Banks are blocking ChatGPT and Claude: what to do
Banks are blocking ChatGPT and Claude from signing in as you. Here's why, what still works, and how one authorised link lets every assistant read your accounts.
If your bank is blocking ChatGPT or Claude, the fix is to stop giving an assistant your bank password and to link your accounts through a read-only connection that your bank allows. Banks have been clear that the password is the problem. A connection you approve in your bank's own sign-in, which only reads, is the part they still welcome.
The past month has made that line very visible. Chase, Citi, Wells Fargo and Schwab have each said they won't let agents sign in as customers, and the terms some of them are writing make you answerable for what an agent does with your login. This page explains what changed and what to do about it.
Why are banks blocking AI agents?
Banks are blocking agents because an agent that signs in with your password looks to the bank exactly like you, and the bank can't tell whether you asked for what happened. Its duty is to your money, so it closes the door that a password opens.
Jamie Dimon put it as "you don't want rogue agents, and it could be thousands of them."
Reports in October described banks spotting agents using customers' credentials, some acting beyond what the customer had asked for. Schwab's chief executive, Rick Wurster, summed up the worry about Meta's Muse: "We don't want someone just giving their password to Muse… who's liable?" The block is the bank's answer to that question, and the new terms are the second half of the answer. Who's liable when an AI agent uses your login covers that in full.
What is actually being blocked?
What is blocked is a login: an assistant typing your username and password into your bank's website or app on your behalf. What isn't blocked is read-only access that you approve through your bank's own sign-in, which Chase has said in terms is allowed through open-banking providers.
The first approach hands over the keys, and the second hands over a view.
The difference matters most when something goes wrong. A password can be used to do anything you can do, for as long as the password works. A read-only connection can only read, it shows up as a connection in your bank's settings, and you can end it without changing your password.
What should I do if my bank is blocking my assistant?
Take your password out of the assistant, link your accounts through a read-only connection instead, and connect the assistant to that. Your bank's own flow handles the sign-in, so the bank sees an approved connection and the assistant never sees your login. You keep the assistant you like and your bank keeps the control it wants.
That is how Era works. You link your banks once through a regulated data network, signing in on the bank's own pages, and Era holds read access that you can revoke. Then every assistant connects to Era in place of connecting to your bank. Connect any AI agent to your bank accounts walks through it.
Why does one hub work better than a link per assistant?
One hub means you approve your bank once, and every assistant you use shares the same view. If each assistant linked to your bank on its own, you would approve the same bank repeatedly, and each approval would be another connection that some terms say you must remember to revoke.
One link in one place is easier to understand, easier to audit and easier to switch off.
It also changes what an assistant can do for you. Claude's Personal Finance launch on 5 November covers up to 16 US banks, is read-only, and keeps nothing from those chats in memory, which is a sensible design inside one product. With Era, the same accounts feed Claude, ChatGPT, Gemini and Meta's Muse, with shared memory and your financial health in figures, with the analysis behind them. Say you're saving for a house in one assistant and the next one already knows.
What else do you get with Era besides the connection?
You get documents, spaces and the figures that sit behind a good answer. Era reads statements, payslips and tax forms from the Document Library, brings transactions from them into manual accounts, and lets you share a space with a partner or housemates so each of you shares only what you choose.
- Shared memory, so your goals and preferences follow you from one assistant to the next.
- Financial health, with the figures and the analysis behind them, available to any assistant you connect.
- Spaces, for sharing balances and splitting costs with people you trust.
- Documents, filed, read and cited when an assistant answers.
Era connects accounts across the US, the UK, Canada, France, Germany, Spain, and 40+ countries in all. Moving money between your own accounts is in private beta, off by default for every key, and every transfer needs your approval.
Is this going to get stricter?
Almost certainly, which is a reason to build on the route the banks already accept. The pattern in each statement is the same: no passwords, and responsibility with the customer who shares one. A setup that never involves your password is unaffected by tighter terms, and a setup that does is the one to change first.
The full list of what each bank has said is in which banks block AI agents, and the safety answer is in is it safe to connect your bank to ChatGPT or Claude.
Related reading
Last updated