Updated October 23, 2026
AI assistants are moving from answering questions to operating software. Banking exposes the difference: summarizing a balance and initiating a transaction require very different levels of trust. New reporting about US banks restricting third-party AI agents makes that boundary a timely issue for anyone connecting an assistant to financial accounts.
The practical takeaway: a customer asking an agent to help does not automatically mean the bank supports that access method. Check both the connection and the permissions before letting software operate inside an account.
What the reporting actually establishes
Reuters reporting republished by Mint says major US banks are blocking third-party agents, citing two people familiar with the situation. It attributes the claim that some actions appeared independent of user instructions to Axios. Reuters also reports that Morgan Stanley blocked outside agents but had not observed unauthorized or rogue agents, according to a person familiar with the matter.
These distinctions matter. The accessible report is evidence of attributed journalism, not a public forensic incident report. It does not establish that every named bank experienced an agent-caused loss, that all outside AI services are banned, or that a particular model was responsible.
Three permissions that headlines can blur
- Permission from the customer: what the person wanted the assistant to do.
- Permission from the bank: whether the institution allows the tool and its connection method.
- Permission enforced by software: what the connection can technically read or change.
A prompt is an instruction; an access control is an enforced limit. Asking an agent to “only look” is weaker than a connection that cannot submit transactions. A careful reader should look for the technical boundary rather than assume a conversational promise creates one.
Blocking agents does not mean blocking every connected app
Chase’s official account-linking guidance describes connections where customers sign in directly with Chase, choose accounts and can remove access. That is evidence of an established data-sharing route, not confirmation that any particular AI assistant qualifies.
Compare an app that requests clearly described account data through a supported connection with a general-purpose agent asking for a banking password in chat. Both may promise budgeting help, but the access arrangements differ. Evaluate the actual integration, not the intelligence or popularity of the chatbot.
What to check before connecting an assistant
- Identify the route. Find the bank’s documentation for connected apps. Do not rely solely on the assistant vendor’s claims.
- Read the scope. Check whether it can view balances, retrieve transactions, create payments or modify settings.
- Limit the accounts. If the task involves one account, question why the connection requests all of them.
- Locate the off switch. Know where the bank and provider let you revoke access.
- Keep consequential actions under direct review. Verify the recipient, amount and timing yourself through the bank’s own interface.
For example, a hypothetical household wanting a monthly spending summary could begin with manually entered category totals rather than a persistent connection. That reduces the data exposed while testing whether the assistant is useful. It does not prove the summary is correct; the household still needs to check it.
What remains unknown
The public material reviewed here does not provide a complete incident count, loss total or bank-by-bank technical specification. Terms such as “rogue” can also hide uncertainty about customer instructions. Do not convert an anonymous-source description into a verified claim that an agent deliberately stole money.
Nor should readers assume a shared-password transaction settles legal responsibility. Outcomes depend on the facts, account terms and applicable rules. If an unexpected action appears, contact the bank promptly through its official app or published phone number.
Questions readers are asking
Are all banking AI tools being banned?
The reporting does not establish a universal ban. Bank-provided tools, supported app connections and unrestricted website automation are different arrangements.
Does read-only access remove every risk?
No. It can restrict actions while still exposing sensitive financial information. Data retention and downstream sharing remain important.
What should I do next?
Review existing connections, remove those you no longer need and keep bank credentials out of AI chats. Recheck the bank’s current guidance before enabling a new integration.
Related reading: evaluating personal AI access and privacy.
Source note: This is an original explainer based on attributed reporting and official account-linking guidance reviewed October 23, 2026. It is not a hands-on investigation or a determination of liability.