US banks blocking AI agents from accessing customer accounts

One analyst's framing captures the real logic at work: an outside agent works for the client, while Schwab's in-house assistant Charley works for Schwab^[3]. That distinction explains why the same bank drawing a hard line against Meta's Muse is simultaneously building a functionally similar tool for itself - Schwab points clients toward Charley for administrative tasks it still controls^[3], while deploying read-only MCP servers that let partner agents view balances but never move money^[1]. Community discussion around JPMorgan's announcement noted the same tension from a different angle: the bank that bars Muse from touching customer accounts is reportedly piloting its own internal cash-optimization tool aimed at the identical yield-chasing behavior it won't let outsiders perform. The liability language banks use publicly - Wurster's 'who's liable?'^[1], sits awkwardly next to a business logic that keeps the client relationship, the data, and the fee revenue inside the bank's own walls rather than ceding it to Meta or any other outside agent maker.

Beneath the liability language is a balance-sheet problem: roughly $1.6 trillion in non-interest-bearing deposits at JPMorgan, Bank of America, and Wells Fargo is exposed if AI agents automatically sweep idle cash out of checking accounts paying a national average of about 0.1% and into fintech alternatives such as Revolut, SoFi, Varo, or Wealthfront offering 3.3% to 5%^[4]. Apollo Global Management's Torsten Slok has framed this as a potential 'agentic bank run,' warning that banks could lose a large share of the cheap deposits they rely on to make loans^[4]. If banks had to pay a competitive 3% on that at-risk deposit base, the added cost would run to roughly $47 billion a year - close to half the combined annual earnings of the banks involved^[5]. That math, more than any single security incident, is a plausible reason banks moved from quiet caution to public blocking policies within days of each other in mid-October 2026. X.com commentary around Slok's warning echoed the same framing, treating deposit flight as a systemic worry for the sector rather than a niche technical one.

JPMorgan CEO Jamie Dimon has been the most explicit about what the bank is actually guarding against: not AI itself, but the liability transfer that happens when a customer hands banking credentials to a third party. 'We're not going to allow just any outside agent to come in inside the company, because it's just too risky for clients and for JPMorgan. You don't want rogue agents, and it could be thousands of them,' Dimon said^[2]. Schwab draws the same line around credential-sharing and browser automation specifically, rather than AI access in general - the bank says it already blocks tens to twenties of unauthorized access attempts a week by agents like Muse trying to log in and transact^[1]. Bank of America has taken a softer version of the same stance, allowing agent-initiated transactions but insisting a human approve any money movement rather than banning agents outright^[5]. The common thread is that none of the banks are rejecting AI as a concept - Schwab, for instance, already runs its own proprietary assistant, Charley, for the tasks it keeps in-house - they are rejecting a model where a client's login credentials, and therefore legal responsibility for what happens next, pass to a company the bank has no authority over.

Banks are setting these policies without a clear regulatory mandate to do so. Agentic AI remains explicitly outside the scope of the interagency model-risk guidance banking regulators currently rely on^[6], and banks say they have briefed Federal Reserve officials on their blocking stance even though no formal rule compels it. That gap matters because generative-AI-related lawsuits in the US grew 978% between 2021 and 2025^[6], meaning banks are making unilateral risk calls in a legal environment that is already becoming more litigious. Reddit and YouTube discussion of the story both situated the banking lockout inside a wider pattern: AI agents are reportedly being blocked or restricted not just by banks but by Amazon, Yelp, Delta, Amtrak, and Cars.com, suggesting the agent-versus-platform standoff triggered by Meta's Muse extends well past financial services. One recurring counterpoint in that discussion is that none of this strictly requires large language models at all - a simple deterministic app could already auto-sweep cash to higher-yield accounts, which some read as a sign that the 'AI agent' framing is doing more rhetorical work than technical work in this story.