Claude Gets Your Bank Details, Not Your Bank Keys

Anthropic announced a feature on November 5 that lets Claude analyse users' bank and card records, bringing sensitive financial data into conversations about spending, bills and affordability. ^[1][2]

Personal Finance is launching as a US beta for Pro, Max and Ultra subscribers, with a rollout scheduled for the week of the announcement. Users connect their accounts through Plaid, the service that links financial institutions to apps, then ask Claude questions in everyday language. ^[1][2]

Anthropic says the connection is read-only. Claude cannot move, manage or invest money, and the feature is not designed to provide investment advice. It can, however, receive balances, credit limits and up to two years of transactions, giving the chatbot a detailed view of a user's finances. ^[1][2]

That is a substantial change from asking a chatbot how to build a budget. Claude can now work from the records of what a person actually earns, owes and spends. The permission deserves scrutiny even when it gives the software no power to make a payment.

What Claude is offering

Anthropic describes tasks including sorting spending into categories, working out which bills remain before payday and considering whether a purchase is affordable. It also says Claude can help identify possible duplicate charges, unexpected fees and bills that have risen. These are the company's examples of what the feature can do, rather than independently verified results. ^[1][2]

The attraction is straightforward: a user can ask a question across connected records instead of inspecting transactions one by one. A question about an expensive month could become a conversation about where spending changed. That is the practical promise behind bringing a language model into personal finance.

But the questions vary in difficulty. Finding two similar charges requires comparing records. Deciding whether a purchase is affordable requires a judgement about the money available and the obligations still to come. A fluent answer to the second question deserves more scrutiny than a list of transactions.

Anthropic says the feature has no ads or sponsored recommendations and offers no financial products of its own. ^[2] If those commitments hold, they would remove an obvious potential conflict: steering someone from a discussion of their finances towards a product sold within the service. They do not establish whether its calculations or recommendations are sound.

OpenAI introduced a similar Plaid-connected personal-finance feature for ChatGPT in May 2026. Anthropic is entering an existing contest to make the chatbot a place where users examine their own money, rather than merely ask general questions about it. ^[1][2]

What the connection exposes

According to Anthropic, the information available through Personal Finance may include the bank's name, the account's name and type, the last four digits of its number, balances, credit limits and transaction history covering up to two years. ^[1][2]

A balance is a snapshot. A long transaction history can reveal recurring commitments and patterns of spending. Read-only access protects against one class of risk, unauthorised movement of money, while still allowing extensive visibility into financial life.

Claude does not see banking passwords when Plaid handles the connection, according to The Verge. Plaid's own help material says the information it can access depends on the product and permissions involved. Its general list includes account-holder contact details, transaction descriptions, balances, account numbers and routing numbers. ^[1][3]

That list is broader than the information Anthropic describes for Claude. It should not be read as proof that every category reaches the chatbot. Plaid also says that information accessible through one set of bank credentials may cover multiple accounts, making the scope of an authorisation worth checking carefully. ^[3]

Before connecting, users should inspect which accounts and categories of information the permission screen includes. The useful question is specific: what will this connection share? Knowing that Plaid provides the link does not answer it.

A training promise is not a deletion policy

Anthropic says conversations within Claude's Personal Finance hub are not used to train its models. That is a feature-specific assurance, separate from the company's broader consumer-data policy. ^[2]

In its August 28, 2025 consumer-policy announcement, Anthropic said Free, Pro and Max users could choose whether their data helped improve its models. It described retention of up to five years when users allowed training and 30 days when they did not. Those general terms do not explain how retention works for information connected through Personal Finance. ^[4]

The launch leaves several practical questions unanswered: the precise permission and account-selection controls, retention and deletion rules for connected records and conversations, whether disconnecting removes information already shared, and whether the no-training assurance covers all transaction data as well as hub chats. Supported institutions and independent accuracy results are also not established. ^[1][2][3][4]

A promise about training describes one use of information. Users also need to know how long it stays and what happens when they leave.

For anyone trying the beta, Claude's answers should be treated as analysis to check against the underlying records. A possible duplicate charge is a reason to inspect the transactions, not a finding that the bank charged twice. An affordability answer should be checked against obligations the conversation may not capture.

Anthropic's immediate task is the US rollout to eligible paying subscribers. ^[1][2] For users reaching the connection screen, the decision comes earlier than the first clever answer: whether the help on offer is worth sharing the history behind it.