AI · Oct 19, 2026
Microsoft ships Guardian Agents, saying frontier models ‘can’t be trusted’Claude gets read-only access to your bank accounts with Personal Finance, pitched as ad-free and unconflicted
· Updated · Signal Desk newsroom
Analysis of reporting by TechCrunch. How we work

Anthropic’s beta lets US Pro, Max and Ultra users link bank and card accounts via Plaid so Claude can answer questions about balances and transactions. It can’t move money, carries no ads or Anthropic financial products, and hub chats stay out of training.
Key takeaways
- Personal Finance is in beta for US Pro, Max and Ultra users, connecting via Plaid.
- Access is read-only: no moving, managing or investing money.
- Data is limited to account basics, balances, credit limits and up to two years of transactions.
- No ads, sponsored recommendations or Anthropic credit cards, loans or savings accounts.
- Chats in the finance hub are excluded from training; ChatGPT shipped a similar feature in May.
What launched
Money talk is where assistants become useful or become a privacy incident. Anthropic is rolling out Personal Finance in beta for Pro, Max and Ultra users in the US, letting Claude connect to bank and credit-card accounts through Plaid and answer questions about balances, transactions and categories. Access is read-only, with no moving, managing or investing money. Anthropic stresses no ads or sponsored recommendations and no Anthropic credit cards, loans or savings accounts, so answers stay unconflicted. ChatGPT shipped a similar connect-your-accounts experience in May, so this is Claude catching up in a high-trust category while protecting its no-ads brand, the same week WIRED chronicled the industry’s rush into chatbot ads (chatbot ads).
The data boundary
- Fields
- Bank name, account name and type, last four digits, balances, credit limits and up to two years of transactions.
- Permissions
- Read-only through Plaid.
- Training
- Chats in the finance hub are excluded from training.
- Conflicts
- No ads, placements or Anthropic financial products.
Plain speech on your books
The pitch: where did the money go, is this more than usual, flag duplicate charges and creeping fees, explain APR with a personal interest total. Narrow defaults are the minimum adult posture for wiring a model into a ledger, consistent with the restraint Anthropic built into its Duet agent, which asks approval before payments or deletions (Claude Duet). Anthropic’s push into legal and financial services shows it wants professional money workflows; Personal Finance is the consumer wedge.
Catching up to ChatGPT matters less than whether Claude stays boringly trustworthy as the category gets lucrative. If Plaid connections normalize faster than user understanding, expect phishing lookalikes, oversharing of two-year histories and pressure to add write actions. Keeping it read-only until the security and UX stories are dull is the right call. More in our Anthropic coverage and finance coverage.
Who feels it
- US Claude paid users
- Everyday-language questions about real spending.
- Privacy-conscious consumers
- Clearer boundaries than many finance apps, still a real account-link decision.
- OpenAI
- Parity pressure on finance UX and trust messaging.
- Banks and fintechs
- Another large AI surface on Plaid rails.
What to watch
- Whether write actions and placements stay off the table.
- Incident response to any connected-account bug or prompt injection.
- Expansion beyond US paid tiers.
- Categorization quality versus ChatGPT’s finance tools.
FAQ
Can Claude access my bank account?
With Personal Finance, US Pro, Max and Ultra users can link accounts via Plaid for read-only access.
Can Claude move my money?
No. Access is read-only, with no moving, managing or investing.
Is my Claude finance data used for training?
Anthropic says chats in the Personal Finance hub are excluded from training.
Companies: Anthropic
More on Anthropic
AI · Oct 20, 2026
Anthropic’s draft federal AI bill would let a judge halt a model release over mass-casualty riskAI · Nov 5, 2026
Perplexity Teammates bets that always-on work agents are worth $200 a monthAI · Nov 10, 2026
New York’s RAISE Act gives frontier labs a 72-hour clock for reporting dangerous AI incidentsAI · Nov 13, 2026
A week with Claude Duet: good at catching what falls between apps, shaky on tone and urgency