JPMorgan Tests AI Agents for Corporate Treasury

Semi-realistic illustration of Hyundai Card AI treasury agent at a desk, connected to Kinexys blockchain, reviewing FX quotes and approving a transfer under guardrails.

J.P. Morgan’s Kinexys has tested an AI-driven treasury workflow with Hyundai Card in which software agents monitored balances, compared foreign-exchange quotes and initiated funding instructions under predefined controls. According to the bank’s Trust in the Agentic Economy case study, the proof-of-concept demonstrated autonomous treasury decision-making while keeping execution subject to bank-enforced identity, mandate and authorization checks. The experiment ran in a test environment and did not transfer real value.

The scenario gave Hyundai Card’s treasury agent a standing objective: maintain a minimum balance in a hypothetical Frankfurt euro account by selling pounds from a named London account when funding was required. The agent checked the balance, requested FX quotes from three banks, selected the most competitive offer and submitted an execution request. A separate Kinexys-operated agent controlled the execution step, separating the AI system making the decision from the infrastructure authorized to move funds.

AI Authority Was Bound by a Machine-Readable Mandate

The control model began with identity. Hyundai Card generated a cryptographic keypair for its agent and registered the public key with Kinexys alongside an identifier and the legal entity represented by the software. The agent then signed outgoing messages using its private key. Kinexys could therefore verify which registered agent submitted an instruction instead of relying on the agent simply asserting its identity.

Authority was handled separately through a machine-executable mandate. The AI agent could draft that mandate, but Hyundai Card had to review and approve it before it became active. The approved version specified permitted transaction types, currency pairs, amounts, accounts, operating periods and escalation conditions. The agent could propose its own operating boundaries but could not grant itself additional authority.

Every request then passed through checks covering agent identity, the principal represented, mandate compliance, transaction coherence and audit logging. An instruction inside all permitted parameters could proceed automatically; one exceeding an escalation threshold required human approval, while activity outside the mandate was rejected. The test deliberately exercised all three outcomes—automatic execution, human escalation and denial—to demonstrate constrained autonomy rather than unrestricted AI control.

That model differs from consumer-facing systems such as Base MCP, where AI agents prepare onchain actions but users still approve the final transaction. Other architectures are experimenting with policy-limited smart accounts for autonomous agents. Kinexys instead tested delegated corporate authority that allows routine actions to proceed without individual human approval when they remain inside an institutionally approved mandate.

Proof-of-Concept Stops Short of Production Deployment

The experiment was intentionally narrow. It covered one funding objective, one currency pair and direction, two named accounts and a single relationship between Hyundai Card and J.P. Morgan. The client agent also operated inside the bank’s own agent-as-a-service environment. The test did not establish interoperability between autonomous agents operating across multiple banks, companies or independent networks.

That limitation is important because Kinexys itself already operates production blockchain infrastructure at much greater scale. J.P. Morgan says the broader platform has processed more than $4 trillion since inception and averages over $7 billion per day, while its JPM Coin deposit token is live for institutional clients on Ethereum. Those production metrics belong to Kinexys as a whole and should not be interpreted as usage generated by the Hyundai Card AI experiment.

The broader agent-payment market is developing through different architectures, including infrastructure for AI agents to purchase cloud services on a pay-per-call basis. Kinexys is approaching the problem from institutional governance: proving who an agent represents, defining what it may do and recording why each action occurred. The Hyundai Card test establishes that this controlled agent-to-agent treasury workflow can be built with existing technology, while cross-institutional standards and live deployment remain unresolved.

Find Us on Socials

Join Our
Newsletter

Subscribe to get latest crypto news!

Latest News

You may also like

The Chain Observer
Privacy Overview

This website uses cookies so that we can provide you with the best user experience possible. Cookie information is stored in your browser and performs functions such as recognising you when you return to our website and helping our team to understand which sections of the website you find most interesting and useful.