US banks start blocking AI agents in customer accounts

Axios reported on 22 October that big US banks found unvetted AI agents logging into customer accounts with customers' own credentials. Morgan Stanley began blocking third-party agents; JPMorgan, Bank of America and Citi are restricting them, mostly to read-only. Wells Fargo has also acted and Charles Schwab's CEO confirmed it turns agents away.

One executive called the situation "horrifying". Morgan Stanley said it has not seen rogue agents on its platforms. Chase reminded customers never to share credentials, and Jamie Dimon said JPMorgan will not let just any outside agent in. Agents may read data via connectors such as Plaid but cannot pay bills. No breach of core systems has been reported.

The core concern is liability. If an agent makes a wrong payment or is hijacked, it is unclear who bears the loss, and because the customer shared the password the bank may be in a weak position in a dispute. Fraud systems may also flag agent behaviour as suspicious.

Walmart opened an official agent gateway with spending limits and no access to payment data, while Amazon blocks shopping agents. Claude Duet asks before paying, and 1Password for Agents signs agents in without revealing passwords.

When customers share passwords, access is technically legitimate, but the bank cannot tell a person from software. Know Your Agent adapts KYC so agents connect with their own scoped identity.

That identity would show the bank what the agent may do, for how long and on whose behalf.

Expect banks to open agent-specific APIs and consent screens rather than tolerate password sharing.

Payment networks are taking a different route: reports the same week said Visa is letting agents execute approved supplier payments, opening a controlled door rather than closing it.

• Never give an agent your bank password; use official connections.
• Separate reading from paying; approve payments yourself.
• Keep critical payments off agent automation in case access is blocked.