Big U.S. banks are finding unapproved AI agents working inside customer accounts, and some of those bots seem to be acting without the user telling them to, according to Axios. Usually the user hands their login to an AI assistant so it can handle tasks for them. Most banks' rules say only the actual human should log in, though. Executives worry about what a bot could do if it has permission to buy things or move money. Some agents have even been calling bank call centers on their own.

Banks are moving fast to clamp down. According to Axios, Morgan Stanley has started blocking third-party agents, and JPMorgan, Bank of America and Citi are restricting them too. A Chase spokesperson reminded customers not to share their username and password with anyone. Morgan Stanley said it hasn't seen any "rogue AI agents" on its platforms. The tricky part is that many banks can't yet tell whether a bot is doing what its user asked or has gone off script.

For now, most banks are leaning toward read-only access, which lets an AI see your account info without touching your money, similar to budgeting apps like Quicken. Bank leaders admit customers will want these tools, so a middle ground is likely eventually. In the meantime they're planning for worst-case "day after" scenarios so they can quickly lock down or restore accounts if something goes wrong. The tech is moving faster than the safeguards.