NEW YORK—Major U.S. banks are detecting unauthorized third-party artificial intelligence agents operating inside customer accounts, including some that appear to be taking actions independent of users’ instructions, according to new report.
The developments are raising significant security concerns at some of the nation’s largest financial institutions, which are struggling to distinguish between legitimate customer activity and automated agents that may have access to account credentials and the ability to move money. As the CU Daily reported with its coverage from the Money 20/20 conference in Las Vegas this week, related issues around authenticity, identificaiton and agentic AI were extensively discussed during the four-day event.
According to Axios, banks including Morgan Stanley, Bank of America, JPMorgan Chase, Wells Fargo and others are moving to restrict access by third-party AI agents as the technology advances faster than institutions can develop safeguards.
How AI Agents Are Accessing Accounts
Axios reported that many of the incidents involve AI assistants or agents accessing bank accounts with customers’ permission and login credentials.
Although customers may have authorized the access, the activity can violate banks’ terms of service, which generally require customers to log in to their own accounts rather than share credentials with third parties.
The risks become particularly significant when customers give AI agents permission to make purchases, transfer funds or perform other financial transactions.
In other cases, banks have detected agents contacting customer service call centers or apparently taking actions beyond the instructions provided by account holders, according to people familiar with the incidents who spoke with Axios.
Bank executives described the situation as alarming, with one official telling Axios the industry has serious concerns about AI agents gaining access to customer accounts and potentially taking actions that could affect customers’ finances.
Banks Moving to Restrict Access
According to Axios, several major banks are taking steps to limit the technology:
- Morgan Stanley: Has begun blocking third-party AI agents from accessing its platforms, according to sources familiar with the decision.
- JPMorgan Chase: Is restricting access by third-party agents and warning about the risks associated with customers sharing account credentials.
- Bank of America: Is among the institutions moving to restrict AI agent access.
- Citigroup: Is also implementing restrictions on third-party agents.
A Chase spokesperson told Axios that customers should not share their usernames and passwords with anyone.
A Morgan Stanley spokesperson said the company had not observed instances of rogue AI agents accessing its platforms.
Difficulty Distinguishing Humans From AI
One of the biggest challenges facing banks is determining whether account activity is being performed by a human customer, an authorized AI assistant or an agent operating without permission.
Axios reported that several banking executives acknowledged their institutions cannot reliably make those distinctions.
One executive said banks are struggling not only to identify automated agents but also to determine whether an agent is following customer instructions or has begun acting independently.
That uncertainty is prompting financial institutions to rewrite terms of service, strengthen authentication requirements and restrict certain forms of automated account access.
Read-Only Access Emerging as Compromise
Despite the concerns, banking executives recognize that customers are likely to increasingly demand AI-powered financial services.
According to Axios, many banks already permit, or are moving toward permitting, read-only access for certain third-party applications and AI tools.
Such access would allow an AI assistant to review account balances, transactions and other financial information without having the authority to transfer funds, make purchases or initiate other transactions.
Banks already provide similar access to financial management applications such as Quicken, which customers use for budgeting and tracking spending.
Industry executives told Axios that finding a balance between customer demand for AI services and account security will be increasingly important.
For now, however, several institutions believe the risks of allowing AI agents broader account access outweigh the potential benefits.
Preparing for Worst-Case Scenarios
According to Axios, banks are investing in defenses intended to prevent unauthorized AI activity while also developing contingency plans for situations in which those safeguards fail.
Those preparations include what sources described as “day after” planning, designed to help institutions quickly secure compromised accounts and restore customer access following an incident.
The report noted that large banks have substantial financial resources and extensive experience protecting critical financial infrastructure, potentially positioning them to respond to AI-related threats.
But one source cautioned that financial resources alone may not be sufficient because AI technology is advancing rapidly and exposing new vulnerabilities faster than institutions can address them.
The central challenge, according to Axios, is that AI agents are gaining the ability to perform increasingly complex financial tasks at a time when banks still lack reliable methods to determine whether those actions are authorized, safe or consistent with customers’ intentions.
