Banking

JPMorgan Tests AI Agents for Autonomous Transactions

4 min read · November 18, 2026
Hero illustration for the article “JPMorgan Tests AI Agents for Autonomous Transactions”

JPMorgan and Hyundai Card have completed a simulated foreign-exchange transaction carried out by AI agents without direct human intervention. The banks’ next planned test, in the first half of 2027, will involve multiple agents and more complex transactions, as JPMorgan works through how to govern autonomous payments.

What the first test involved

The proof-of-concept centred on a Hyundai Card AI agent requesting a foreign-exchange transaction to maintain euro liquidity in its Kinexys blockchain deposit account. JPMorgan described it as the first private blockchain-based test of an autonomous bank agent interacting with a client agent for corporate treasury management.

The trial was simulated; it was not a commercial service. JPMorgan has not set a timeline for making the technology available commercially.

Moving from execution to interaction

The planned expansion is intended to test what happens when agents do more than carry out a transaction: they would interact with one another in an ecosystem. That means the next phase will examine transactions involving multiple AI agents, rather than a single request within the initial proof-of-concept.

JPMorgan and Hyundai Card are also exploring whether to extend their work to other businesses within Hyundai Motor Group. The group’s broader operations were one reason JPMorgan chose Hyundai Card as a partner, but no specific projects have been announced.

Controls are the harder problem

JPMorgan’s Kinexys unit has developed a framework that confines agents to predefined transaction parameters. Requests outside the authorised limits are rejected automatically. The example given by the bank is an agent permitted to transact up to $5 million: it would be blocked from making a larger transaction.

JPMorgan’s Wee Kee Toh, global head of Kinexys Insights at Kinexys by JP Morgan, said the technical build was relatively straightforward compared with establishing governance and internal controls. The harder questions concern accountability and the risks of letting software move money without direct human oversight.

Why safeguards matter

Autonomous agents can create risks when they act on bad information or pass it to other agents. Data management and protection software company AvePoint said its State of AI 2026 report found that unauthorised autonomous agent actions were the third most common AI agent-related security breach among companies reporting such incidents, affecting 34.1%. The finding covers agent-related incidents broadly, not just corporate treasury payments.

AvePoint CEO Tianyi Jiang warned that interactions between agents could amplify mistakes and trigger transactions before a person can step in. For payments, that makes limits on spending and account permissions, explicit authorisation, activity monitoring and the ability to revoke access central to the design—not optional features.

What comes next

JPMorgan and Hyundai Card are targeting the first half of 2027 for expanded testing with multiple agents. The next concrete test will be whether the controls can govern not only an individual agent’s authorised actions, but also the transactions that emerge when agents interact.

Sources

  • cryptobriefing.com — “JPMorgan builds AI agents that can carry out financial transactions on their own”
  • cnbc.com — “JPMorgan is working on AI agent transactions that don’t involve humans”
  • timesofindia.indiatimes.com — “JPMorgan is working on AI agent transactions, and wants to keep humans out of it to make sure … – The Times of India”
  • threads.com — “JUST IN: JPMorgan develops AI agents capable of carrying out financial transactions without human involvement”
  • Linas Beliūnas — “Wild: JPMorgan, Morgan Stanley, and other major US banks are now finding AI agents inside customer accounts”