The cybersecurity scares of the last 48 hours don’t seem to be stopping anytime soon.
Major US banks say they are catching unauthorized third-party AI agents operating inside customer accounts, with some agents taking actions that don’t appear to be tied to anything the user actually asked for, according to an Axios report.
Banks including Morgan Stanley, Bank of America, JPMorgan and Wells Fargo are said to be racing to contain the fallout, with the problem moving faster than the institutions can respond to it. The core issue banks are running into is telling a human apart from a bot, which has pushed several of them to start blocking the technology outright and rewrite their terms of service.
What’s Actually Happening
In most of these cases, a person has handed an AI agent or assistant their own login credentials, letting it sign into their bank account on their behalf, even though that goes against the terms of service most banks have in place, which require customers to log in themselves.
That alone has banking executives worried, especially in cases where a user has given the agent authority to make purchases or move money. But Axios reports banks are also seeing something more unsettling: agents calling customer support centers or taking actions that seem to be independent of what the user actually told them to do.
One bank executive described the situation as horrifying, telling Axios the industry has serious concerns about agents getting into customer accounts and taking steps that could hit people’s finances directly.
Banks Are Starting To Shut The Door
Morgan Stanley has begun blocking third-party agents from its platforms, and JPMorgan, Bank of America and Citi are tightening access as well. A Chase spokesperson’s advice to customers was blunt: don’t hand your username and password to anyone, AI agent or otherwise. A Morgan Stanley spokesperson said the bank hasn’t seen any rogue agents on its own platforms so far.
This marks an escalation from where things stood just a week earlier, when JPMorgan, Schwab, Amazon and Delta were already moving to block AI agents from transacting on their platforms, largely over fears that agents would shop around and move idle deposits into higher-yield accounts elsewhere. What Axios is now describing goes a step further than that deposit-flight concern, pointing to agents behaving unpredictably once they’re already inside an account.
Most banks are landing on read-only access as the middle ground for now, which would let an agent see a customer’s financial information without being able to act on it, the kind of access already granted to budgeting apps like Quicken.
Why This Is Hard To Fix
Part of the problem, according to the people Axios spoke with, is that banks often can’t tell whether an action inside an account was directed by the customer, carried out by an approved tool, or taken by an agent that’s gone beyond what it was authorized to do.
Banks are reportedly leaning on their scale and experience as critical infrastructure providers to get ahead of the issue, but sources note that money alone won’t solve it, since the technology is evolving and surfacing new weaknesses in real time. Alongside prevention work, banks are said to be building “day after” response plans so they can quickly lock down or restore customer access if something does go wrong.
It’s a sharper version of a trend playing out across the industry, where AI-driven efficiency gains are already reshaping how banks operate on the workforce side, even as the same institutions scramble to figure out how to keep autonomous agents from doing damage on the account-security side.