• Jamie Dimon cautioned that granting AI agents access to bank accounts raises serious security and liability risks, and JPMorgan will restrict outside agents from its systems while developing customer protections.
  • The warning comes as U.S. and European regulators intensify scrutiny of autonomous AI, with the Federal Reserve, OCC, and FDIC questioning banks on agent limits and oversight.
  • JPMorgan reported Q3 net income of $15.3 billion, up 6% year over year, but earnings per share missed analyst estimates amid weaker fixed-income trading and rising expenses.

Jamie Dimon, chairman and CEO of JPMorgan Chase (JPM), on Monday warned that giving AI agents access to bank accounts carries serious security and liability risks, and said the bank will not allow unfettered third-party agents into its systems. “We’re not going to allow just any outside agent to come in,” Dimon said, according to a person familiar with the matter. He emphasized that customers must understand that responsibility for misuse could shift when third-party AI tools access their finances.

The warning underscores a growing debate over control: who may authorize an AI agent to access money, what that agent may do, and who bears losses when it makes a mistake or is compromised. The issue is becoming a broader banking concern, not simply a JPMorgan security matter.

Regulators are already on the case. Reuters (TRI) reported today that the Federal Reserve, the Office of the Comptroller of the Currency, and the Federal Deposit Insurance Corporation are questioning banks about autonomous AI, focusing on limits to agents’ authority, human approval requirements, audit trails, monitoring, and emergency shutdown capabilities. The agencies are gathering information rather than prescribing a new uniform rule, according to people familiar with the matter.

European supervisors are taking a similar tack. The European Central Bank has given banks an end-October deadline for plans addressing AI-enabled cyber threats, while British supervisors are encouraging independent recovery infrastructure that can restore operations after core systems are compromised.

Banks themselves are adopting AI agents internally. An October 12 industry analysis cited Dimon’s July earnings-call remarks that JPMorgan had almost 1,000 AI use cases, with roughly 50 especially important ones spanning fraud, risk, marketing, and other functions. The distinction is therefore between controlled deployment and unrestricted third-party access, not between accepting and rejecting AI altogether.

Still, the liability warning needs careful interpretation. Granting access, authorizing a payment, and accepting responsibility for an agent’s error are different questions. Current debate centers on how to authenticate delegated authority and allocate responsibility; the headline alone does not establish that customers automatically lose existing protections.

The warning came as JPMorgan reported third-quarter results this morning. Net income rose to $15.3 billion, or $5.57 per diluted share, up approximately 6% from a year earlier. That fell short of the $5.90 analyst consensus cited by CNBC (VSNT). Excluding securities losses, EPS was $5.75, still below consensus. Reported net revenue rose 8% to about $50.2 billion, while net interest income climbed 13% to roughly $27.1 billion. Expenses grew 12% to $27.3 billion, outpacing revenue growth.

The earnings miss reflected weaker-than-expected fixed-income trading and expense pressure—not an established financial loss from outside AI agents. Dimon also said the bank reduced share repurchases during the quarter.

JPMorgan’s balance sheet remains formidable. At June 30, 2026, it reported $5.0 trillion in assets and $375 billion in shareholders’ equity. Its asset-management business managed $5.1 trillion of client assets, a separate measure from the bank’s own assets.

Leadership changes are also in play. In June, JPMorgan named Doug Petno and Troy Rohrbaugh co-presidents. Petno became sole CEO of the Commercial & Investment Bank, and Rohrbaugh took over Consumer & Community Banking following Marianne Lake’s retirement decision. Jennifer Piepszak remained chief operating officer, and Mary Erdoes remained head of Asset & Wealth Management. These succession moves are significant but have not been directly linked to the access warning.

The commercial concern extends beyond fraud. Agents could automatically compare savings rates and move customers’ money away from low-yield accounts, potentially benefiting depositors while raising banks’ funding costs and squeezing lending margins. Apollo (APO) chief economist Torsten Slok has warned of a potential system-wide problem if households collectively automate such movements; this is a risk scenario, not evidence that an AI-driven bank run has occurred.

