Major U.S. banks warn that external AI agents are accessing customer accounts, with some taking unauthorized actions, prompting them to block or restrict access by third-party artificial intelligence tools.
According to an Axios investigation, financial institutions have raised concerns about the difficulty of distinguishing customer actions from those taken by AI agents, prompting them to review their terms of service.
AI agents taking actions banking customers have not requested
The report warns that some artificial intelligence agents could contact call centers and take actions their customers had not requested.
However, the activity did not involve hackers compromising customer accounts, as the AI agents gained access after customers provided them with their account login credentials to help manage finances or complete certain tasks.
Additionally, some AI tools obtain credentials through scraping customer banking interfaces, a risky method that some banks block to protect customer information.
The report warns that banks increasingly cannot distinguish human actions from those performed by artificial intelligence tools.
Consequently, Bank of America, JPMorgan Chase, Morgan Stanley, and Wells Fargo have blocked or restricted AI agents from accessing customers’ accounts.
Chase also discourages customers from disclosing their account login credentials to third parties, including AI agents. Nevertheless, Morgan Stanley admits that it has not observed rogue artificial intelligence tools accessing customer financial accounts.
Nevertheless, many banks worry they cannot keep up with the pace of AI development, prompting them to take precautionary measures. They also worry that granting these AI tools access to customer accounts poses a real-time risk, which could be difficult to predict or address.
Nevertheless, banks already permit various third-party financial management tools, such as Quicken, to access customer account data and offer services such as budgeting and expense tracking. They also deploy chatbots to assist customers with banking services. However, unlike autonomous AI agents, these productivity tools are limited to predefined tasks.
Beyond customer-authorized AI agents, financial institutions have also faced automated attacks by AI-powered autonomous tools.
In October 2026, several South Korean banks were targeted using an AI tool, exposing the personal information of over 140,000 customers. The cyber attack involved ARTEX AI, an open-source, Chinese AI-powered penetration-testing tool that could enable low-skilled hackers to breach into computer systems.
“There is no ‘machine gone rogue’ in this story,” said Ted Miracco, CEO, Approov. “There are only irresponsible people handing banking credentials to AI software, and vendors shipping agents that ask for them. Both are human decisions, and both should carry the financial consequences; however, right now neither clearly does.”
“We are building agents to impersonate customers and then using shared credentials which means a bank cannot tell an authorized agent from an attacker using the exact same login. That is more of an identity failure than an AI failure. Until agents carry verifiable identities, scoped permissions and an audit trail, blocking them is the best defensible control a bank has,” added Miracco.
Banks propose a middle ground on AI agents
To address the risk posed by AI agents, some major U.S. banks have revised their terms of service to prohibit customers from allowing third-party artificial intelligence tools to access their financial accounts.
However, most U.S. banks also propose a middle ground on AI agents and recognize customers’ need to use artificial intelligence tools in the future.
Some have also resorted to granting read-only access to customer accounts, allowing AI agents to view customers’ financial statements without the ability to take unauthorized actions, such as moving money.
Additionally, some are developing internal agents and are open to granting access to third-party artificial intelligence tools in the future. They are also considering developing secure channels that would let AI agents connect to their banking platforms safely.
