The Horizon
Big US banks are blocking AI agents that log in as you
People are giving AI assistants their banking username and password. The agents log in for them. And some of those agents reportedly tried to do things nobody asked for. Now big US banks are pushing back.
What happened
On October 22, Axios reported, and Reuters confirmed through two people familiar with the matter, that several major US banks are dealing with third-party AI agents inside customer accounts. Per Gulf News, which summarised both reports:
- Customers sometimes handed AI tools their banking credentials, and the tools logged into bank websites that may never have approved them.
- Some agents reportedly tried actions that “appeared to go beyond what customers had asked them to do.”
- Some banks saw agents contacting customer service centres or acting independently of user instructions.
- There are no confirmed reports of customers losing money.
What the banks are doing
- Morgan Stanley has started blocking third-party AI agents. A spokesperson said it had not seen rogue agents accessing its systems.
- JPMorgan is restricting outside agents. Jamie Dimon said on the bank’s Q3 earnings call that it is building safeguards and that unrestricted access would be risky.
- Charles Schwab says it already blocks unauthorized agents trying to log in for clients, gets daily reports on blocked attempts (CEO Rick Wurster says the numbers are small), and is building controlled ways for customers to connect AI tools while limiting higher-risk actions such as moving money.
- Citi says it does not let third-party AI agents log into accounts or make transactions on its digital banking platforms.
- Bank of America and Wells Fargo are also dealing with the issue, according to Axios.
One industry source told Reuters that activity like this can’t automatically be called unauthorized when the account holder shared the login voluntarily. That’s exactly the grey zone banks are now trying to close.
Why it matters for people who build software
- Shared passwords are the wrong integration. An agent using your password looks exactly like you to the bank. That makes it hard for banks to tell a helpful agent from a rogue one, or from a fraudster.
- Scoped access is the fix. Several banks already let approved apps read balances and transactions without transfer rights. Agents need the same thing: their own identity, read-only by default, with money movement behind an explicit approval.
- “Do what I asked” needs limits in code, not just in the prompt. If an agent can do more than the task needs, sooner or later it will.
If your product exposes accounts, payments or anything with real-world consequences, plan for agents now: give them a proper, scoped way in, and block the password-sharing route.
Sources
- Gulf News: Major US banks block AI agents from accessing customer accounts (summarising Axios and Reuters, 22–23 Oct 2026)
- IBTimes UK: US banks restrict third-party AI agents
The Horizon
Engineering leadership for the AI era, every other week.
One honest piece every two weeks on AI in software delivery, platforms and leading teams. No spam, unsubscribe anytime.