The next artificial-intelligence security fight may be happening inside your bank account. Major U.S. financial institutions are restricting third-party AI agents after reports that automated assistants have entered customer accounts—and, in some cases, tried to take actions involving money.
The concern is not necessarily that hackers broke through bank defenses. In many cases, customers reportedly gave an AI assistant their own login credentials and permission to complete a task. The bank, however, did not authorize the software, may not know exactly what the agent was told to do and may struggle to distinguish an automated action from the customer’s own click.
Which banks are tightening access?
Axios reported Thursday that Morgan Stanley, Bank of America, JPMorgan Chase, Wells Fargo and other institutions are racing to contain the risks posed by unapproved AI agents. The report, based largely on people familiar with the matter, said some banks have observed agents contacting call centers or appearing to take steps that did not clearly match a user’s instructions.
Important caution: Reuters said it could not independently verify those reported incidents. Morgan Stanley told Axios that it had not observed “rogue AI agents” accessing its platforms, while a Chase spokesperson advised customers not to give their usernames and passwords to anyone.
The broader access crackdown is better established. The Financial Times has reported restrictions at Citi, Bank of America and JPMorgan. Morgan Stanley has begun blocking third-party agents, according to Axios. Some institutions are considering read-only access that would let approved tools retrieve information without authorizing transfers or other account changes.
Schwab’s CEO confirms the money-moving concern
Charles Schwab CEO Rick Wurster publicly confirmed that his company is blocking unauthorized AI agents that attempt to move clients’ money. In comments reported by Bloomberg Law, Wurster said the firm does not want customers handing an assistant their password and allowing it to act when neither the bank nor the customer can fully control the result.
That statement highlights the central problem: permission is not the same as control. A customer may ask an agent to pay one bill or move a limited amount. The bank must still determine what authority the agent actually has, whether its action matches the request and who is responsible if it makes a costly mistake.
Why an “unauthorized” agent may still have your permission
In this context, unauthorized often means unauthorized by the financial institution—not necessarily secretly installed by a criminal. A customer may voluntarily share credentials with an assistant that promises to compare spending, negotiate bills or make purchases. That can still violate the bank’s rules requiring the account holder to log in personally.
Traditional financial apps typically connect through approved interfaces with defined permissions. An agent using a person’s username, password and browser session can look more like the customer. It may also gain broader access than necessary, including the ability to view balances, change settings or initiate payments.
What customers should do now
- Do not give a general-purpose AI assistant your bank username, password, PIN or one-time verification code.
- Use only connections and financial tools that your bank explicitly supports.
- Review connected apps and devices in your bank’s security settings and remove anything unfamiliar or no longer needed.
- Turn on transaction alerts and multifactor authentication. An authenticator app or security key is generally stronger than a text-message code when the bank offers it.
- If credentials were shared with an unapproved service, change the password and contact the bank through the number in its official app, website or statement.
The Federal Trade Commission recommends strong, unique passwords and two-factor authentication. It also warns consumers never to share verification codes. Those basic protections become even more important when software can operate a browser and complete multi-step tasks on a user’s behalf.
The technology is moving faster than the rules
Banks recognize that customers will want AI assistants capable of organizing finances, comparing products and handling routine tasks. The question is how to offer those conveniences without handing an outside system unlimited access to deposits, investments and personal data.
For now, the safest line is simple: let AI explain your statement if you manually remove sensitive information, but do not give an unapproved assistant the keys to the account itself. Until banks establish clear permission controls and liability rules, convenience can quickly become exposure.
Sources: Axios; Reuters; Bloomberg Law; Federal Trade Commission.
