JPMorgan and Hyundai Card test AI agents for autonomous payments
The financial giants are testing a system where AI agents negotiate and execute transactions independently, aiming for a larger rollout by 2027.

The short version
- JPMorgan and Hyundai Card are testing AI agents to handle financial transactions without direct human oversight.
- The initial proof-of-concept involved a simulated foreign exchange transaction using a private blockchain.
- The companies target the first half of 2027 for a complex phase involving multiple interacting AI agents.
- A new governance framework restricts agents to predefined limits to prevent unauthorized spending.
- Kinexys, a JPMorgan unit, has processed over $4 trillion in transactions since its inception.
Why it matters
This shift toward autonomous finance could significantly speed up corporate treasury management by removing human bottlenecks. It also sets a precedent for how companies will manage the risks of AI-driven spending and accountability.
Testing autonomous finance
JPMorgan Chase and South Korea's Hyundai Card are currently exploring how artificial intelligence agents can manage financial matters independently. The goal is to create a system where these agents can handle payments and negotiations without requiring a human to click a button for every step. This move targets high-potential areas like corporate treasury management, where moving money quickly and accurately is essential.
The two companies recently finished an initial proof-of-concept that simulated a foreign exchange transaction. During this test, a Hyundai Card AI agent requested a transaction to maintain euro liquidity in a blockchain deposit account. This marked the first private blockchain-based test of an autonomous bank-agent-to-client-agent system for corporate use.
While the initial trial is complete, the companies have not yet set a specific timeline for when this technology will be available to the general public. They are currently aiming for a more complex phase to begin in the first half of 2027. This next phase will focus on how multiple agents can interact within a larger ecosystem to settle debts and agree on terms.
Moving beyond simple execution
The current phase of the project focused on the basic execution of a single request. However, the next step involves moving toward a model where agents can communicate with one another to reach agreements. Wee Kee Toh, the global head of Kinexys Insights at Kinexys by JP Morgan, explained that the biggest goal is seeing agents interact in an ecosystem. He noted that the technology itself is not the primary hurdle for the project.
The real challenge lies in establishing the governance and internal controls necessary to manage autonomous spending. Because AI agents can move money, companies must find ways to ensure they do not perform actions that are unintended or unauthorized. The project seeks to answer what it actually looks like when two different AI agents negotiate a price and then execute the payment directly with each other.
To solve these concerns, Kinexys developed a specific framework designed to restrict the agents' behavior. This framework forces the AI to operate within predefined transaction parameters. If an agent receives a request that exceeds its authorized limits, the system is designed to automatically reject the transaction to protect the company's assets.
Drawing a box around AI
The safety framework works by creating a digital boundary for what the AI is allowed to do. Wee Kee Toh described this process as drawing a box around the agent. He explained that the agent will never be able to do anything out of that box, which provides a layer of security for the humans overseeing the system.
For example, if a company authorizes an AI agent to transact up to $5 million, the system will block any request that goes above that specific threshold. This ensures that even if the AI makes a mistake or encounters an unusual situation, it cannot spend more than the human operators intended. This approach balances the speed of automation with the safety of human-defined limits.
By using these constraints, JPMorgan and Hyundai Card hope to build trust in autonomous agents. The goal is to allow the AI to handle the
Blockchain and corporate scale
The project relies heavily on blockchain technology to facilitate these autonomous transactions. Hyundai Card has a history of experimenting with blockchain for cross-border payments, including a previous test involving a $20,000 transfer between U.S. and Mexican subsidiaries. That specific test used a stablecoin and took an average of seven minutes to complete both the transfer and the verification process.
JPMorgan's Kinexys unit is already a major player in the blockchain payments space. The unit has processed more than $4 trillion in transactions since it began, which averages out to about $7 billion in transactions every single day. This scale provides a significant foundation for the new AI agent experiments being conducted with Hyundai Card.
The partnership was chosen because of Hyundai Card's connection to the broader Hyundai Motor Group. JPMorgan is exploring whether this collaboration can be extended to other businesses within the group. While no specific projects have been announced for those other entities yet, the scale of the group's operations makes it a primary target for these new payment technologies.
The path to 2027
As the companies look toward the 2027 expansion, the focus will shift from single transactions to complex, multi-agent interactions. This will require more sophisticated testing of how different AI systems from different companies can
Sources and links
- Officialjpmorgan.com/payments/newsroom/kinexys-usd-digital-deposit-tokens jpmorgan.com
- Officialsaid in August jpmorgan.com
- ReportingCNBC: the article we sourced
First reported by CNBC. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.