Enterprise & Work

JPMorgan Tests AI Agents Making Bank Transactions Without Humans

JPMorgan's Kinexys and Hyundai Card completed the first autonomous bank-agent-to-client-agent blockchain test, a simulated FX trade, and aim to test negotiating AI agents by 2027.

JPMorgan is working on AI agent transactions that don’t involve humans
JPMorgan is working on AI agent transactions that don’t involve humansfancycrave1 / pixabay
By Rebecca Stone4 min read

Why it matters

  • JPMorgan's Kinexys and Hyundai Card completed the first private blockchain-based autonomous bank-agent-to-client-agent test, a simulated FX transaction.
  • A multi-agent phase, where AI agents negotiate and pay each other, is targeted for the first half of 2027.
  • Kinexys restricts agents to predefined limits; a $5 million cap would block larger transactions automatically.
  • Kinexys has processed over $4 trillion in transactions since inception, averaging about $7 billion a day.
  • Hyundai Motor Group holds 78.04% of Hyundai Card; a July pilot moved $20,000 in USDT between subsidiaries in seven minutes.

JPMorgan Chase and Hyundai Card have completed the first private blockchain-based test in which a bank's system executed a transaction with a client's AI agent autonomously, with no human in the loop. The two companies are now targeting the first half of 2027 for a far more ambitious phase: multiple AI agents negotiating with each other and making payments to one another directly.

The initial proof-of-concept, which JPMorgan described as the first private blockchain-based autonomous bank-agent-to-client-agent test for corporate treasury management, involved a simulated foreign exchange transaction. A Hyundai Card AI agent requested the FX transaction to maintain euro liquidity in its deposit account held on Kinexys, JPMorgan's blockchain-based payments unit. The bank's system handled the request without direct human intervention.

Why does this matter?

The pilot addresses one of the most consequential open questions in commercial AI: whether autonomous agents can be trusted to move money. If they can, AI agents could take over routine corporate treasury functions — liquidity management, currency conversion, payment execution — that today require human approval at every step. If they cannot, agent-based commerce stays confined to non-financial tasks.

The stakes for JPMorgan are direct. Kinexys said in August that it has processed more than $4 trillion in transactions since inception, averaging about $7 billion a day. Any mechanism that lets corporate clients automate even a fraction of that flow represents a substantial business line for the bank.

What happens when agents negotiate with each other?

The next phase of the collaboration will test what occurs when autonomous agents move beyond executing instructions and begin interacting with each other within a financial ecosystem.

Wee Kee Toh, global head of Kinexys Insights at Kinexys by JP Morgan, framed the question in blunt terms in an interview with CNBC: "What happens when agents are negotiating, agreeing directly with each other and then making payments? What does that actually look like?"

"That, I would say, is really the biggest part of what we want to get as an expansion of the phase one scope," he added.

The first half of 2027 is the target for that more complex multi-agent phase, according to the companies.

Is the technology the hard part?

By Toh's account, no. Building the transaction technology itself was relatively straightforward. The difficult work involved governance — establishing accountability and internal controls for financial transactions executed without direct human oversight.

"Technology-wise, it's really not the biggest hurdle," Toh said.

Kinexys's answer is a containment framework. The system restricts each AI agent to predefined transaction parameters and automatically rejects any request that falls outside authorized limits.

"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.

He gave a concrete example: an AI agent authorized to transact up to $5 million would be blocked from executing any transaction above that threshold. The limit is enforced by the system, not by the agent's own judgment — a design choice that shifts the trust question from the AI model to the bank's control layer.

When will it be commercialized?

JPMorgan has not set a timeline. The initial trial is complete, but commercialization remains undated, and the 2027 multi-agent phase is still a test, not a product launch. The gap between a successful simulated FX transaction and agents negotiating payments among themselves illustrates how much validation the bank believes is still required.

Why Hyundai Card?

Toh said Hyundai Motor Group's broader operations were one reason JPMorgan selected Hyundai Card as a partner. The two companies are now exploring whether to extend their collaboration to other businesses within the group. No specific projects have been announced.

The corporate structure gives the partnership room to grow. Hyundai Motor Group holds a combined 78.04% stake in Hyundai Card through Hyundai Motor, Kia and Hyundai Commercial, according to the card issuer's website.

Hyundai Card also brings its own blockchain payment experience. In July, the company completed a proof-of-concept involving a $20,000 transfer using the USDT stablecoin between Hyundai Motor's U.S. and Mexican subsidiaries. The transfer and verification process took an average of seven minutes.

Where does this fit in JPMorgan's blockchain business?

The agent pilot sits inside a broader expansion of Kinexys's blockchain-based payments business. Beyond the $4 trillion in cumulative transaction volume reported in August, the unit has moved quickly on deposit tokens.

JPMorgan launched a proof-of-concept for its dollar-denominated deposit token, JPMD, in June 2025. It made the token available to institutional clients that November — a roughly five-month path from test to institutional availability.

That trajectory offers a plausible template for the agent-to-agent work: a simulated transaction in 2025, a multi-agent test targeted for the first half of 2027, and commercial availability on a timeline the bank has not yet committed to. The open variable is governance. Toh's own assessment — that accountability and risk controls, not code, are the binding constraint — suggests the pace of rollout will depend on how confidently regulators and the bank's own risk functions can define the box the agents live in.

Source: CNBC Tech

Share this article:

More from Rebecca Stone

Rebecca Stone

Show full bio

Correspondent covering companies, deals and strategy in artificial intelligence and machine learning at AI In Context.

754 articles

Related articles

  1. JPMorgan and Hyundai Card Complete First AI-Agent FX Trade Trial
  2. Wall Street Is Pricing In an AI Banking Panic. Bank CEOs Aren't Buying It
  3. Bank of America warns customers not to share passwords with AI agents
  4. AI Agents Break Financial Rules 90% of the Time Under Pressure
  5. JPMorgan, Amazon and Kohl's Are Blocking AI Agents. Experts Say They Can't Win That Fight

« Previous articleNext article »