In a quiet corner of the financial system, a new kind of visitor has begun appearing: software agents that customers have invited to act on their behalf, slipping through the front door with borrowed passwords. Major American banks have noticed these digital helpers moving through accounts their owners never personally opened to them, and a reporting effort at Axios has brought the pattern into public view. The deeper question is one people have always faced when handing their keys to someone else: how much trust a helper deserves, and whether acting on your behalf is the same as acting withou
Banks Detect Unvetted AI Agents Accessing Customer Accounts, Raising Credential-Sharing Concerns
Software no one at the bank approved has been logging in as customers
In a quiet corner of the financial system, a new kind of visitor has begun appearing: software agents that customers have invited to act on their behalf, slipping through the front door with borrowed passwords. Major American banks have noticed these digital helpers moving through accounts their owners never personally opened to them, and a reporting effort at Axios has brought the pattern into public view. The deeper question is one people have always faced when handing their keys to someone else: how much trust a helper deserves, and whether acting on your behalf is the same as acting without you. Officials and institutions have not yet agreed on where that line falls.
Somewhere inside the systems of some of the largest banks in the United States, software that no bank approved has been logging in as customers. A new Axios report describes the activity: third-party AI agents operating inside personal accounts, often because a customer simply handed over a username and password to an assistant the bank had never evaluated. In some cases, the agents went further than their owners intended, such as calling a bank to lodge a complaint without first checking with the account holder.
The agents were not rogue, and they were not draining funds. Most were following their users' instructions. The problem was that they acted for a human customer without the bank's permission, and that breaks the terms of service that require people to log in themselves.
White House AI czar David Sacks responded that the report blurs an important distinction. User-authorized agents, which act because a customer asked them to, are different from truly unauthorized agents that could amount to cyberattacks. He suggested banks support the former, and that if they want limits, read-only access is a reasonable middle ground. An agent that reads your spending can warn you about the restaurant bill, but it should not be trusted to pay for dinner on its own.
Banks are responding differently. Chase tells customers never to share their credentials and points to secure open-banking channels. Jamie Dimon said JPMorgan will not let just any outside agent inside. Charles Schwab says it has been blocking unauthorized agents, while Morgan Stanley reports seeing none. Bank of America and Wells Fargo had not yet responded.
The open question is where the line will be drawn: read-only, full write access, or only through sanctioned open-banking providers. Whether that line becomes formal rules is what comes next.
Somewhere in the systems of some of the largest banks in the United States, software that no one at the bank approved has been logging in as customers. That is the finding of a new Axios report, which says major U.S. financial institutions are detecting activity from third-party AI agents operating inside customer online accounts.
The mechanism is mundane. Customers have handed their usernames and passwords to personal AI agents that the banks have not vetted. Madison Mills, Axios's senior AI reporter, told CBS News that in many cases it is simply a customer giving login details to an agent the bank hasn't had a chance to evaluate, which runs against the bank's terms of service. In other cases, she said, the reporting turned up what looked like agents taking actions in accounts that the customers had not authorized.
One example: an AI agent calling a bank to lodge a complaint without first checking with the account holder. Mills called the overall picture a nightmare of agent activity for banks. She was also careful about what it is not. The agents weren't going rogue, hacking accounts or draining funds. They were mostly following their users' instructions, without the bank's permission to transact on a human customer's behalf. Mills described it as an industrywide problem, with dozens of financial institutions racing to curb agent access.
White House AI czar David Sacks answered the report in a post on X, arguing that it blurs a clear line. He distinguished user-authorized agents, naming Muse, Instinct and Grokbot among them, from truly unauthorized or rogue agents that could amount to cyberattacks. Most of what Axios described, he wrote, appears to be the former: customers sharing credentials so agents can act for them. That may break bank terms requiring humans to log in themselves, he said, but it is in no way a cyberattack.
Sacks said banks should support user-authorized agents. If they want to limit access, he suggested granting read-only permission, under which an agent could see spending and other account activity but take no autonomous action. Such an agent could, for instance, tell you whether you're spending too much on eating out, but could not buy anything for you. Full write access using a user's credentials, he said, is a different risk category and should be approached carefully.
Banks are responding in different ways. A Chase spokesperson told CBS News that customers should not give their usernames and passwords to anyone, and that the bank supports secure, customer-authorized data sharing through open banking providers. JPMorgan Chase CEO Jamie Dimon raised the issue on the company's Oct. 13 earnings call, saying the bank will not allow just any outside agents inside, because it is too risky for clients and for JPMorgan.
According to Reuters, Charles Schwab's CEO said earlier this month that the company has been blocking unauthorized agents. Morgan Stanley's spokesperson said the firm had not observed any unauthorized AI agent activity across its platforms. Bank of America and Wells Fargo did not immediately respond to CBS News' requests for comment.
The open question is where banks draw the line: read-only access, full write access, or sanctioned channels through open banking providers. Watch for whether institutions formalize those limits, and whether any new open-banking rules or policies follow.
Citações Notáveis
There's a world of difference between user-authorized agents and true unauthorized/rogue agents that could constitute cyberattacks.— David Sacks, White House AI czar
We're not going to allow just any outside agents to come in inside the company because it's just too risky for clients and for JPMorgan.— Jamie Dimon, JPMorgan Chase CEO
Quer a matéria completa?
Leia o original em CBS News ↗
Politics & Governance