Hyundai Card and Kinexys test AI agents under a treasury mandate
A proof of concept tested how an agent could check balances, request foreign-exchange quotes and initiate a funding operation within bank-enforced limits. The test moved no real money.

Hyundai Card and Kinexys by J.P. Morgan tested how an AI agent could help manage euro liquidity in a company account. The agent checked balances, sought foreign-exchange quotes and requested an operation when the account needed funding. The case study describes a test environment. No real value moved.
The company gave the agent a digitally signed mandate. When a request reached the bank, Kinexys checked the identity and authority behind it and evaluated it against bank controls. That separation makes the instruction a request the bank can independently verify. The model’s output alone carries no authority to move money.
The experiment gives treasury teams a concrete design question. Which parts of the job can software handle on its own, and what evidence must the bank have before acting? Balance monitoring and obtaining a quote carry different consequences from executing an exchange. An explicit mandate lets teams define that boundary and document how it was enforced.