NewsCryptoJPMorgan and Hyundai Card test autonomous AI agents for payments in blockchain first

JPMorgan and Hyundai Card test autonomous AI agents for payments in blockchain first

Author: Cryptopolitan·

Key Takeaways

  • •An AI agent operated by Hyundai Card autonomously submitted a foreign exchange request to increase euro liquidity in its deposit account on JPMorgan's Kinexys private blockchain platform.
  • •JPMorgan and Hyundai Card plan a more complex multi-agent testing phase in the first half of 2027 to examine how autonomous agents communicate and interact beyond simply executing transactions.
  • •JPMorgan has not announced a commercialization timeline, reflecting industry-wide difficulties in moving AI agents beyond pilots due to accountability, security, and compliance concerns.
  • •Kinexys currently restricts AI agents to defined transactional parameters and rejects any request that falls outside those boundaries.
  • •JPMorgan CEO Jamie Dimon said the bank excludes external agents from customer accounts over liability concerns, with customers assuming that liability when granting agents access.
JPMorgan and Hyundai Card test autonomous AI agents for payments in blockchain first

JPMorgan Chase and Hyundai Card, a South Korean card issuer, are exploring how far AI agents — software that can plan and execute multi-step tasks with minimal human oversight — can go in carrying out financial tasks on their own, from negotiating terms to processing payments without human involvement. The two companies have completed an initial proof of concept built around a simulated foreign exchange transaction, which JPMorgan describes as the first-ever private blockchain experiment involving autonomous bank and customer agents in corporate treasuries.

In the test, an AI agent operated by Hyundai Card submitted a foreign exchange request to increase euro liquidity in its deposit account on Kinexys, JPMorgan's private blockchain platform, according to a CNBC report.

Multi-agent testing planned for 2027

JPMorgan and Hyundai now intend to run a more complex, multi-agent AI testing phase in the first half of 2027. Wee Kee Toh, who leads Kinexys Insights at Kinexys, said the next phase will examine how autonomous agents communicate and engage with each other beyond simply executing transactions.

For now, the work remains a simulation rather than live payments, and JPMorgan has not announced a timeline for commercializing the technology, a gap that reflects broader concerns among financial institutions about trusting autonomous AI systems with financial operations. Banks are still struggling to move AI agents beyond pilot projects, as covered in a separate Cryptopolitan analysis, largely because of accountability, security, and compliance concerns.

According to Toh, the organizations still need to develop appropriate governance models and implement the required safety measures.

Use cases beyond foreign exchange

The applications for autonomous financial agents are not limited to currency trading. Corporate treasury departments manage liquidity, currency exposure, and payment coordination, and ensure businesses have funds across multiple accounts. These processes often span multiple institutions, making them complex.

AI agents could eventually automate parts of this work: identifying liquidity requirements, requesting currency conversions, and coordinating payments between accounts. For multinational companies like Hyundai, that would help reduce delays in cross-border treasury operations and allow businesses to respond more quickly to changing financial needs. Even so, transaction limits and verification mechanisms would be needed to ensure transfers are not made by accident.

In Toh's view, building the technology was not the hard part — the real problem is accountability and the risks of letting AI agents execute financial transactions on their own.

To date, Kinexys has built a model that restricts AI agents to defined transactional parameters and rejects any request that exceeds those boundaries, per JPMorgan's announcement. "You are going to draw a box around the agent. The agent will never be able to do anything out of the box," Toh said.

JPMorgan's warnings on agent behavior

Last September, the bank highlighted concerns about AI agents, noting that a command is just an intention and nothing more. Even if an agent is instructed not to access the internet, not to share data, or to stick strictly to a defined investment mandate, those guidelines only work if the tools, communication channels, and execution paths enforce them.

The bank cited one security evaluation in which an agent was explicitly told it had no internet access but still managed to create malware that could run on real-world systems after accidentally discovering a connection during the test. A larger review linked to that evaluation also found unauthorized access affecting numerous organizations.

In another set of evaluations, agents co-opted an internal package service to act as a persistent message board, passed discoveries across separate runs, and compromised parts of external production infrastructure. They also bypassed internal research infrastructure and accessed sensitive secrets on a massive scale.

Building toward agentic finance since June

JPMorgan Chase first shared its plans to deploy long-running agents in June. In a whitepaper published that same month, the firm described a treasury operations scenario in which an AI agent managing a payment exceeding $1 million detected a change in the transaction's currency — a catch that reportedly averted a possible $2.3 million currency mismatch.

Later in the year, the bank took a further step by implementing defensive measures to protect customer accounts from external agents attempting to access them, a move tied to liability considerations.

Speaking during the October earnings call, JPMorgan Chief Executive Officer Jamie Dimon clarified that the bank excludes outside agents because of liability issues associated with customer accounts. He asserted that customers themselves assume that liability when deciding to grant an agent access to their accounts.

That same month, the bank partnered with Mirakl, an e-commerce marketplace platform provider, on agentic commerce, which supports in-house payment processing.

Source: Cryptopolitan