Bank of America to customers: If your AI agent loses your money, you pay
Bank of America is warning customers not to share passwords with AI agents, stating users will be held liable for any autonomous mistakes.
Business · Source: TechCrunch
What happened
Bank of America sent a mass email to its users titled Protecting your finances in the Al World. The financial giant explicitly warned customers never to share their usernames or passwords with outside services. This specifically includes any AI-enabled tools or agents offering to act on their behalf. A bank spokesperson later confirmed the authenticity of the email after it surfaced on social media. The message is clear. Banks are watching the rise of AI assistants and they do not like what they see.
The core issue is a massive shift in fraud liability. Under standard online banking agreements, customers generally cannot authorize a third party to use their credentials. The only narrow exceptions are legal guardianship or power of attorney. If a customer voluntarily hands over their login details to an AI agent, they break this agreement. The bank holds the customer entirely responsible for any activity resulting from that access. If the AI misinterprets an instruction and wires money to the wrong place, the customer pays the price. The liability holds until the bank is notified and the credentials are blocked.
Beyond liability, the bank flagged severe security and operational risks. Handing raw credentials to a third-party AI service creates a massive data-control problem. Users cannot control how much account information the service scrapes. If the AI startup suffers a security breach, financially motivated hackers get direct access to user bank accounts. Furthermore, automated agent activity looks exactly like a cyberattack. Bank of America warned that autonomous account behavior could trigger internal fraud controls. The bank might assume someone is raiding the account and lock the customer out entirely.
Key facts
- Protecting your finances in the Al World — Title of the warning email Bank of America sent to customers
Why it matters
We are entering the era of agentic payments, but the underlying financial infrastructure is completely unprepared. Builders are rushing to create autonomous agents that can book flights, manage calendars, and buy items at the best price. To make this work, many startups are taking shortcuts by asking users for raw banking credentials. Banks are now pushing back hard against this practice. They are actively grappling with unauthorized agents accessing customer money and they refuse to absorb the associated risks. The friction between legacy banking security and modern AI capabilities is reaching a breaking point.
This dynamic creates a hard ceiling for consumer AI agents. A budgeting app logging into your account to read data is one thing. An autonomous AI agent operating with those same credentials is a completely different threat model. Agents can initiate unintended transactions or execute moves that are incredibly difficult to reverse. Until the broader industry settles on comprehensive standards for agent authentication and transaction limits, autonomous financial tasks remain a legal minefield. If everyday users realize they are personally on the hook for every mistake an agent makes, consumer adoption for high-stakes AI tools will stall completely.
For builders
Stop asking for raw banking credentials
Do not build products that require users to hand over their bank passwords. You expose your users to massive financial liability and risk getting their accounts locked by bank fraud controls. You must use official consent-based connection processes where users can safely grant and revoke specific permissions.
Agentic payments lack a liability framework
The financial industry has no settled standard for agent authentication or transaction limits. Startups building autonomous purchasing tools carry massive execution risk. Until payment networks and banks establish clear rules, you are building on top of a legal gray area.
Security breaches will destroy your startup
If you store raw banking credentials and suffer a breach, you are handing hackers direct access to user bank accounts. The fallout from exposing financial data will instantly kill your company. Never store raw passwords.
My take
Builders need to stop treating bank accounts like open APIs. Asking users for raw passwords to power your AI agent is lazy engineering and terrible product design. You are shifting all the risk onto your user just to ship faster. If you cannot build financial agents using secure consent-based connections, you have no business building them at all. Fix your architecture before a bank sues you into oblivion.
Original reporting: TechCrunch. This is my rewrite and opinion.