Bank of America just told customers in writing that handing banking credentials to an AI assistant makes the account holder personally liable for anything the agent does. If you are building AI automations that touch money - payments, invoicing, reconciliation - do not use a human login as the agent's entry point; use scoped API keys, bank-approved integrations, or a human-in-the-loop approval step.
The bank sent an email titled "Protecting your finances in the AI World" warning customers never to share username and password with any outside service, including AI tools acting on their behalf, as TechCrunch reports. The account agreement is explicit: if you share credentials, you own the outcome - including a frozen account when fraud systems see an unknown agent logging in from a strange location.
What this means for small business automations
This kills the quick hack of letting an agent "just log in" to online banking to pull statements or push payments. The clean pattern is boring but holds up: pull data via accounting software APIs (QuickBooks, Xero, 1C) that already carry a bank feed, keep the agent read-only wherever possible, and gate every outbound transaction behind a human click - the same guardrails I wire into finance-adjacent flows, including a personal assistant that handles routine tasks without ever touching raw login secrets.
Why it matters: If your automation plan assumes an AI agent will type passwords into a bank portal, the liability sits entirely on you - redesign the integration around proper APIs before you build.
Source: TechCrunch