Anthropic's Claude Can Now Read Your Bank Accounts, but Never Move Money

TL;DR

Personal Finance in Claude, announced November 5 by Anthropic, connects your bank and card accounts so you can ask Claude about your money in plain language. The beta is US-only on Pro, Max, and Ultra plans, access is read-only, and Plaid handles the bank sign-in so Claude never sees your password. Claude reads balances, credit limits, and up to 24 months of transactions from up to 16 banks, and conversations inside the Personal Finance hub are excluded from model training. ChatGPT got the same kind of bank view in May. Anthropic is differentiating on what it refuses to do with the data.

What shipped

A Personal Finance section is appearing in Claude's sidebar on web, desktop, iOS, and Android, rolling out gradually. You open it, pick goals or skip them, and click Link accounts with Plaid. Plaid opens its own window, you sign in to your bank there, you choose which accounts to share, and the link has to finish inside a 10-minute window. Claude asks for nothing else: your password stays with Plaid, and account numbers arrive trimmed to the last few digits.

The hub shows widgets for spending, cash flow, subscriptions, and upcoming bills, and you can ask questions in the hub or in a regular chat. Anthropic's announcement walks through its worked example: a month at $4,309.37, more than half of it rent and utilities, $1,024.51 to food and drink, then four upcoming checking-account payments leaving $1,393.22 before payday. Ask what looks off and Claude reads across every connected account for duplicate charges, creeping fees, and renewals you forgot about.

What Claude actually sees

The data scope is narrower than "connected to your bank" sounds. For each account you share, Claude reads the bank's name, the account name, type, and last four digits, balances and credit limits, and up to 24 months of transactions with amount, date, merchant, and category. You can link up to 16 banks. For investment, loan, and card accounts it reads balances and credit limits only: no holdings, no trades, no rates or terms.

The isolation rules are the product

Most of the design work went into what the feature refuses. Conversations in the Personal Finance hub are not used to train Anthropic's models. Chats that touch your linked accounts stay out of Claude's memory, out of search, and out of the monthly recap, and in those same chats Claude may not run code or produce Word, Excel, PowerPoint, or PDF files. The hub works like a clean room in a chip fab: your money walks in, and nothing about it walks out into the parts of the product with a longer memory.

Revocation is explicit too. Unlink a bank and Claude stops reading every account under that login, all at once. You can also cut Claude off from my.plaid.com, Plaid's own dashboard, which keeps working even if your Claude plan lapses. The help center has sharp edges worth reading before you connect anything: artifacts can read the linked accounts of whoever opens them, a shared chat link shows your figures to anyone who has it, and data already shown stays in chats and artifacts until you delete them.

Scope caps run through the docs. US banks only. Personal plans only, with no Free, Team, or Enterprise access. Linked accounts cannot appear in scheduled tasks, Cowork, or Claude Code. Move your account into a Team or Enterprise org and Anthropic unlinks your banks automatically. Banks can also end the connection themselves, a reminder that the whole arrangement rides on their tolerance.

The business model gap

Anthropic states the commercial positioning in one breath: no ads, no sponsored recommendations, and no credit cards, loans, or savings accounts of its own. The feature is explicitly not investment advice, and the company tells you to check important figures with your bank because Claude can be wrong. The Next Web draws the contrast with OpenAI, which brought Plaid reads to ChatGPT in May starting on its $200-a-month Pro tier and has since started showing ads in ChatGPT. Anthropic is selling the absence.

The announcement closes with the wedge. For life's bigger money decisions, Anthropic wants Claude working alongside licensed professionals and regulated firms, and it is inviting banks, wealth managers, and financial services companies to help shape what comes next, as PYMNTS reported. TNW also points to an existing adviser-facing version and an integration with the accounting platform Xero. This beta is the consumer on-ramp to that pipeline.

What to steal

• Ship eyes before hands. Read-only is the entire permission model here: Claude can explain your money, and it cannot move, manage, or invest it. Hands can come later, once the read path has earned trust.
• Scope the data policy to the surface. Hub chats skip training, account chats skip memory, search, and recaps, and tool use shrinks in exactly the chats that touch money. A privacy boundary drawn per feature beats one global setting.
• Put the kill switch in two places. Unlinking inside the app plus disconnecting at Plaid's own dashboard covers the user whose app access is already gone.
• Trim the tool list where the risk lives. Blocking code execution and Office/PDF output in account-touching chats removes obvious exfiltration paths, and shrinking the toolkit in the risky room is a design decision worth copying.
• Design artifacts for the second viewer. An artifact that reads the opener's accounts turns a shared link into a statement you did not mean to publish. Assume every rendered surface will be seen by someone else.

Caveats

• This is a gradual beta on US personal plans, and if you do not see it yet, the rollout is the usual explanation. The help center lists the rest.
• Anthropic's own docs warn that balances can be a day or two old, that an account linked through two logins can be counted twice, and that Claude can make mistakes. Reconcile anything that matters with your bank.
• The privacy commitments are current policy statements from a beta, and the docs already show persistence quirks around shared links and artifacts. Treat the boundary as real but young.
• No pricing, revenue, or partner terms were announced, and the OpenAI timeline comes from TNW's reporting; neither lab has published a side-by-side.

Key Takeaways

• On November 5, Anthropic announced Personal Finance in Claude, a US beta on Pro, Max, and Ultra plans rolling out across web, desktop, iOS, and Android.
• Access is read-only through Plaid: up to 16 US banks, with balances, credit limits, and up to 24 months of transactions; investment, loan, and card accounts share balances and limits only.
• Conversations in the hub are excluded from model training, and chats that use linked accounts stay out of memory, search, and monthly recaps, with code execution and Office/PDF output blocked in those chats.
• Anthropic promises no ads, no sponsored recommendations, and no financial products of its own, and positions the feature as explicitly not investment advice.
• ChatGPT gained Plaid-based bank reads in May on its Pro tier, and The Next Web reports OpenAI has since started showing ads in ChatGPT.
• Anthropic is inviting banks, wealth managers, and financial services firms to help shape the next version, pointing toward partnerships with regulated firms for the bigger decisions.

Sources: Anthropic, Introducing Personal Finance in Claude, Claude Help Center, The Next Web, PYMNTS