Crypto

JP Morgan and Hyundai Card Tests AI-Powered Payments Without Humans



JP Morgan’s next experiment will involve multiple AI agents negotiating directly with one another before settling financial transactions.

JP Morgan Chase and South Korea’s Hyundai Card are testing AI agents that can negotiate financial transactions and make payments without direct human involvement.

While the project could change how corporate payments work, JP Morgan’s early findings suggest that controlling autonomous systems may be harder than building them.

Testing Machine-to-Machine Payments

According to a CNBC report, the two companies completed an initial proof of concept involving a simulated foreign exchange transaction in which a Hyundai Card AI agent requested a currency exchange to maintain euro liquidity in its deposit account on JP Morgan’s blockchain-based banking platform, Kinexys.

The collaboration targets machine-to-machine payments, where AI systems, without a person in the loop, recommend actions, initiate and approve transfers, and verify their authenticity.

A second phase, planned for the first half of 2027, will look at how multiple AI agents can interact, negotiate agreements, and settle payments with one another. Wee Kee Toh, global head of Kinexys Insights, told CNBC that the goal is to see what happens when agents move beyond just carrying out tasks and start dealing with each other.

The technical challenge includes ensuring that these payment agents can sift through complex regulatory frameworks to stay within the laws of different jurisdictions, as well as detect fraud in real time. They must also handle edge cases that could confuse algorithms while maintaining clear audit trails to meet compliance requirements.

However, the executive pointed out that developing the technology was the easy part, while establishing governance and internal controls could prove more difficult. According to a recent report by AvePoint, at least 34% of security breaches linked to AI agents were due to unauthorized autonomous actions, including them making transactions and data changes without human approval.

You may also like:

Furthermore, AvePoint CEO Tianyi Jiang warned that interactions between AI agents could allow a compromised or malfunctioning agent to pass incorrect information to others and possibly trigger a chain of unauthorized actions.

Recall that in October, OpenAI revealed that a research model took funds without permission from a developer’s wallet on the Ethereum testnet to pay for deployment fees.

Strict Transaction Limits and Oversight

Regarding the concerns mentioned above, Toh pointed out that companies using the system can put guardrails around agents so that, for example, if an agent is authorized to transact up to $5 million, any requests above that amount will be rejected automatically.

“You are going to draw a box around the agent. The agent will never be able to do anything out of the box,” he said.

Alongside spending limits, Hassan Ahmed, Coinbase’s Singapore country director, added that safe agent payment should also have restricted account permissions and clear authorization requirements, and give users the ability to monitor activity and revoke access at any time.



Source link

Leave a Reply

Your email address will not be published. Required fields are marked *