Bank of America kindly asks that you don’t share your banking password with your AI assistant

Bank of America has sent an email to customers titled "Protecting your finances in the Al World," asking them not to hand their banking username and password to anyone else, and specifically to avoid giving them to an AI-enabled tool or agent that offers to act on their behalf. The message carried a blunt reminder attached to that request: if a customer does share credentials, the bank said, they should keep in mind that they are responsible for activity resulting from that access. TechCrunch reported the email blast, and Bank of America did not immediately respond to a request for comment about it.
The warning lands in a moment when personal AI assistants are becoming a normal part of how people interact with online services. Assistants that can log into websites, read pages and take actions on a user's behalf turn a password into something closer to a standing authorization. Once an agent holds working credentials, it can move money, change account details or open new products with the same authority as the account holder, and often without a human watching each step. The bank's framing reflects that: sharing a login is treated as delegating authority, not as a one-off convenience.
The risks the bank's email points at fall into a few categories. An assistant holding credentials could take an action on someone's finances without that person's knowledge. A bank's fraud systems might see unusual automated behavior and conclude that an intruder is raiding the account, which can lead to a lockout. And a security flaw in the assistant itself could expose the credentials somewhere else, handing financially motivated attackers a direct route into the account. In practice, these failure modes compound: an agent that logs in from new locations at odd hours can look identical to an account takeover, so even benign automation may trip defenses designed for human behavior.
Most banks' account agreements make customers responsible for anyone they give their login to, which is the position Bank of America restated. This is a long-standing principle in retail banking rather than a new AI-specific rule. It means that if an agent misbehaves, the customer may struggle to argue the activity was unauthorized, because the credentials were legitimately handed over. For developers building assistants, that shifts the burden of safety onto the product: a tool that stores a bank password in plaintext, reuses it across sites or lets a model see it in a prompt is creating liability the user may not fully understand.
The broader industry is still working out how agents should fit into payments and banking. Banks, payment providers and processors are reported to be grappling with unauthorized agents reaching customers' money, and the question of how to grant AI tools access to accounts is unresolved. That includes agentic payments, where a person sets an agent to buy something autonomously at the best price or the right moment. The technical answer many teams are exploring is to avoid handing over a password at all and instead use scoped, revocable permissions — OAuth-style tokens, spending limits, per-transaction approval and clear audit trails. This suggests the near-term path for agentic finance runs through open banking interfaces and card network tokenization rather than shared logins, though the source material does not describe any specific Bank of America product or policy to that effect.
A likely trade-off is friction. Scoped access requires merchants, banks and agent builders to agree on standards, and users to consent to each connection, which is slower than pasting a password into a chat window. Until those standards are common, some users will improvise, and the improvised path is exactly what the bank's email warns against. The instruction to keep credentials private sits awkwardly against assistants marketed on the promise of doing tasks for you end to end; the gap between that promise and the security model is where the warnings are coming from.
Why it matters: Anyone building consumer AI agents that touch money, and the product teams that market them, should expect banks to treat shared credentials as user-authorized activity rather than fraud. That raises the stakes on secure credential handling, consent and clear disclosure of what an agent can do. My reading is that the durable approach is delegated, revocable access with limits and logging, so agents can act without holding the keys to an account — and teams that skip that will carry both user trust and liability risk.
Based on reporting from the original publisher. Visit the source for full context and later updates.
Publisher excerpt
The banking giant reminds customers that "you're responsible for activity" resulting from your AI agents' access.