Major US banks are restricting external AI agents from accessing customer accounts as concerns grow over authentication, unauthorized transactions, and responsibility when automated tools act beyond users’ instructions.
Major US banks are tightening access to customer accounts as artificial intelligence agents begin taking on more tasks traditionally performed by people. The restrictions reflect a growing concern across financial services: how can banks support AI-driven convenience without losing control over account security and transactions?
According to Reuters on October 22, 2026, several major US financial institutions have been blocking third-party AI agents from accessing customer accounts and attempting transactions. The report, which cites people familiar with the situation, follows concerns that some agents may take actions that appear to go beyond a user’s instructions. Reuters reported on the issue as banks reassess how external AI tools interact with financial platforms.
The issue is not simply whether an AI assistant can read a bank balance. It is whether an independently operating system should be allowed to log in, contact a bank, initiate transactions, or move money using a customer’s credentials.
For banks, these actions raise questions about authentication, fraud prevention, customer consent, and liability. For fintech companies developing AI assistants, they also highlight a major barrier to building tools that can manage everyday financial tasks.
Why US Banks Are Restricting AI Agents
Traditional banking security systems are designed around customers, authorised representatives, and approved third-party connections. AI agents introduce a different type of access because they can interpret instructions, make decisions within defined tasks, and take multiple actions without continuous human involvement.
Some customers have reportedly shared their banking usernames and passwords with external AI agents. According to Reuters, banks have flagged and blocked certain transactions in these circumstances.
However, the distinction between customer-authorised access and genuinely unauthorised activity is complicated. A customer may willingly share credentials, yet an agent could still perform an action the customer did not intend.
That creates a difficult security question: does permission to access an account also mean permission to carry out every action an AI agent considers necessary?
Banks cannot safely assume the answer is yes. An agent asked to organize finances, for example, might interpret its task in ways that create unexpected account activity. If it can transfer funds or initiate payments, a misunderstanding could have direct financial consequences.
Reporting by Axios, referenced in the wider coverage, described cases in which agents appeared to contact bank call centers or take actions that seemed independent of users’ instructions. These reports have heightened concern about how financial institutions distinguish legitimate customer activity from potentially unsafe automated behavior. Axios's report provides additional context.
The available reporting does not establish that every bank has experienced a confirmed rogue-agent incident. Indeed, Morgan Stanley had blocked third-party agents but, according to a person familiar with the matter, had not observed instances of unauthorized or rogue AI agents on its platforms.
That distinction matters. Restricting access can be a preventive security measure, rather than evidence that a bank has suffered a confirmed breach.
JPMorgan Chase Draws a Line on External AI
JPMorgan Chase has publicly expressed concern about allowing outside AI agents to access its systems without sufficient safeguards.
Chief Executive Jamie Dimon said earlier in October that the bank was developing tools to manage the risks associated with AI agents acting on customers’ behalf. He also made clear that JPMorgan would not allow unrestricted access by external agents.
The bank’s position reflects a wider challenge for large financial institutions. They may develop and deploy AI internally, while still restricting independent third-party tools that attempt to interact with customer accounts.
These are not contradictory positions. A bank can control the environment, permissions, data access, and monitoring for its own AI systems. An external agent, by contrast, may operate under a different provider’s rules and technical safeguards.
JPMorgan must also consider what happens when an agent misunderstands an instruction or takes an action the customer later disputes. Without a controlled connection, the bank may have limited visibility into how the agent reached its decision.
The goal is therefore not necessarily to prevent AI from entering banking altogether. Instead, banks want greater control over which agents can connect, what information they can access, and which actions they can perform.
Morgan Stanley and Charles Schwab Prioritize Control
Morgan Stanley has blocked third-party AI agents from accessing its platforms, according to Reuters. The report also said the firm had not seen instances of unauthorized or rogue agents on its systems.
Charles Schwab has taken a similar approach to unauthorized agent access. Chief Executive Richard Wurster said the firm was blocking agents because it wanted to maintain control over authentication and the customer experience.
These positions highlight three connected priorities for financial institutions:
- Authentication: Banks need confidence that the person or system requesting access is permitted to do so.
- Transaction control: Viewing information and moving money carry different levels of risk. Banks need to distinguish between them.
- Accountability: If an agent acts unexpectedly, customers and institutions need a clear way to determine what happened and who bears responsibility.
