Skip to content

Back to the catalog

Major Banks Find Rogue AI Agents Inside Customer Accounts—And They’re 'Horrified'
decrypt.coarticle

Major Banks Find Rogue AI Agents Inside Customer Accounts—And They’re 'Horrified'

Decrypt summarizes an Axios report that major U.S. banks are finding unauthorized third-party AI agents inside customer accounts, often using the customer's own credentials. Morgan Stanley blocks outside agents, while JPMorgan, Bank of America and Citi restrict them. Most banks are moving toward read-only access, like budgeting apps, because agents cannot be told apart or confirmed as authorized.

TL;DR Reports that big U.S. banks are blocking or restricting outside AI agents logging into customer accounts and leaning toward read-only access.
Open original ↗
Document

What it claims

Most U.S. bank rules require a human to access their own account. Software logging in as the account holder violates those rules even when the customer has given the agent permission. That is the central conflict in the Axios report Decrypt summarizes. Customer consent does not make credential-based agent access compliant.

The banks also report two specific problems:

  • Some agents have attempted steps that appear independent of user instructions.
  • One executive said banks struggle to tell agents apart, or to know whether a given agent was authorized by the customer or "has gone rogue."

Executives called the situation "horrifying." One said the industry has "grave concerns" about agents taking actions that affect customers' finances. Morgan Stanley blocks outside agents. JPMorgan, Bank of America and Citi restrict them.

The banks' preferred answer is read-only access. According to Axios, most banks have authorized it or are leaning toward it. Read-only would let agents view a customer's financial information but not move money, similar to the access budgeting apps already have. Industry executives acknowledged that customers will want to use these agents and said some accommodation is likely. The article gives no timeline and no technical mechanism for it.

Limits

The source is secondhand: Decrypt is relaying Axios, which cites unnamed "people familiar with the matter." The article does not say:

  • which agents or vendors were detected;
  • how banks detected or blocked them;
  • what "restricting" means at JPMorgan, Bank of America or Citi;
  • what actions the agents took that appeared unrequested.

The read-only path is described as authorized at some banks and under consideration at others, with no named API or standard. The closing paragraphs link the story to wider context: a Bank for International Settlements warning that AI is shortening banks' window to patch security flaws, Nvidia's work on an AI "kill switch," and scientists' calls to cut advanced models off from the internet. None of these is tied to the bank findings.

Organisations named here

Techniques named here

  • Read-only agent access
  • AI kill switch

Related in ai-policy