Is it safe to connect your bank to ChatGPT or Claude?
Is it safe to connect your bank to ChatGPT or Claude? What each assistant can see, why your password should stay out of it, and the read-only route to use.
It is safe to connect your bank to ChatGPT or Claude when the connection is read-only, you approve it in your bank's own sign-in, and your password never goes near the assistant. It is not safe to type your banking login into any assistant, and the banks that have spoken on the subject agree with that.
Plenty of people are uneasy about this, and with good reason: an ABA and Morning Consult survey in late October found that 77% of Americans are worried about an AI agent moving their money. This page explains what each route lets an assistant see and do, so you can pick the one you're comfortable with.
Is it safe to connect my bank to ChatGPT?
Connecting your bank to ChatGPT is safe when it can only read your accounts, you signed in through your bank's own pages and you can switch the connection off yourself. That is a simple test to apply to any assistant, whichever company makes it.
Finances in ChatGPT, which began rolling out to Free and Go users in the US on 2 October, works that way: you link accounts through a data network and ChatGPT sees the results.
What to check is simple. Ask whether your password goes anywhere other than your bank's own sign-in, whether the connection can move money, and how you end it. If the answers are no, no and in the settings, you're in good shape. If an assistant ever asks you to paste a banking password, stop, because that is the one thing the banks have said they will not accept.
Is Claude Personal Finance safe?
Personal Finance in Claude is read-only: Anthropic says Claude "can't move, manage, or invest your money", and it links up to 16 US banks for Pro, Max and Ultra plans. Anthropic also says Claude is designed not to save anything from those chats to memory. For a single assistant in the US, that is a careful design.
Its limits are the other side of the same care. It covers US banks only, it keeps no memory of what you discussed, and it lives inside Claude, so ChatGPT and Gemini don't see what you linked. If you use more than one assistant, you end up linking your accounts again each time, and each link is something to remember and revoke.
What can an assistant see when it's connected?
A read-only connection shows an assistant your account names, balances and transactions, and sometimes holdings, and nothing else. It does not show your password, it cannot sign in to your bank, and it cannot move money. Everything it learns comes from the accounts you chose to link.
With Era, the assistant also sees whatever you have chosen to keep in Era, such as your goals, your financial health figures and the documents you have filed.
You decide the scope. In Era you can hide an account from every assistant, make an API key read-only, and set it to expire on a date you choose. Hidden accounts stay hidden everywhere, so an assistant that only needs your everyday spending never learns about the savings you'd rather keep private.
What is the risk of giving an AI your bank password?
Your password can do everything you can do on your bank's website, for as long as the password works, and a bank that sees an agent using it can't tell the agent from you. That is why Chase, Citi, Wells Fargo and Schwab block it, and why Bank of America tells customers they are responsible for activity from shared credentials.
There is a second cost, which is liability. Wells Fargo's terms, from 24 November 2026, treat an agent's instructions as yours and make you responsible for revoking its authority, and Chase's from 1 January 2027 hold customers responsible for an agent's actions. A password handed to an assistant therefore leaves the money at risk and also leaves you holding the responsibility. Who's liable when an AI agent uses your login explains where that leaves you.
How does Era keep the connection safe?
Era keeps your bank login out of the picture: you sign in on your bank's own pages through a regulated data network, and Era holds only a read-only token for each connection. Era never sells your data and never uses it to train models, and it is encrypted at rest.
You can disconnect an institution, which ends access for every assistant at once.
Because every assistant connects to Era in place of connecting to your bank, you have one place to look and one place to switch things off. You can see which assistants have access, make a key that can only read, set it to expire and revoke it in a single step. An edit trail in Era shows what any assistant has changed.
- Hide any account from every assistant.
- Keep a key read-only, with an expiry date you choose.
- Revoke one assistant without touching the others.
- Disconnect a bank to end access everywhere.
What should I do first?
Link your accounts in Era, connect the assistant you use most, and ask it something you can check, such as your balance on a card or last month's grocery spending. When the answer matches your own records, you've confirmed the connection works.
How Era protects your financial data goes through the rest of the protections, and the guides for Claude and ChatGPT cover the steps.
Related reading
Last updated