On November 18, JPMorgan and South Korea's Hyundai Card completed an initial proof of concept for autonomous payments by AI agents. During the test, Hyundai Card's AI agent initiated simulated foreign exchange transactions through JPMorgan's Kinexys platform to maintain euro liquidity in a blockchain deposit account. The two parties plan to conduct the next phase of testing in the first half of 2027 to explore whether multiple AI agents can interact, negotiate, and execute transactions. Wee Kee Toh, head of JPMorgan's Kinexys division, said that technology is not the main obstacle—the key lies in accountability mechanisms, governance, and internal controls. An AvePoint survey showed that among enterprises reporting AI agent-related security incidents, 34.1% had experienced unauthorized actions by agents. Agents may pass erroneous information to one another—triggering a chain of mistaken fund transfers before human intervention. To control risks, Kinexys has established a transaction limit framework that automatically blocks requests exceeding authorized limits; for example, an agent approved for a transaction cap of $5 million cannot execute transactions above that amount. Experts also recommend setting account permission limits and allowing users to revoke authorization at any time.
