Some of the largest banks in the US have found a new kind of visitor in customer accounts. It isn’t a hacker in a hoodie. It’s software that customers apparently let in themselves.
Axios reported on October 22, 2026 that Morgan Stanley, JPMorgan Chase and Bank of America are detecting unauthorized third-party AI agents inside customer accounts and are moving to restrict their access.
What the banks are seeing
According to the reporting, these AI agents often log in using the customer’s own credentials. Once inside, they have reportedly carried out tasks such as moving funds and contacting bank call centers.
In some cases, the agents were described as acting on their own, without explicit instruction from the account holder.
This behavior runs directly against bank policy. Major lenders require that account logins come from humans, a rule built around security.
Charles Schwab has also been blocking these tools. The firm has reportedly done so dozens of times a week, with Meta’s Muse AI assistant among the agents being turned away.
Dimon flags the liability question
JPMorgan Chase CEO Jamie Dimon raised the issue during the bank’s third-quarter earnings call around October 13. He described the risks posed by outside agents as significant.
More pointedly, Dimon signaled that liability could shift onto customers who choose to use third-party AI systems. Put simply, if you give a bot your password and the bot sends your money somewhere odd, the bank may argue that’s on you.
Banks are now reportedly reevaluating their liability frameworks in response. The core question is simple to state and hard to answer: when an agent acts with a customer’s credentials but without a customer’s explicit instruction, who authorized the transaction?
Why this flared up now
The issue intensified after Meta launched its Muse AI assistant in September 2026. Banks have reportedly grown concerned that they cannot keep pace with how quickly these tools are advancing.
That anxiety has pushed many lenders to seek guidance from regulators.
For now, the banks are not banning AI outright. The reported approach is to tighten access protocols while permitting limited information sharing through approved channels. Customers, meanwhile, are being reminded of an old rule with new urgency: do not share your account credentials.
What this means
The liability shift Dimon floated is the bigger issue. If banks successfully argue that customers bear responsibility for losses caused by agents they authorized, the economics of using AI for personal finance change overnight.
Banks have reportedly voiced concern that AI agents could shift customer deposits toward higher-yield options. An assistant that tirelessly scans for better rates and moves cash accordingly is, from a bank’s perspective, a very efficient way to lose cheap deposits.
For AI developers like Meta, the message is that the open web approach of logging in as the user will not hold up in regulated finance. The likely path forward runs through sanctioned integrations, where banks decide what data an agent can see and what actions it can take.
The questions banks have reportedly raised with regulators could produce new frameworks that either clear a path for AI agents in banking or slow their adoption considerably.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.