Anthropic has begun testing a feature that connects users’ bank accounts directly to Claude through the financial data network Plaid, allowing the assistant to analyze spending, receipts, and subscriptions without any manual uploads. The capability, branded Personal Finance, is deliberately constrained: Claude can read account data but cannot operate on it, meaning no transfers, no new contracts, and no touching the money itself.
- Read-only by design: Claude accesses account names, balances, credit limits, and up to two years of transaction history, but cannot move, manage, or invest funds.
- US-only beta: The feature launched on November 5, 2026, exclusively for Pro, Max, and Ultra subscribers in the United States, leaving Europe and other markets in the cold for now.
- Privacy caveats: Anthropic says Personal Finance conversations are excluded from model training, yet data persists in chats, artifacts, and shared links until users manually delete it.
What Claude Can See and Do
The scope of access is broad. According to details confirmed across multiple reports, Claude can view the bank’s name, account name and type, balances, credit limits, the last four digits of each account, and up to 24 months of transaction history complete with merchants, categories, and descriptions. From that raw material, it builds charts, bill calendars, and dashboards that Anthropic describes as easy to understand.
The analytical output is where the value sits. Claude can flag duplicate charges, surface unexpected fees, detect a subscription creeping upward, calculate interest paid on a loan, and help users anticipate upcoming expenses. It can answer questions like whether spending exceeds income. What it cannot do is equally important: it cannot pay a bill, send funds to a third party, or execute an investment. Anthropic frames this as a deliberate boundary, keeping the AI in the realm of analysis without handing over the keys to the account.
Authentication happens directly with the bank, so the bank password never reaches Claude. Plaid acts as the intermediary, governed by its own privacy policy. Will Robinson, Plaid’s chief technology officer, framed the partnership around comprehension rather than more data: most people don’t need additional financial information, he said, they need help making sense of it.
Availability and Competitive Context
Personal Finance remains tied to the US market and paid tiers. Free users and anyone outside the United States are excluded, and usage also depends on Plaid’s compatibility with individual banking institutions. The paid-first rollout fits Anthropic’s broader push to embed Claude in daily life, following earlier connections to services such as Spotify, Uber, Booking, and Instacart, and the introduction of more autonomous agents.
The move lands on ground already seeded by OpenAI, which launched personal finance tools with Plaid inside ChatGPT in May. Anthropic is positioning Claude as a privacy-focused competitor, emphasizing that the feature carries no ads or sponsored recommendations and that it offers no cards, loans, or savings accounts of its own. The company has also said it wants to explore partnerships with licensed financial advisers and regulated firms for bigger decisions, while stressing that the current tool does not provide investment advice and that Claude can make mistakes.
The Privacy Ledger
Anthropic’s assurances come with edges worth reading carefully. Conversations inside the Personal Finance hub are not used to train models, but the data stays in chats, artifacts, and shared links until the user deletes it. Unlinking a bank does not erase prior answers; that requires manual cleanup. A shared link could expose balances, figures, and charts. Cancelling a subscription stops Claude from reading accounts, but the bank connections persist until users disconnect them through Plaid’s own portal.
Neither company’s official documentation specifies how long financial data is retained in chats and shared artifacts, or how many banks a single user can link. No independent evaluation of Claude’s accuracy on this kind of data exists, so the described capabilities rest on the companies’ own accounts.
The read-yes, operate-no boundary is the right call, and it is what makes the feature defensible. Automated analysis genuinely helps people who never audit their subscriptions or notice a fee inching upward. But the privacy architecture follows a familiar pattern: comfortable to enable, awkward to reverse. Two years of transaction history inside a chatbot is an enormous portrait of a person’s life, and the persistence of that data after disconnection is precisely the loose end that turns a useful tool into a liability for the careless. Before handing over the ledger, users should weigh whether the analysis justifies the exposure, or whether a spreadsheet that never leaves the laptop would do the job just as well.