JPMorgan AI Agents Ran an FX Test Alone; Kinexys: $7B a Day

JPMorgan and South Korea’s Hyundai Card have completed a proof-of-concept in which AI agents carried out a simulated foreign exchange transaction without direct human intervention, CNBC reported on 17 November. The companies are targeting the first half of 2027 for a more complex phase involving multiple AI agents, according to Wee Kee Toh, global head of Kinexys Insights at Kinexys by JP Morgan.

Our number: Toh’s example of an agent authorized to transact up to $5 million would be about 0.07% of the $7 billion that Kinexys says it processes on an average day, by our arithmetic. JPMorgan has not set a timeline for commercializing the technology, CNBC reported.

Business Pill · Least Privilege

An evergreen Business Engineer explainer on giving an AI agent only the access its task needs. It teaches the general idea only and does not discuss JPMorgan or Hyundai Card.

The key insight: JPMorgan’s Kinexys and Hyundai Card had AI agents carry out a simulated foreign exchange transaction without direct human intervention, CNBC reported; JPMorgan’s Wee Kee Toh says the harder part is governance, with agents confined to preset limits.

What JPMorgan and Hyundai Card Tested

CNBC reported that the initial trial involved a Hyundai Card AI agent requesting a foreign exchange transaction to maintain euro liquidity in its Kinexys blockchain deposit account. JPMorgan said it was “the first private blockchain-based autonomous bank-agent-to-client-agent test for corporate treasury management”, according to CNBC.

The next step is agents dealing with each other. Toh told CNBC: “What happens when agents are negotiating, agreeing directly with each other and then making payments? What does that actually look like?” He added: “That, I would say, is really the biggest part of what we want to get as an expansion of the phase one scope”.

Kinexys average daily transactions of $7 billion (J.P. Morgan Payments, August 2026) against the $5 million pe
Kinexys average daily transactions of $7 billion (J.P. Morgan Payments, August 2026) against the $5 million per-agent limit Wee Kee Toh gave CNBC as an example of how limits work. $5M / $7,000M is about 0.07%, by our arithmetic.

The Limit Is the Product

Toh said building the technology was relatively straightforward and that governance and internal controls were harder. “Technology-wise, it’s really not the biggest hurdle”, he told CNBC. Kinexys developed a framework that restricts AI agents to predefined transaction parameters and automatically rejects requests outside authorized limits, CNBC reported.

“You are going to draw a box around the agent. The agent will never be able to do anything out of the box”, Toh said. His example: an agent authorized to transact up to $5 million would be blocked from executing transactions above that threshold.

The Rail Underneath

Kinexys is JPMorgan’s blockchain payments unit. Its August release says Kinexys “has processed over $4 trillion in transactions since inception, with a $7 billion daily transaction average”. Against that average, Toh’s $5 million example is about 0.07% of one day’s flow, by our arithmetic.

The test sits next to banks’ caution on outside agents. In October we reported that JPMorgan and Citi curbed AI agents logging in with customers’ passwords; this week AI agents got a payment card. Companies are also putting limits on their own staff’s AI spending: Salesforce caps AI tokens at $1,000 a month and Apple caps some at $200.

The Business Lever

The lever is permissioned autonomy on a bank-owned rail. In JPMorgan’s framing to CNBC, the value is not the agent but the limit around it: a client’s agent can act only inside parameters the bank enforces on its own blockchain deposit accounts. That keeps the transaction, and the control layer, inside JPMorgan’s network.

The Structural Read

As we read Toh’s account to CNBC, the product being tested is the limit as much as the agent: a client agent acts only inside parameters the bank enforces, and requests outside them are rejected automatically.

As we read the scale, Toh’s $5 million example would be about 0.07% of an average Kinexys day of $7 billion, by our arithmetic, so the trial is about control design rather than volume.

Wee Kee Toh, global head of Kinexys Insights, to CNBC

“You are going to draw a box around the agent. The agent will never be able to do anything out of the box”

Three Implications

FOR CORPORATE TREASURERS The first test was an agent requesting a foreign exchange transaction to keep euro liquidity in a Kinexys blockchain deposit account, CNBC reported.

FOR JPMORGAN Kinexys says it has processed over $4 trillion since inception, with a $7 billion daily average, per its August release; no commercial timeline for agents has been set, CNBC reported.

FOR HYUNDAI CARD The companies are targeting the first half of 2027 for a phase with multiple agents negotiating and paying each other, according to Toh.

The Business Engineer Lens

This story maps onto the Business Engineer framework AI & The Harness Theory.

The framework says: “It is an operating system that wraps your AI — a harness that compresses your thinking, your team’s feedback, and the model’s raw capability into something replicable.”

As we read it, Kinexys’s predefined transaction parameters are a harness of this kind around the payment agents. The framework predates this article and does not discuss JPMorgan or Hyundai Card.

What Is Not Established

The size of the simulated foreign exchange transaction was not disclosed, and the $5 million figure was Toh’s example of how a limit works, not a stated limit on the trial. JPMorgan has not set a commercial timeline, and the companies have announced no specific projects with other Hyundai Motor Group businesses, CNBC reported. We did not contact JPMorgan or Hyundai Card.

This Week In The AI Supercycle

AI Lost Its Central Banker, the Lab Oligopoly, and the Future of Open Source

This week’s issue: the Nvidia leadership change, read as the AI build-out losing its central banker: who now finances the GPU landlords, backstops the data centres and sponsors open models, and why two labs are becoming the bank.

Read the full issue →

Business Engineer Framework

AI & The Harness Theory

A Business Engineer analysis of the operating system you build around AI models to make their output replicable.

Read the Map of AI →

The Bottom Line

JPMorgan and Hyundai Card ran a simulated foreign exchange transaction executed by AI agents without direct human intervention, on JPMorgan’s Kinexys blockchain network, and plan a multi-agent phase in the first half of 2027. JPMorgan’s Toh told CNBC the harder part is governance: agents confined to preset limits.

97,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.

A note on sourcing. The proof-of-concept, the next-phase timing, Wee Kee Toh’s quotes and the $5 million limit example are as reported by CNBC on 17 November 2026; we read the article in full. The Kinexys transaction figures are from J.P. Morgan Payments’ August 2026 release. The 0.07% comparison is our arithmetic. We first saw the story via a relay post on X. We did not contact JPMorgan or Hyundai Card.

Sources: CNBC, 17 Nov 2026 · J.P. Morgan Payments, Kinexys release, Aug 2026

Scroll to Top
FourWeekMBA