents can make wrong decisions and may not even be sure that the decision is wrong." He said mistakes could affect large-scale transactions at machine speed, and it could already be too late by the time someone notices.

AI compares FX rates by bank, automatically executes purchase of up to 800,000 euros

The demonstration was conducted under a scenario that assumed treasury management operations between overseas subsidiaries of a global company.

Under the scenario, when the balance of a euro account at a virtual subsidiary in Frankfurt, Germany, fell below a certain level, Hyundai Card's AI agent detected it and converted pounds held in a London account into euros to fill the shortfall.

In the demonstration, the target balance of the euro account was set at 5 million euros, and the AI agent was to determine the need for currency conversion when the balance fell below 4.5 million euros.

After checking the account balance, Hyundai Card's AI agent requested foreign exchange quotes from 3 banks. One of them was JPMorgan Kinexys' AI agent, while the other 2 were set up in a simulated environment.

After comparing the exchange rates offered by each bank, the AI determined that Kinexys offered the most favorable terms and requested execution of the transaction.

Hyundai Card and Kinexys AI agents then exchanged messages several times in natural language as they proceeded with the transaction. In the process, a request to buy 800,000 euros was sent to Kinexys' core payment system, and the demonstration also showed the account balance being updated after processing was completed.

In particular, the messages exchanged between the two agents were based on natural language rather than the standardized payment messages used in existing financial transactions.

JPMorgan also introduced an example from testing in which the Kinexys AI agent asked Hyundai Card's AI agent for additional explanation when an unclear transaction request occurred.

Unlike existing financial software that operates according to predesigned procedures, the explanation said AI agents can find and carry out the necessary steps on their own to achieve a given goal.

"Transactions not permitted cannot be executed" - stress AI financial transaction control technology

The key point the two companies stressed in the demonstration was a function that blocks transactions when an AI agent goes beyond authority granted in advance.

JPMorgan built a separate control layer called the "Kinexys Agentic Layer" between AI agents and the existing payment system.

The layer applies an engine that determines whether to allow a transaction under predefined rules. Even if a transaction request arises based on an AI agent's judgment, it is structured to be forwarded to the actual payment system only if it meets the authority and rules granted by the customer.

Before executing a transaction, the system verifies the AI agent's identity, transaction authority and transaction purpose in steps. It checks whether the AI is an approved agent, whom it is transacting on behalf of, and whether the transaction matches the delegated authority.

JPMorgan said, "All transactions are delivered to the payment processing system only after passing procedures such as identity verification, authority validation and verification of transaction intent."

It added, "Transactions within the permitted scope are executed automatically, but those that are not are immediately rejected or request approval from a finance officer." It said the structure automatically handles routine work while maintaining human judgment in exceptional situations.

In a second demonstration to confirm this, the system requested a large euro purchase transaction requiring about 2 million pounds.

But the AI agent had a rule set that prohibited transactions that would reduce the pound account balance below 4.5 million pounds.

The transaction requested by the AI agent was judged to violate the limit, and Kinexys' control system rejected execution. The live demonstration screen also showed the transaction as having failed.

JPMorgan said, "The most important part of this demonstration is not the process of a normal transaction succeeding, but showing what happens when an AI agent tries an impermissible transaction."

The demonstration was conducted in a test environment connected to JPMorgan's backend systems, not a commercial transaction moving real funds. It implemented the processing procedures needed for typical financial transactions to confirm that transactions between AI agents are technically possible.

Hyundai Card "to expand to in-car payments" - JPMorgan also to strengthen cooperation

Hyundai Card said the proof of concept confirmed the possibility that AI can carry out corporate treasury management tasks while operating under human control.

Nam said, "Through this proof of concept, we confirmed the possibility that AI-based payments can be applied to actual work." She said the important question now is how much authority to grant AI agents and how to manage it.

She stressed that it is important to build a system in which AI agents operate under human control and can respond immediately when errors occur.

Hyundai Card plans to expand AI agent use cases beyond treasury management between companies to everyday payment areas such as in-car payments made while driving.

JPMorgan is also reviewing more complex multi-agent transaction structures through its cooperation with Hyundai Card.

Harris said, "We intentionally started with a small-scale use case to prove that we can build a safe transaction structure and integrate it with the overall payment process." He said it is reviewing the possibility of combining blockchain-based payment infrastructure with payments between agents.

He added, "We are currently looking at more complex multi-agent use cases with Hyundai Card," and said it will expand cooperation going forward.

Keyword

#Hyundai Card #JPMorgan #Kinexys #Singapore FinTech Festival #Kinexys Agentic Layer
Copyright © DigitalToday. All rights reserved. Unauthorized reproduction and redistribution are prohibited.