US banks tighten control over AI agents accessing customer accounts

Major US financial institutions have discovered external automated tools accessing customer accounts, according to a report by Axios published on Thursday, October 22. This issue has already impacted JPMorgan’s earnings communication, with CFO Jeremy Barnum stating on October 13 that the bank blocks services based on credential sharing and screen scraping whenever possible.

User authorization does not guarantee bank access

JPMorgan CEO Jamie Dimon said during the same conference call that the company uses agents internally but considers it risky to allow unrestricted access for external tools. Barnum added that the bank is working on a suitable way to enable these connections.

Who is liable when an agent makes a mistake?

Chase’s terms of use require agents to identify themselves as AI, flag automated traffic, and answer truthfully to questions designed to distinguish them from humans. The bank also reserves the right to limit, suspend, or block such access.

The digital services contract, updated in October, goes further: instructions sent by an agent with access provided by the account holder are treated as if they came from the customer themselves. The document assigns liability to whoever enables the tool for errors in interpretation or actions that exceed the granted authorization.

This is the bank’s contractual position, not a judicial conclusion about liability in any potential dispute. The risk of a tool acting unexpectedly has already spurred businesses focused on monitoring and insuring against agent failures.

Consulting data is not the same as moving money

There are integrations designed for more limited functions. Plaid describes Perplexity’s Personal CFO connection as read-only access, where the user chooses which data to share, and the tool uses it for financial analysis and dashboards.

This differs from granting an agent permission to perform operations within an account. In Brazil, an integration by Belvo with banking data via Open Finance, also based on user consent, has already been publicly presented.

Restricting external tools does not mean abandoning technology. Charles Schwab announced in September the launch of Charley, its own assistant, with a gradual rollout starting in October, to consult information and perform certain tasks. The company warns that results may be inaccurate and advises customers to verify that requested changes were made correctly.

The banks’ rules and warnings reflect concerns about control and liability. They do not, in themselves, prove that money has been stolen or that improper transfers have been made by these systems.

Frequently Asked Questions

Can a bank block an agent authorized by a customer?

Chase’s terms reserve the right for the bank to restrict or prevent access by these tools.

What does Chase require from agents?

Identification as AI, flagging of automated traffic, and truthful answers about the nature of the interaction.

Does a read-only connection allow for moving money?

The Personal CFO integration described by Plaid allows for consulting information, not executing banking operations.

Have institutions stopped using AI?

No. JPMorgan and Schwab also use or develop their own tools.