Anthropic has introduced Personal Finance in Claude, a feature that lets eligible users connect bank and card accounts and ask questions about their financial activity in everyday language. The company says the connection is read-only: Claude can analyse account balances and transactions but cannot move money, manage accounts or invest. The beta began rolling out in the United States to Pro, Max and Ultra subscribers during the week of November 5. The feature can help users understand spending patterns without requiring them to open several banking apps and calculate totals manually.
Anthropic says Personal Finance can summarise balances, categorise transactions and explain financial terms. Users can ask where their money went during a month, compare spending with previous periods, identify recurring charges or estimate what remains after upcoming bills. The company’s examples show Claude combining information from connected accounts to produce a consolidated view. It can also point to unusual or repeated charges, although users should verify any potential discrepancy against the original bank statement before contacting a provider or making a financial decision.
The connection is handled through Plaid. Users sign in to their bank through Plaid and approve read-only access; Claude does not ask for or see the bank password. For connected accounts, Anthropic says the service can see the bank name, account name and type, last four digits, balances, credit limits and up to two years of transactions. Anthropic also says conversations inside the Personal Finance hub are not used to train its models, and users can unlink an account at any time. These are important boundaries, but users should still read the permissions and privacy terms shown during setup.
Read-only access reduces transaction risk, not every risk
Restricting the connection to read-only access is an important design choice. It means the assistant is not authorised to initiate payments, transfer funds or change account settings through this feature. That reduces the consequences of a mistaken instruction or a compromised conversational session. It does not eliminate other risks. Account balances may be delayed, transaction descriptions may be ambiguous and categorisation can be imperfect. A model may also misunderstand a question or draw an incorrect conclusion from incomplete records.
Financial analysis depends on context that may not be visible in a transaction feed. A transfer between a user’s own accounts can look like spending, a refund can distort a category total and a pending payment may not appear in the same way as a completed charge. Shared accounts, credit card billing cycles and cash transactions can also complicate a monthly picture. Users should treat the assistant’s summaries as a way to organise information, not as a replacement for official statements or professional advice.
The feature may be useful for basic budgeting and explanation. Someone trying to understand a recurring subscription or compare grocery spending can ask a question without manually exporting transactions. The same convenience can be valuable for people who find banking interfaces difficult to navigate. But any decision involving debt, investments, taxes or a major financial commitment requires a higher standard of accuracy than a conversational estimate. The model’s answer should be checked against source records and the relevant institution’s own information.
Privacy and availability boundaries
Anthropic says Personal Finance starts as a beta in the United States for Pro, Max and Ultra plans. The company’s announcement does not establish availability for users in other countries or for every financial institution. Bank connectivity, account types and the amount of historical data available can vary. The company describes access to balances and transaction history, but users should check the exact permissions presented during account linking rather than assuming every account exposes the same information.
Personal Finance shows how AI assistants are moving into sensitive, practical workflows. The value is not simply that Claude can explain a financial term; it can combine records from several accounts and respond to questions about a user’s own situation. That makes accuracy, data provenance and privacy controls central product requirements. Anthropic’s read-only design limits what the assistant can do directly, but users should retain control over connected accounts and verify important conclusions before acting on them.