JP Morgan and Hyundai Card experimented with a future in which software, not humans, drives cross-border payment decisions during the proof-of-concept, autonomous agents from two different organizations communicated, checked balances, matched foreign-exchange quotations and carried out a simulated currency transfer, all without a human signing off on every step. It did not involve real funds. That’s relevant because much of the early coverage of the concept emphasized “live autonomous finance.”
What it demonstrated was more limited and more relevant: how agent-to-agent workflows could be configured, verified, and limited within a compliant payment system. The experiment used JP Morgan’s Kinexys platform together with Hyundai Card, South Korea’s card issuer. Hyundai Card’s agent kept watch on euro liquidity within a fictitious treasury account, flagged a deficit below a set threshold, compared FX quotations from several, different, banks, and instructed JP Morgan’s Kinexys agent to perform the GBP-to-EUR FX conversion.
Table of Contents
Why Does it Matter That it Was Simulated?
In the institutional world, trust is often established through simulation. Financial firms will not give autonomous systems access to live treasury accounts until they have tested identity, permissions, limits, and audit logs. The proof-of-concept so proved less a matter of speed than governance. The design points toward a model in which an agent interacts within guardrails.
Source: Tekedia
A treasury agent, for instance, may only speak to approved currency pairs, only in units up to a predetermined size, only when liquidity falls below an absolute limit and only after an agent at an approved counterparty has validated that the treasury agent is permitted to do so. A bank agent may only act after identity has been verified, must do so in line with set compliance policies and with settlement instructions pre-approved.
Any of those constraints fails, the workflow terminates or escalates to a human. That matters because agentic finance presents a different question of risk than automation. In traditional automation, rules are set. In agentic finance, the AI interprets the goal, selects the tools and crafts the sequence of actions.
That flexibility can save costs and help solve new challenges, but it raises new questions over why agents did what they did and whether they stayed within their bounds. JP Morgan’s research on trust in the agentic economy has been on establishing verifiable identity, permissioned access and an audit trail.
Also Read: JP Morgan JLTXX Soars To $695M On Ethereum, Boosting Liquidity
What Kinexys Brings to Agentic Payments
Kinexys (previously Onyx) is JP Morgan’s blockchain and digital-payments division. It offers a permissioned network of tokenised deposits with blockchain settlement. The significance is that it is the territory for the digital assets, which could verify identity, balance, and settlement finality for settlement in the near-real time.
In context, stablecoins, and tokens are becoming the settlement asset that can be held and transferred digitalmoney by software, under regulatory supervision, via systems like the US GENIUS Act with bank-issued tokenised deposits.
Source: Fireblocks
Also Read: JP Morgan Predicts $1.2T Equity Issuance Rise Due To AI Costs
Agents Negotiate Directly
The next test, due in the first half of 2027, will see several agents negotiate, agree, and pay each other directly. Putting prices in agents will negotiate messages, accept, and refuse terms and settle without a central adjudicator.
Regulators and risk teams will want to understand how agentic disputes can be traced, who is liable and how to replay the agent’s movements. Apart from payment APIs, developers will also need to specify agent identities and permissions, and how to test those permissions before going live.
Source: Bombay Chamber
For exchanges and blockchain communities, they will need to decide whether agentic payments will settle on public blockchains, permissioned bank networks or in some hybrid fashion. The experiment isn’t a proof that AI agents are ready to run treasuries, only that banks are starting to define the rules under, which they might be given a shot.
Also Read: JP Morgan Chase Sued Over Alleged $328 Million Crypto Ponzi Scheme
Ananthyka J is a market reporter at Tronweekly, reporting on cryptocurrency news. She covers cryptocurrency markets, blockchain technology, and digital asset regulation, focusing on Bitcoin, Ethereum, DeFi, altcoins, and crypto policy. Her reporting emphasizes clear and accurate market coverage, including crypto market movements, regulatory developments, and blockchain adoption. She holds a BA in Journalism and Mass Communication and an MA in Communication and Media Studies. She has also completed multiple media internships, follows strict editorial and fact-checking standards, and discloses potential conflicts of interest when reporting.



