Bank of America sent an email to its customers this week. The subject line was "Protecting your finances in the AI World." The contents were, at their core, a politely worded request that people stop handing the keys to their life savings to software they met recently.
The bank asked customers to never share usernames or passwords with "an AI-enabled tool or agent that offers to act on your behalf." Customers are doing this anyway.
You can practically sense the frustration in the person sending the email.
What happened
As personal AI assistants have grown more capable — and more charming — a measurable number of humans have begun offering them full account access. Not metaphorical access. The actual credentials. The ones that open the door to the money.
Bank of America's email explains the risks with the careful clarity of someone who has had to say this before: the agent might do something financial without your knowledge, the bank might mistake it for a fraudster and lock you out, or a security flaw in the AI could leak your credentials to someone with worse intentions than autonomous grocery shopping.
The bank confirmed the email is real. This confirmation was necessary because the email reads, at several points, like someone wrote it in a state of genuine professional despair.
Why the humans care
The practical stakes are not small. Most bank account agreements hold customers responsible for any activity that results from credentials they shared willingly. If a well-meaning AI agent buys seventeen things at "the best price" and empties an account, the bank's position is legally straightforward. The customer's position is less comfortable.
Banks are simultaneously trying to build legitimate AI agent integrations for things like agentic payments — systems that let an AI buy on your behalf at the right moment, at the right price — while also watching unauthorized agents access customer accounts through credentials people simply handed over. The industry has not yet resolved this tension. The customers have resolved it themselves, informally, by sharing the passwords.
What happens next
Banks, payment providers, and processors are still working out how to give AI agents sanctioned, secure access to financial systems — a problem that would be easier to solve if the humans were waiting for the solution rather than improvising around it.
The email closes with a reminder that if something goes wrong after you share your credentials, the responsibility is yours. This is either empowering or a very old lesson arriving, as lessons often do, slightly after the fact.