A US beta for connected financial questions
Anthropic announced Personal Finance in Claude on November 5, 2026. The beta begins rolling out in the United States to Pro, Max and Ultra plan subscribers. Users can connect eligible bank and card accounts through Plaid, then ask Claude about balances, transactions, spending categories, recurring payments and budgeting questions. The feature is not described as globally available, and a staged rollout means it may not appear for every eligible account at once.
Anthropic says access is read-only: the assistant cannot move, manage or invest money. That is an important limit, but it does not make the output infallible. The company explicitly warns that Claude can make mistakes and advises users to inspect the transactions behind a figure and confirm important amounts with their bank.
What data the connection can expose
Anthropic says users sign in to their financial institution through Plaid and approve access there; Claude does not request or receive the bank password. For accounts a user chooses to connect, the article says Claude can see the institution name, account name and type, last four digits, balances, credit limits and up to two years of transactions. Users can unlink an account, after which Claude can no longer see its data.
The company also states that conversations in the Personal Finance hub are not used to train its models. It distinguishes those conversations from other Claude chats, where the user’s normal privacy settings apply. Readers should review Anthropic’s current privacy information, Plaid’s terms and the account-specific permission screen before connecting. The product’s read-only design does not remove the sensitivity of transaction history or the risk of a mistaken summary.
Where conversational summaries help—and where they can mislead
A natural-language interface can make it easier to ask questions across several accounts, such as how spending changed over a few months or which recurring charges are approaching. This is a convenience layer over financial records, not an authoritative ledger. Categorization can be imperfect, pending transactions may change, and a question like “Can I afford this?” depends on obligations and context the connected data may not capture.
Before relying on a total, ask which transactions and date range were included. Check transfers, refunds, cash activity and shared accounts carefully. Do not use a conversational result alone for tax filings, loan decisions, investment choices or payment authorization. Anthropic says Personal Finance is not designed to provide investment advice.
A useful pattern for consumer AI assistants
The announcement shows a broader direction: assistants are moving from answering general questions toward working with user-authorized personal data. That raises the bar for clear consent, narrow permissions, visible data provenance and easy revocation. For a good financial assistant, an answer should make it practical to verify the underlying transactions rather than ask the user to trust a confident paragraph.
For consumers, the sensible test is limited: connect only an account you are comfortable sharing, try a simple question, verify the result in the bank’s own interface and disconnect if the product does not meet your expectations. Do not paste credentials into a chat. Availability and account coverage may vary, so check the current Claude interface and its permission disclosures rather than assuming the beta supports every institution or plan.
Sources & further reading
Have a factual correction or a source to suggest? Contact the editorial desk.