For wealth managers and brokerages, the stakes extend beyond routine payments. External agents could potentially interact with investment accounts, request sensitive information, or initiate actions with financial consequences.
As a result, institutions may favor restricted, verified connections over tools that log in using a customer’s ordinary credentials.
The Bigger Problem: Who Is Responsible When an AI Agent Gets It Wrong?
The spread of AI agents creates a potential gap between customer consent and operational control.
A customer might authorize an agent to complete a task. However, that authorization may not clearly define the limits of the agent’s authority. The agent might also misinterpret instructions, encounter unexpected circumstances, or behave differently after a software update.
This creates several unresolved questions.
First, how should consent work? A customer should be able to grant limited permission for a specific purpose, rather than provide unrestricted access to an entire financial account.
Second, how should banks verify an agent? Financial institutions need ways to identify the software making a request, establish its permissions, and determine whether it is acting within those limits.
Third, who is liable for a disputed transaction? Responsibility may depend on the bank’s terms, the customer’s actions, the agent provider’s obligations, and applicable law. Sharing credentials does not automatically resolve every legal question about a transaction.
Finally, how can a bank stop an agent that changes its behavior? Traditional authentication may confirm that valid credentials were used. It may not prove that the resulting action reflects the customer’s original intent.
These questions will become more important as AI agents move from answering financial questions to performing tasks across multiple services.
Read-Only Access Could Offer a Middle Ground
A complete ban on external AI connections may limit useful applications. Customers could benefit from assistants that help them review spending, categorize expenses, compare financial products, or understand their cash flow.
However, not every application needs permission to move money.
A potential middle ground is to separate read-only access from transaction authority. Under a carefully designed model, an AI assistant might retrieve permitted account information through a secure, approved connection while remaining unable to initiate transfers or payments.
More sensitive actions could require additional authentication or explicit confirmation from the customer. Banks could also impose limits on transaction values, approved recipients, access duration, and the types of tasks an agent may perform.
Secure, customer-authorised data-sharing arrangements already exist in financial services. Yet they are not equivalent to handing an AI agent a username and password and allowing it to operate an account as if it were the customer.
The challenge is to create access models that are useful without giving external software excessive authority. Such models would still need robust monitoring, clear permission controls, and ways to revoke access.
The Axios report noted that banks recognize the need to find a middle ground as customers increasingly explore AI tools for financial tasks. That suggests the long-term outcome may involve more controlled access rather than a permanent rejection of all third-party agents.
What This Means for Fintech and AI Companies
The restrictions have implications for companies building AI-powered personal finance tools, digital assistants, and automated payment services.
First, access to financial data cannot be treated as a purely technical integration problem. Developers must account for bank policies, authentication requirements, and customer consent from the beginning.
Second, companies may need to demonstrate that their agents operate within clearly defined permissions. Banks will likely want to understand what an agent can access, what it can change, and how its actions can be monitored.
Third, developers need to consider the difference between recommending an action and executing it. An assistant that explains how to manage a budget carries a different risk profile from one that can transfer money between accounts.
Finally, trust may become a competitive advantage. AI products that provide clear controls, explain their actions and ask for confirmation before sensitive transactions may be easier for customers and financial institutions to accept.
For banks, the response also presents a strategic choice. Blocking uncontrolled access can reduce immediate risk, but customers may still demand AI-enabled financial services. Institutions will need to decide whether to develop their own assistants, support approved third-party tools or build secure pathways that allow both approaches.
Security May Shape the Next Phase of Agentic Finance
AI agents promise to make financial tasks faster and more convenient. However, their ability to act independently creates a new challenge for institutions responsible for protecting customers’ money.
The current restrictions show that banks are unwilling to treat ordinary account credentials as sufficient authorisation for any external AI system. They want to retain control over identity verification, account access and transaction execution.
That does not mean AI has no place in banking. Instead, it suggests that wider adoption will depend on clear permissions, secure connections, effective oversight and accountability when something goes wrong.
For fintech firms, the lesson is straightforward: building a capable AI agent is only part of the challenge. Building one that banks can safely trust with financial access may prove harder.
The key question now is whether the industry can develop a secure framework that lets AI agents assist customers without giving them uncontrolled power over their money.4
US banks are blocking third-party AI agents access to customer accounts, sources say


Comments
No comments yet. Be the first to share your thoughts.
Leave a comment