Anthropic Pilots 'Claude Money' to Let AI Access Personal Banking Data

Artificial intelligence firm Anthropic has begun field testing a new feature dubbed "Claude Money," which enables its Claude chatbot to link directly with users' bank accounts and other financial services. The beta program aims to let the AI analyze transaction histories, categorize spending, and answer questions about personal finances, effectively acting as a digital financial assistant.

Participants in the trial can authorize the connection through standard banking APIs, granting Claude read‑only access to account balances, recent transactions, and recurring payments. Once linked, the model can generate summaries such as monthly spending breakdowns, highlight unusual charges, or suggest budgeting adjustments based on the user's historical patterns.

Anthropic says the feature is built on the same safety and privacy principles that guide its broader AI products. Data is processed on secure servers, and the company asserts that no personal identifiers are stored beyond the session needed to fulfill a query. Users retain the ability to revoke access at any time, and the integration does not permit the AI to initiate transfers or modify account settings.

The move reflects a growing trend among AI developers to embed financial‑management capabilities into conversational agents. Competitors have introduced similar tools that can pull in banking data to answer queries about balances or upcoming bills. By offering a more nuanced analysis—such as identifying subscription churn or forecasting cash flow—Anthropic hopes to differentiate Claude as a more comprehensive personal finance aide.

Industry observers note that the rollout raises regulatory considerations, particularly around data protection and the handling of sensitive financial information. While Anthropic emphasizes compliance with existing banking security standards, regulators may scrutinize how AI models interpret and act on personal finance data. The company plans to expand the feature beyond the current test group later this year, pending user feedback and any necessary adjustments to privacy safeguards.