JPMorgan Chase and South Korea’s Hyundai Card have successfully completed a proof of concept in which artificial intelligence agents carried out simulated foreign exchange transactions and fund transfers without direct human intervention.
The trial, conducted on JPMorgan’s Kinexys blockchain platform, marks one of the first instances of autonomous bank-to-client AI agent communication for corporate treasury management on a private blockchain.
In the simulated scenario, a Hyundai Card AI agent monitored account balances of hypothetical overseas subsidiaries.
Register for the next Tekedia Mini-MBA.
Register for Tekedia AI in Business Masterclass.
Join Tekedia Capital Syndicate and co-invest in great global startups.
When a euro-denominated account in one location fell below a predetermined threshold, the agent compared foreign exchange quotes from multiple banks, selected the most favorable rate, and instructed a corresponding Kinexys AI agent to convert currency and transfer the necessary funds.
The entire process from detecting the liquidity need to requesting and authorizing the transaction occurred without a human typing an order or approving each step in real time.
Speaking on this, a Hyundai Card official said,
“This PoC is significant in that it validated a next-generation financial transaction model in which AI agents handle corporate cash management and foreign exchange operations without human intervention. We plan to expand our cooperation with JPMorgan into new areas such as in-vehicle payments.”
During the exercise, no actual money moved. The focus remained on testing technical feasibility, secure communication between agents belonging to different institutions, and the control frameworks required for autonomous financial activity.
JPMorgan developed the AI agents for both sides and implemented identity verification, authorization checks, and validation mechanisms. Hyundai Card designed the transaction scenarios and defined the agents’ authority limits.
A critical element of the system is the set of guardrails known as the Kinexys Agentic Layer. This control layer sits between the AI agents and the underlying payment infrastructure.
It verifies the identity and authority of the requesting agent, confirms the transaction falls within pre-approved parameters, and either executes the action automatically or blocks it and escalates it for human review.
Officials have emphasized that the technology itself is less challenging than establishing clear governance, transaction limits, and accountability. An agent authorized up to a certain amount, for example, simply cannot complete a larger payment.
The companies presented the results at the Singapore FinTech Festival and have indicated plans to expand testing in the first half of 2027.
The next phase will explore more complex interactions in which multiple AI agents negotiate and transact with one another within a broader ecosystem. Commercial deployment has not yet been scheduled.
The experiment highlights both the potential and the caution surrounding agentic AI in finance. Autonomous agents could eventually streamline routine treasury operations such as liquidity management and cross-border payments, operating around the clock on blockchain infrastructure.
At the same time, banks and regulators continue to stress the need for robust controls, clear liability frameworks, and strong cybersecurity measures before such systems move beyond controlled simulations.
The successful proof-of-concept demonstrates that AI agents can already communicate, analyze financial conditions, and initiate transactions under carefully defined boundaries.
As digital banking adoption accelerates, banks are deploying AI-powered technologies to improve operational efficiency, strengthen security, reduce transaction costs, and deliver more personalised financial services.
AI is also changing how banks process and manage financial transactions. Through intelligent automation, financial institutions can streamline payment processing, verify transaction details, automate reconciliation, and reduce the manual intervention required to complete routine banking operations.
Looking ahead, the deployment of agentic AI could mark a significant shift in how banks and financial institutions manage transactions, particularly in corporate treasury operations, foreign exchange, and cross-border payments.
As AI agents become more sophisticated, financial institutions could move towards systems where routine financial decisions, liquidity monitoring, currency conversions, and fund transfers are handled autonomously within predefined limits.
JPMorgan and Hyundai Card’s planned expansion of testing in 2027 could provide further insight into how autonomous financial transactions can operate at scale.



