
Key Points
- 01JPMorgan and Hyundai Card ran a simulated FX trade executed by AI agents with no direct human input
- 02The trial used JPMorgan’s Kinexys blockchain and focused on corporate treasury euro liquidity
- 03A multi-agent test phase is targeted for the first half of 2027, with agents negotiating and paying each other
- 04JPMorgan stresses governance and strict transaction limits as it develops long-running AI agents
AI agents execute a simulated FX trade
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. The initial trial took place on JPMorgan’s Kinexys blockchain network and was framed as the first private blockchain-based autonomous bank-agent-to-client-agent test for corporate treasury management. In the test, a Hyundai Card AI agent requested a foreign exchange transaction to maintain euro liquidity in its Kinexys blockchain deposit account. The size of the simulated trade was not disclosed.
The companies now aim to expand testing in the first half of 2027 to more complex scenarios involving multiple AI agents. In this planned phase, agents are expected to negotiate with each other, agree on terms and execute payments inside the Kinexys infrastructure. The goal is to observe how autonomous agents behave when they move beyond simple execution into ongoing interactions within an ecosystem.
Governance, limits and control frameworks
JPMorgan characterizes technology development for these AI agents as relatively straightforward compared with establishing governance and internal controls. A central concern is accountability and the risk of allowing agents to execute financial transactions without direct human oversight. To address this, Kinexys has created a framework that confines each AI agent to predefined transaction parameters and automatically rejects any request outside authorized limits.
An illustrative example describes an agent authorized to transact up to $5 million being blocked from executing a larger transaction. The bank’s approach is often described as drawing a box around the agent so it cannot act beyond the limits set on the Kinexys platform. Despite completing the proof-of-concept, JPMorgan has not set a timeline for commercializing autonomous agent-based transactions.
Broader AI agent strategy at JPMorgan
The Kinexys experiment sits within a wider shift at JPMorgan from simple automation to long-running digital workers. In June 2026, the bank outlined plans to deploy agents that can operate for up to two hours without human input, managing complex financial workflows. These internal agents are designed to support regulatory compliance and strengthen internal control rather than replace oversight.
By late 2026, JPMorgan’s strategy also included defensive measures to block external AI agents from accessing customer accounts, reflecting concerns about liability. In October 2026, the bank partnered with Mirakl to support agentic commerce, focusing on robust payment processing while keeping control inside JPMorgan’s own systems. Together with the Hyundai Card pilot, these initiatives show the bank experimenting with autonomous capabilities while keeping transaction control on its own rails.
Next steps with Hyundai Card and Kinexys
Beyond the initial trial, JPMorgan and Hyundai Card are considering extending their collaboration to other businesses within Hyundai Motor Group, though no specific projects have been announced. The existing proof-of-concept demonstrates how a client AI agent and a bank AI agent can coordinate on liquidity management within a permissioned blockchain environment.
Kinexys says it processes about $7 billion in transactions on an average day and has handled more than $4 trillion since inception, highlighting the scale of the underlying payments network on which these agent tests are running. As the two companies move toward the planned multi-agent phase in 2027, the emphasis remains on combining permissioned autonomy for clients with strict, bank-enforced limits on activity.
Key Takeaways
- 01JPMorgan and Hyundai Card have moved beyond theory to a proof-of-concept where AI agents independently executed a simulated FX workflow.
- 02The main challenge is organizational rather than technical, with JPMorgan prioritizing governance, accountability and hard transaction limits over rapid commercialization.
- 03The Kinexys pilot and internal long-running agents both point to a model of tightly controlled autonomy, where agents act inside bank-owned systems rather than on open networks.