Bank of America warns customers against sharing logins with AI agents
Bank of America has told customers not to give their banking usernames or passwords to artificial intelligence tools. In an email headed “Protecting your finances in the AI World,” the financial institution said customers should never share credentials with another person or an outside service, including an AI-enabled tool or agent that offers to act on their behalf.
The message is a direct warning for customers experimenting with personal AI assistants. Bank of America said that, if credentials are shared, the customer remains responsible for activity resulting from that access. The advice applies even where an agent is presented as a convenience layer for carrying out tasks on a customer’s behalf.
Why AI access creates a banking risk
An AI agent with a username and password may be able to act within an account without the customer’s knowledge. Its activity could also appear suspicious to a bank, potentially prompting controls that lock the customer out of the account while the activity is reviewed.
There is a further exposure if the AI service has a security weakness. A flaw could leak credentials elsewhere, giving financially motivated attackers a route into the account. The warning therefore concerns both what an authorised tool might do and what could happen if the tool or service handling the credentials is compromised.
Banks are still defining agent access
The notice comes while banks, payment providers and processors are working through how AI agents could interact with customer accounts. One possible use case is agentic payments, where a person sets an agent to buy something autonomously at the best price or at a chosen time. The industry is also grappling with unauthorised agents accessing customers’ money.
That tension matters because a useful automated payment experience requires a way to distinguish customer-approved activity from access that should not be trusted. Bank of America’s email does not describe a customer-authorised method for handing an agent a conventional username and password; instead, it explicitly tells customers not to do so.
What customers and businesses should do
Customers should keep online-banking credentials out of AI prompts, connected assistant profiles and third-party agent workflows. They should use only access methods their bank explicitly supports, rather than treating a password as a reusable connection mechanism for an AI service.
For businesses deploying AI assistants, the practical implication is to design financial workflows around approved, bounded access rather than employee or customer logins. That reduces the chance that an agent is given unrestricted account access and helps keep responsibility aligned with the bank’s stated security rules.