Big Banks Block Outside AI Agents From Moving Money

Four of the biggest names in American banking drew the same line in a single week: outside AI agents may look at your accounts, but they may not move your money. Between October 13 and 17, JPMorganChase, Bank of America, Citigroup and Charles Schwab each said that outside AI agents would not be allowed to log into customer accounts or initiate transfers, according to reporting from PYMNTS, the Financial Times and Bloomberg.

The lockdown arrives just as AI agents go mainstream. Meta's Muse and OpenAI's Dots launched as full-fledged consumer agents in recent weeks, and millions of users are already handing outside AI agents tasks that end at the bank's front door. The banks' answer, so far, is no — and the reason is the same everywhere: nobody has figured out who is liable when an outside AI agent moves money wrong.

What the four banks actually said

The week started on Tuesday, October 13, on JPMorganChase's third-quarter earnings call, where CEO Jamie Dimon warned of "rogue agents" and said outside AI agents would not get inside the bank, according to PYMNTS. A bank spokeswoman later confirmed that the bank blocks outside AI agents from performing tasks for retail customers, and Dimon told analysts that customers who use outside AI agents should understand there is a liability transfer, as reported by WebProNews, citing CNBC's coverage of the call.

Bank of America followed a day later with a condition rather than an outright ban. The bank is limiting outside AI agents' access to consumer deposits while it writes protocols that decide which party is liable for a payment an AI authorizes, according to the Financial Times. CEO Brian Moynihan said on the October 14 earnings call that a human should stay in the loop to approve money moving out of an account.

By Friday, Citigroup had drawn the same line: the bank does not let third-party agents log into accounts or transact, as reported by the Financial Times. And Charles Schwab gave the sharpest numbers of the week. CEO Rick Wurster told Bloomberg on October 16 that Schwab actively blocks unauthorized AI agents — including Meta's Muse — in the "tens and 20s" each week, across all of its clients.

The Schwab quote that traveled furthest was Wurster's liability question: he said he did not want clients handing passwords to an agent and letting it take action on their behalf, because neither the bank nor the client could control it, something could go wrong, and there was no clear answer on who would be liable, according to Digital Trends, which covered the Bloomberg interview. He added that he gets a weekly report on the blocks, which he takes as a sign that people are not yet ready to trust AI with their full accounts.

Read-only is open, moving money is closed

All four firms kept one door open while slamming the other. Outside AI agents can still see balances when a customer authorizes a connector such as Plaid at JPMorgan, according to PYMNTS — but they cannot send money or pay bills, and the bank gave no timeline for changing that. Citigroup lets certain authorized open-banking providers retrieve account information in read-only form, according to the Financial Times. And Schwab plans to build Model Context Protocol servers so a client's chosen AI model can check balances and run research under rules the brokerage sets.

The fight echoes the open-banking battle over customer data, in which JPMorgan pushed back on third-party access to its infrastructure and began charging aggregators for data access, a dispute that ended in a September 2025 data-access agreement with Plaid, as reported by PYMNTS. But the stakes are higher with agents: unlike data aggregators, which only read account information, outside AI agents can initiate transactions and move money, which raises risks around fraud, unauthorized payments and accountability for what goes wrong.

The demand is clearly there. PYMNTS Intelligence found in February that 37% of AI power users relied on native AI platforms as their primary tool for managing finances and banking, and a June report found that 39% of US adults had used AI for a payment-related activity — with consumers insisting the technology ask permission before completing a purchase.

Notably, the banks are not ceding the money-moving job to anyone else: they plan to do it themselves. Dimon wrote in his 2025 shareholder letter that JPMorgan's Smart Cash capability would move client money automatically between checking and higher-yield brokerage products in 2026, according to PYMNTS — the same sweep an outside agent would offer, inside the bank's own app. And Schwab is rolling out Charley, its own generative AI assistant announced September 30, which answers questions on transfers and trading but hands off to a human representative, according to the company.

The permission problem AI agents still have to solve

Strip away the posturing and every bank's statement converges on the same missing piece: the permission layer. Congress is still debating how AI agent liability should work, and until there is a framework that says who pays when an outside AI agent misfires, banks will default to no.

Some institutions are working toward yes instead of just saying no. Bank of America is writing liability rules directly with AI agent developers, according to the Financial Times. And one firm has already shipped the permission model the others are waiting for: Robinhood lets outside AI agents trade through its banking MCP server with a customer-set spending limit, optional manual approvals and no access to the primary card number, as reported by PYMNTS.

The identity infrastructure is racing to meet the moment. World's World ID for Agents, unveiled at Money20/20 this week, lets a verified person delegate identity proofs to an AI agent so banks can distinguish an agent acting for a real human from anonymous automated traffic. And new transaction-signing protocols create signed evidence of exactly what a user agreed to, so a bank can verify consent before executing a transfer.

For the agent economy, the week's lockdown is less a verdict than a specification. The banks told outside AI agents exactly what it will take to earn access: limits set outside the agent, a named human approving the precise action, and an audit trail that survives a dispute. Outside AI agents that cannot bring those three things will keep getting blocked. The ones that can may find Wall Street's door opens faster than anyone expects — the contrast with crypto, where autonomous trading agents already run under no bank at all, could hardly be starker.