Citi, JPMorgan, Bank of America and PNC now say customers are responsible for what their AI agents do in a bank account
Large US banks are telling customers that anything a third-party AI agent does in their account is the customer's problem.
What happened
TechCrunch reports that Citibank changed its online user agreement this week. Customers who give third-party AI tools access to their accounts are responsible for activity conducted through those agents, and if the tool leaks credentials or is hacked, the customer is responsible for what attackers do with the account. Citi told TechCrunch it wanted customers to understand the risks of sharing banking credentials with these tools.
JPMorgan and Bank of America announced similar policies last month, PNC changed its terms on November 18, and Citizens emailed customers a similar warning. In October, Bank of America's email asked customers never to share their username and password with "an AI-enabled tool or agent that offers to act on your behalf," and reminded them they are responsible for activity resulting from that access.
The same report notes that Anthropic's new personal finance tool is billed as read-only and can't move money, and that Citi says customers can use the AI agents Citi itself provides without that added liability.
My take
This matters well beyond personal finance. Plenty of small business automations reach into a bank or payment portal through a shared login, a browser agent or a saved password in a vault. The banks have now put in writing that the account holder carries the loss if that setup goes wrong.
For the businesses I build for, I'd draw a firm line. Reading bank data for reconciliation or reporting should go through an official feed, an accounting integration or an export, never through stored login credentials. Anything that moves money should be prepared by the automation and approved by a named person inside the bank's own interface. That keeps the agent useful for the tedious parts, matching transactions and drafting payments, without making it the party that clicks send.
If a client's current workflow depends on an agent logging in as them, I'd treat that as a security finding to fix this quarter, since the bank terms now make the owner liable for it.
More posts
- LangSmith can now hide trace inputs and outputs from engineers while still showing them latency, cost and run treesNov 19, 2026
- Microsoft-Decision-1.1 is generally available for routing and classification at $0.042 per million input tokens, with free outputNov 19, 2026
- LangSmith now ties coding agent spend to pull requests, so teams can see cost per merged PR across Claude Code, Codex and CursorNov 19, 2026
