Jamie Dimon cautions that AI payment agents could “may not” behave properly without bank-grade controls

JPMorgan Chase CEO Jamie Dimon warned that artificial intelligence agents capable of initiating payments for customers may not act properly, even if they are built with good intentions. Speaking in public remarks covered by Yahoo Finance, Dimon said such systems would need customers’ bank codes or equivalent credentials, effectively placing the AI inside the bank’s payment environment.

The concern, as framed in the report, is less about whether AI can understand instructions and more about how it behaves when it has authority. If an AI agent has the ability to execute transactions, Dimon suggested the chain of responsibility and controls becomes critical, because the agent’s actions could deviate from what a customer expects or from what internal safeguards are designed to enforce.

Dimon compared the broader AI shift to a force that must be managed. The report characterizes his remarks as describing an “AI revolution” that could arrive like a “tsunami,” implying that the pace and scale of deployment could outstrip existing governance, risk frameworks, and operational protections.

JPMorgan is not the only financial firm exploring AI for customer service, fraud detection, and internal workflows, but Dimon’s comment targets a specific use case: agents that execute payments rather than simply advising customers. In that setting, the operational risk is immediate and measurable, because payment errors, misuse, or misinterpretation can trigger downstream financial and customer impacts.

The report does not provide further technical detail on what JPMorgan’s approach would be, or whether the bank is actively restricting agent-driven payments. It also does not specify which internal systems would be involved if such agents were used, or what approval layers Dimon believes are necessary for safe operation.

Still, the underlying message aligns with a recurring theme in banking technology risk. Granting an automated system authority over credentials, payment initiation, or account access raises questions about authorization, auditability, error correction, and recovery when an agent makes a mistake. Without strong controls, an agent that is “helpful” in normal situations could become unpredictable when confronted with edge cases or imperfect inputs.

For JPMorgan, the stakes are unusually high because payment infrastructure is deeply regulated and tightly monitored, and because consumer trust depends on accurate execution. Dimon’s framing suggests he wants governance to keep up with capability, especially when AI decisions are tied to real money movement.