NewsStocksJPMorgan Builds AI Agents That Execute Financial Transactions on Their Own

JPMorgan Builds AI Agents That Execute Financial Transactions on Their Own

Author: CryptoBriefing·

Key Takeaways

  • •JPMorgan plans to deploy long-running AI agents capable of operating for up to two hours without human input, with the rollout framed around regulatory compliance and internal control.
  • •A June 2026 whitepaper described an AI agent that detected a currency change on a payment exceeding $1 million and proposed a hedge that avoided a potential $2.3 million foreign exchange mismatch.
  • •CEO Jamie Dimon said on the October 13, 2026 earnings call that JPMorgan blocks outside AI agents from customer accounts to limit liability, and that customers who grant access assume that risk themselves.
  • •In October 2026, JPMorgan partnered with Mirakl to support agentic commerce while keeping payment processing control within its own systems.
  • •JPMorgan's AI initiatives are estimated to have generated approximately $2 billion in annual savings by 2025, along with a 20% increase in gross sales within private banking.
JPMorgan Builds AI Agents That Execute Financial Transactions on Their Own

JPMorgan has developed AI agents capable of executing financial transactions without a human involved, pushing Wall Street's largest bank past chatbots and document summarizers into more consequential territory: software that acts on its own. The shift matters because it moves AI from systems that draft text into systems that move money, raising questions about oversight, error handling, and who is accountable when an autonomous transaction goes wrong. The bank is simultaneously building defenses against AI agents it does not control, keeping outside systems away from customer accounts.

What JPMorgan's Agents Actually Do

JPMorgan Chase's agents are designed to run complex financial workflows with little ongoing oversight. Some operate autonomously, while others work in a semi-autonomous mode.

In June 2026, the bank laid out plans to deploy what it describes as long-running agents — systems capable of working for up to two hours without human input. The bank framed the rollout around regulatory compliance and internal control.

A whitepaper JPMorgan published that same June offered a concrete example drawn from treasury operations. An AI agent handling a payment worth more than $1 million noticed that the transaction's currency had changed. The agent flagged the switch and proposed a hedge to cover the exposure, then helped move the trade through approval. According to the whitepaper, the intervention avoided a potential $2.3 million foreign exchange mismatch. The example illustrates the operating pattern JPMorgan is standardizing on: the agent detects a problem and initiates a fix, while a human approval step remains inside the loop.

Building Agents, Blocking Agents

By late 2026, the bank's approach had expanded to include defensive measures that stop external AI agents from accessing customer accounts, with liability cited as the stated concern.

CEO Jamie Dimon addressed the issue on the bank's October 13, 2026 earnings call. He indicated that JPMorgan blocks outside agents to avoid the liability risks tied to customer accounts, and stressed that customers who choose to grant an agent access shift that liability onto themselves. The posture draws a clear boundary at a time when agentic commerce — software transacting on a user's behalf — is an emerging focus across the payments industry.

In October 2026, the bank partnered with Mirakl to support agentic commerce. The arrangement centers on robust payment processing while keeping control inside JPMorgan's own systems.

The Bigger AI Picture at JPMorgan

JPMorgan's AI efforts are estimated to have generated savings of approximately $2 billion annually by 2025. AI tools have contributed to a 20% increase in gross sales within private banking, while employees using these tools have seen efficiency gains of 30-40%.

The bank frames its next step as dynamic, decision-making digital workers that can stitch together multiple workflows, evolving from single-task automation. The open questions to watch from here: how banking supervisors treat autonomous execution, and whether other institutions adopt a similar stance of running their own agents while blocking outside ones.

Disclosure: The original report was edited by Diego Almada Lopez.