Blog
Why JPMorgan and Schwab Are Blocking AI Agents From Your Accounts
3 November 2026
Quick answer
Major financial firms, including JPMorgan Chase and Charles Schwab, are blocking unauthorized third-party AI agents from logging into customer accounts and moving money. Their reasons are security, unclear liability, and the risk of “rogue agents” acting on instructions the customer never really gave. The banks are not rejecting AI itself. They want AI access to happen through controlled, bank-approved channels.
What Happened in October 2026
In the middle of October, two of America’s biggest financial names said the same thing within days of each other. On its quarterly earnings call, Charles Schwab CEO Rick Wurster said the firm blocks unauthorized agents that try to sign in on a client’s behalf. His view was blunt: an agent arriving at the website with a client’s login is not a safe way to reach Schwab.
Around the same time, JPMorgan Chase CEO Jamie Dimon spoke about “rogue agents” and warned that there could be thousands of them. He also said people who hand tasks to agents should understand that liability shifts when they do.
Other large lenders followed a similar path. Reports also named Bank of America, Citigroup and Morgan Stanley as restricting outside agents, and Royal Bank of Canada later said it would block them too, citing unacceptable client risk.
The most important detail is that this is not a blanket “no AI” rule. Several of these banks still allow read-only access, where a tool can view information. What they refuse is letting an outside agent act, especially when it comes to moving money or paying bills.
What Is an AI Agent, and Why Is It Different From a Chatbot?
A chatbot answers questions. An AI agent does things. It can plan a sequence of steps, use connected apps, click through websites, fill in forms and complete tasks with very little supervision.
That difference matters in banking. A chatbot that gives a wrong answer wastes your time. An agent that misreads an instruction, goes beyond its permissions or behaves unpredictably can cost you real money, and it can do so in seconds.
Why Banks Are Blocking AI Agents
1. Credential sharing is a security problem
For an outside agent to act in your account, it often needs your username and password, or it has to automate a browser session. Handing those details to a third-party system widens the attack surface. If that system is compromised, so is your account.
2. Banks cannot easily tell humans from agents
Security teams build their fraud detection around human behavior. When software logs in and behaves like a person, existing controls get confused. Some banks reportedly try to spot agents by their IP addresses, but traffic can be routed through proxies, so detection is imperfect.
3. Liability is unclear
If an agent sends money to the wrong place, who pays? The customer who authorized it? The company that built the agent? The bank that processed the transfer? In the United States, the question of AI agent liability is still being debated, and there is no settled answer. Until there is, banks would rather not carry that risk.
4. Rogue or misdirected behavior
Dimon’s warning about “rogue agents” captures a real worry: automated tools doing things the customer never meant. An agent given a vague goal like “optimize my savings” could take steps the owner would never have approved.
5. Business interests are also in play
It would be incomplete to say safety is the only motive. Analysts have pointed out that agents could automatically sweep idle cash out of low-yield accounts into better-paying options, which could dent the interest income brokerages and banks earn on that cash. An outside agent works for the customer, while a bank’s own assistant works within the bank’s rules. That is a real tension, and readers should weigh the safety argument and the commercial argument together.
What the Banks Are Offering Instead
Rather than shutting the door entirely, firms are building controlled alternatives.
- Their own AI assistants. Schwab has pointed clients to its in-house assistant, which can handle limited actions such as updating an address or naming a trusted contact.
- Secure, rule-based connections. Schwab has said it plans to use Model Context Protocol (MCP) servers, which let a client’s chosen AI model connect under rules the firm sets. Through these links, outside tools could check balances and do research but could not move funds.
- Read-only access. Several banks keep view-only access while blocking payments.
This points to a likely future: AI that can see and analyze your finances, but needs explicit, verifiable approval to touch your money.
What This Means for Everyday Customers
If you use AI tools for budgeting or investing: Do not give any AI agent your banking username and password. Prefer tools that connect through official, bank-supported methods with limited permissions.
If you want help from AI without the risk: Use AI for analysis and planning. Ask it to explain statements, compare options or model scenarios, and then make the actual transfers yourself.
If you trade or invest: Remember that an agent acting on a vague instruction can make decisions you would not make. Keep final approval in your own hands.
Practical safety checklist:
- Never share online banking passwords with an AI tool.
- Turn on multi-factor authentication everywhere.
- Use read-only connections where possible.
- Review your bank’s terms of service, since some banks now restrict customers from letting third-party AI tools access accounts.
- Check account activity regularly for anything unfamiliar.
- Treat any tool that asks for full account control as a red flag.
What This Means for Businesses and Developers
For companies building financial AI, the lesson is clear. The limiting factor is not how smart the model is. It is permission, identity and accountability. An agent that can reason over account data may be acceptable, while an agent that can trigger payments is not.
Security experts have suggested that blocking will remain the most defensible control until agents carry verifiable identities, scoped permissions and an audit trail. In practice, the winners will likely be products that offer:
- Verifiable agent identity, so a bank knows who or what is calling.
- Scoped permissions, such as view-only or capped transaction limits.
- Audit trails, so every action can be traced.
- Clear liability terms between customer, provider and bank.
Consulting firms such as Accenture have also been telling financial institutions that governance and readiness, rather than raw model access, are what hold agentic AI back. That fits what we are seeing: the technology is moving faster than the rules around it.
Will Banks Eventually Allow AI Agents?
Probably, but on their terms. History offers a pattern. When new ways of accessing accounts appeared in the past, banks and technology companies eventually agreed on standards for identity, authentication and liability, and restrictions then loosened. The same path looks likely here.
Until those standards exist, expect more banks to block unauthorized agents, more to offer read-only access, and more to launch their own assistants. Regulators in the US and Europe are also asking banks how they handle AI agents, so formal rules may arrive sooner than many expect.
Key Takeaways
Verifiable identity, scoped permissions and audit trails are what will eventually let agents in.
JPMorgan and Charles Schwab are blocking unauthorized AI agents from customer accounts, and other major banks are doing similar things.
The main concerns are credential security, difficulty detecting agents, unclear liability and “rogue” behavior.
Banks still support AI through their own assistants, read-only access and secure, controlled connections.
Business motives, like protecting cash balances, also play a role.
Customers should never share banking passwords with AI tools and should keep final approval of payments.
Frequently Asked Questions
They cite security, control and liability concerns. Agents that log in with customer credentials can act beyond what the customer intended, and banks cannot easily verify who or what is behind the activity.
No. Many banks allow read-only access and are building their own AI assistants. The restriction is mainly on outside agents that log in as the customer or move money.
It is an automated tool that takes actions the user did not intend or authorize, either by misreading instructions or by exceeding its permissions. Jamie Dimon used the phrase to warn about uncontrolled agents operating at scale.
It is not recommended. Sharing credentials exposes your account to anyone who compromises that tool, and some banks’ terms now restrict this kind of access. Use official, limited-permission connections instead.
There is no settled answer in the United States yet. Liability between customers, AI providers and banks is still under discussion, which is one reason banks are cautious.
Model Context Protocol is a way for an AI model to connect securely to a company’s systems under rules the company defines. Banks like Schwab plan to use it so outside AI tools can view balances and do research without being able to move money.