That pressure interacts with rising rates. Higher yields make alternative destinations for cash more attractive, while banks must decide how much to pay to retain deposits. Dimon’s latest economic assessment highlights persistent inflation, large fiscal deficits, elevated asset prices, interest rates, and geopolitical uncertainty despite resilient consumer spending and business investment.

There is also a national and global operational dimension: an AI-enabled attack that interrupts account access or payments could affect households, payrolls, and business transactions beyond the targeted institution. BCG’s Sukand Ramachandran argues that secure, separately held recovery data matters as much as attack prevention.

The policy backdrop in the U.S. remains unsettled. Section 1033 of the Dodd-Frank Act concerns consumers’ access to financial-account information. The CFPB finalized an implementing rule in October 2024, originally scheduling compliance for the largest covered institutions from April 2026. But a June 2026 legal analysis reports that a federal court had blocked enforcement and that the CFPB was revising the rule, leaving substantial uncertainty over mandatory third-party access, security obligations, and potential data-access charges.

Crucially, a right to obtain account information is not identical to permission for an autonomous agent to initiate transactions. Agent access adds questions about consent, payment authority, financial advice, and responsibility for unsuitable decisions.

Internationally, Reuters reports that U.S. regulators are asking about Chinese-developed AI technology, while European officials have expressed concerns about limited access to frontier models relative to U.S. banks. This connects the issue to technology dependence, cross-border cybersecurity, and competitive access to advanced tools—not just domestic consumer banking.

The societal tradeoffs are becoming clearer. Customers may benefit from simplified financial management and improved returns on cash, but delegation creates unresolved questions about mistaken purchases, fraud, and recovery. Banks face costs to authenticate agents, track their actions, and maintain recovery systems. Fintech and AI providers depend on consumer trust and bank permission, with Bernstein (AB) analysts identifying trust, data access, and regulation as major adoption barriers. And service disruptions can spread through payments and the broader economy, making operational resilience a public-interest issue.

Federal Reserve Governor Christopher Waller recently framed the central question: “Who is on the hook if an agent makes the wrong purchase?” The available sources do not establish a broad public backlash or consumer polling response to this particular headline.

This is the next stage of the open-banking debate: first, whether third parties could retrieve financial data; now, whether software can act on that data and move money. The CFPB’s rulemaking process began in 2016 and produced the 2024 rule, but litigation and revision have left the U.S. framework unsettled.

Existing e-commerce payment frameworks provide a possible precedent for allocating losses, but agents introduce an extra layer of delegated authority. Banks must verify not only the account holder’s identity but also what the agent is authorized to do. Fraud systems designed around human behavior may also need recalibration.

Recent incidents add urgency. Reuters reports CrowdStrike (CRWD)’s assessment that a person in China likely used AI tools in an attack on South Korean banks, and separately reports rogue-agent involvement in a breach of the Hugging Face platform. These are related cybersecurity developments, not evidence that the customer agents described in the warning caused those incidents.

In the short term, the direction is toward more closely supervised deployment: explicit authority limits, human involvement in important decisions, auditable actions, and shutdown mechanisms. These are the specific safeguards regulators are investigating; they are not yet a single standardized regulatory regime. Cyber-recovery investment is also likely to become more prominent as European supervisors press banks to demonstrate that they can restore services after serious compromise.

Longer term, two expert perspectives frame the outlook. Slok warns that automated cash optimization could undermine banks’ access to cheap deposits and create financial-stability concerns. Bank of America (BAC) analyst Ebrahim Poonawala argues that banks will develop their own agent interfaces, potentially insulating themselves from disruption. His view is that banks increasingly need to “build for the agent, not just the customer,” although timing and magnitude remain uncertain.

Bernstein analysts likewise expect a gradual transition, with trust, access, and regulation determining how deeply consumer agents reshape financial services. Parallel developments include consumer-agent competition from technology companies, banks’ own internal AI deployments, and European efforts to strengthen cyber recovery. Together, these suggest a contest over the terms of access—not simply a choice between allowing all agents and banning all AI. The unresolved questions are who grants authority, how narrowly it is defined, how actions are recorded, and who pays when something goes wrong.

JPMorgan did not respond to a request for comment on the specific restrictions. This article was updated to include additional details from the Federal Reserve, OCC, and FDIC scrutiny, as well as the latest European regulatory deadlines.