Claude can now read your bank accounts. What it sees, and what stays in your chats after you unlink
Anthropic's new Personal Finance beta links US bank and card accounts through Plaid. Access is read-only, but the fine print on data that stays behind matters more than the headline.
Anthropic announced on November 5 that Claude can now connect to your bank and card accounts and answer questions about your money. The feature is called Personal Finance, it’s in beta, and it’s limited for now to Claude Pro, Max and Ultra subscribers in the US. The headline promise is “read-only”: Anthropic says Claude can’t move, manage or invest your money.
That’s the part everyone will repeat. I care more about the smaller print, so here it is as questions a normal user would ask before tapping Connect.
What can Claude actually see?
Per Anthropic’s own announcement, for the accounts you choose to connect (and only those), Claude sees the bank’s name, each account’s name, type and last four digits, balances and credit limits, and up to two years of transactions.
Anthropic’s example is a user asking “Where did my money go this month?” and getting a category breakdown: $4,309.37 in total, more than half to rent and utilities, $1,024.51 to food and drink. Those are Anthropic’s sample numbers, not real user data. From there you can ask about duplicate charges, surprise fees, or what’s left before payday, and ask for charts or a dashboard.
The Next Web, citing Anthropic’s help pages, adds two limits. For investment, loan and card accounts, Claude reads only balances and credit limits. And you can link up to 16 US banks.
Does Claude see my banking password?
No, according to Anthropic. You sign in to your bank yourself through Plaid, the service that connects apps to banks, and approve read-only access there. Anthropic says Claude never asks for or sees the password. The Verge makes the same point and adds that banks don’t have to deal with unvetted AI bots crawling around their customers’ accounts.
What happens to the data after I unlink a bank?
This is the line I’d read twice.
Anthropic says you can unlink an account at any time and Claude can no longer see its data. But The Next Web reports, from Anthropic’s help pages, that figures already shared stay in chats, artifacts and shared links until you delete them, even after a bank is unlinked.
My read: unlinking stops new data coming in, and it doesn’t clean up what’s already in your conversation history. If you build a spending dashboard and share the link with a partner, unlinking the bank won’t pull that dashboard back. Delete the chat and the artifact too.
Is my financial chat used to train Claude?
Anthropic says conversations in the Personal Finance hub aren’t used to train its models. For anything you share with Claude outside the hub, your normal privacy settings apply. So asking a money question in a regular chat isn’t covered by the hub promise. The Next Web also reports that chats using linked accounts stay out of Claude’s shared memory.
Who gets it, and who is the rival?
| Detail | What Anthropic and outlets say |
|---|---|
| Plans | Pro, Max or Ultra, US only, beta (Anthropic) |
| Rollout | Starts the week of November 5 (Anthropic) |
| Connection | Plaid, read-only (Anthropic, Plaid) |
| Ads | None, and no cards, loans or savings products (Anthropic) |
| Rival | ChatGPT added a Plaid bank connection in May, first for its $200-a-month Pro tier (The Next Web) |
Anthropic didn’t name OpenAI in its post, but Engadget notes the company points to its no-ads policy as a contrast, and The Next Web says OpenAI has started showing ads in ChatGPT. I haven’t verified how ChatGPT’s feature has changed since May, so check OpenAI’s own page before comparing the two.
What I make of it
My read: this is a sensible, boring product, and that’s a compliment. A chatbot that explains an APR and finds a double charge is useful. A chatbot that could move money is a different risk. Anthropic has kept that door shut for now.
But “first step” is Anthropic’s own phrase. The announcement says the company is exploring ways for Claude to work alongside licensed professionals and regulated firms for bigger money decisions, and invites banks, wealth managers and other firms to talk to it. My read: that is a lead-gen line as much as a roadmap. Read-only today doesn’t mean read-only forever, and I’d treat any future change to those permissions as a new decision, not an update to click through.
Anthropic itself warns that Claude can make mistakes and tells users to ask which transactions sit behind a number and to check important figures with their bank. My read: good advice, because a wrong total in a polished chart still looks right.
My read: there is a business angle too. With enterprise AI spending reportedly flattening, as we covered in Ramp’s AI spending data, a consumer feature that gives paying subscribers a daily reason to open Claude is worth watching, though Anthropic hasn’t said that’s the motive.
Note on sources: I read Anthropic’s announcement and The Next Web’s article in full. For The Verge, Engadget and Plaid I read the opening excerpts only, so claims attributed to them come from those excerpts. I did not open Anthropic’s help pages, so the 16-bank limit and the “stays after unlinking” detail are as The Next Web reported them.