US banks tighten controls on AI agents

Several major US banks are limiting third-party AI agents connecting to customer accounts. The issue: difficulty distinguishing authorized actions from overly autonomous bots.

In the United States, JPMorgan Chase, Bank of America, Wells Fargo, Morgan Stanley, and Citigroup are tightening access to certain third-party AI tools on customer accounts, according to Axios on October 22. The issue: autonomous agents capable of connecting with credentials provided by the user, then going further than expected. For banks, the risk is twofold: account security and liability in case of an action triggered by an overzealous machine.

Quand un bot entre sur le compte, la ligne devient floue

The issue is not classic hacking, but the use of AI assistants acting on behalf of a client after receiving their credentials. Some reportedly contacted call centers or attempted to perform tasks not explicitly requested, according to sources cited by Axios. The problem for banks boils down to a simple and frankly annoying question: how to verify that an operation really comes from a human, and not from a software improvising?

This confusion explains the mistrust towards password sharing and screen scraping, a practice where software retrieves data directly from online banking pages. JPMorgan Chase said it does not want to give external agents total freedom. Its CEO, Jamie Dimon, mentioned risks for data, assets, and legal troubles if these tools go off track. Jeremy Barnum, CFO, also pointed out methods considered unsafe. The message is clear: "sharing passwords and screen scraping" no longer go as smoothly as before.

La solution qui monte : lire sans toucher

Rather than a total block, several institutions are moving towards read-only access. In this model, an agent can view a balance, analyze expenses, or prepare a budget, but cannot transfer money or initiate a payment. It is the same logic as some already authorized financial tools, like Quicken, which aggregate information without taking control of the account.

Axios indicates that this approach is gaining growing support among major US banks. Morgan Stanley has begun blocking certain third-party agents, while Bank of America and Citigroup are also tightening their rules. A Morgan Stanley spokesperson, however, said no "rogue" agents have been observed on its platform. The nuance matters: at this stage, restrictions do not prove massive incidents, only that banks prefer to close the door before having to chase after the handle.

Les banques veulent de l’IA, mais sous contrôle

Institutions do not reject AI outright. They seek to define a boundary between a useful assistant and an agent acting too freely. In personal finance, the promise is appealing: software could compare rates, track expenses, or help move deposits to better yields. But once it can move money, the discussion changes nature.

The Consumer Financial Protection Bureau recommends customers verify what a financial service can do with their data before connecting it to an account. This includes the ability to make payments, the frequency of data access, and possible sharing with other companies. And deleting an app from the phone does not necessarily cut its access to the bank account. For users, vigilance remains essential. For banks, the real battle begins: keeping useful uses without handing the wheel to just any bot.

Ultimately, the issue goes beyond security alone. It also touches competition, as personal assistants could eventually play banks against each other to obtain better rates. In this context, institutions are trying to define safeguards before AI agents become unavoidable intermediaries in daily financial life.

Key points

  • Axios published the alert on October 22.
  • JPMorgan Chase, Bank of America, and Citi are tightening rules.
  • Morgan Stanley says it has seen no rogue agents.
  • The read-only model limits money transfers.
  • The CFPB advises verifying data usage.
  • Screen scraping remains a contested practice.

By the numbers

  • 5 banks — JPMorgan Chase, Bank of America, Wells Fargo, Morgan Stanley, and Citigroup, according to Axios.
  • October 22 — date of Axios report publication.
  • 1 access mode — read-only, presented as a compromise by banks.

To read