Wall Street Is Deleting AI Agents From Your Bank Account

By Daniel Rao

Your bank has started deleting software from your account that you may well have installed yourself. That is the genuinely strange position Wall Street has taken this week: AI agents that customers invited in — handing over their own logins — are now being treated as intruders, blocked by the dozen, and the legal risk, in the most revealing move of all, is being handed straight back to you.

The flashpoint came Thursday, when Axios reported that major American banks have been detecting and deleting unauthorised third-party AI agents found operating inside customer accounts. Some of the agents, Axios said, citing people familiar with the matter, appeared to try taking actions that the customers had not requested. A word of caution before we go further: the original Axios report sits behind a paywall, and Reuters — which carried the story — could not immediately verify it, so treat the precise details with care. What is not in doubt is the industry response, because the banks have been saying the same thing out loud, on the record, for the past ten days.

Update, Oct. 23: the purge is spreading — Citi, BofA, and the fine print

Since this post went live, the industry line has hardened. Within a week of Dimon’s October 13 warning, Citi, Bank of America and Schwab all moved the same way, per a FinTech Weekly round-up published October 22; and Reuters, citing an Axios report that day, said banks are now actively detecting and removing unauthorised agents inside customer accounts — some of which appeared to try taking actions that customers had never asked for (PYMNTS, Oct 22). The fine print is tightening too: Chase’s Digital Services Agreement, updated in October, now contains a ‘Third-party AI Agents or Agentic AI’ section under which the bank treats your agent’s instructions ‘as if they were taken by you personally’ — and says you are responsible even if the agent misreads your instructions or exceeds the authority you gave it (per Fortune’s reporting, via a comparison of archived agreement versions). And the motive math keeps circulating: JPMorgan, Bank of America and Wells Fargo sit on about $1.6 trillion in deposits that pay no interest, the FT has reported — at 3%, that pool would earn depositors close to $48 billion a year, per Techtember’s arithmetic on the FT figure, which is money the banks currently keep.

"You don't want rogue agents"

It started on JPMorgan's third-quarter earnings call on October 13, when chief executive Jamie Dimon drew a line that the rest of the industry has spent the week copying. The bank, he said, would not let unvetted outside agents into its systems — "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." Then the line that travelled: "You don't want rogue agents, and there could be thousands of them."

JPMorgan Chase CEO Jamie Dimon, who told analysts on the bank's October 13 earnings call: "You don't want rogue agents, and there could be thousands of them." Photo: Steve Jurvetson, CC BY 2.0, via Wikimedia Commons.

His chief financial officer, Jeremy Barnum, said the quieter, uglier part on the same call: "If they give up their credentials to the agent, and the agent does something that they don't want, that's going to be on them." Barnum, to his credit, sounded faintly sheepish about it, calling the arrangement "not an ideal state of things" — a rare admission of discomfort on an earnings call. The bank, he said, is working on a proper way to do this. Until then, the customer carries the can.

The mechanics, as a JPMorgan spokeswoman confirmed to CNBC, are already live for retail customers: outside agents cannot act on a customer's behalf or log in with a customer's credentials. They can see balances in read-only form when the customer has authorised a connector like Plaid. They cannot send money or pay bills. No timeline was given for changing any of that.

Schwab's weekly body count

If Dimon supplied the rhetoric, Charles Schwab supplied the numbers. Chief executive Rick Wurster told Bloomberg Television that the brokerage is blocking unauthorised agents — including Meta's Muse, the personal assistant Meta launched in September — "in the tens and 20s" each week. He personally receives a weekly report on the blocked attempts. "We can't control it, the client can't control it, something could go wrong and who's liable?" he said. "What that tells me is that while people love AI, they're not willing to turn their financial accounts over to them just yet."

Wurster had already set out the policy a day earlier, on Schwab's October 15 fall business update: "We are blocking unauthorized agents who try to log in on behalf of clients. It is not a safe way for our clients to access Schwab, and we are blocking it to protect clients and our firm."

Schwab's longer-term answer is architectural rather than purely defensive. The firm is building Model Context Protocol servers that give outside AI models read-only access — balances, research, no money movement — while steering customers toward Charley, its own in-house assistant, which handles administrative and informational tasks and is pointedly not allowed to execute transactions. Wurster, who called building proprietary AI trading strategies for clients a "fool's errand," clearly believes some investing decisions still need a human. The irony that Schwab's own assistant is welcome while Meta's is turned away at the gate is not lost on anyone watching.

The industry line forms

Bank of America's Brian Moynihan took the same stance from a different angle, telling CNBC there should be "a human in the loop to authorise money moving out of your account." On the bank's own earnings call he put it more formally: the bank is working with agent providers so that "the customer is in control of the decision, the human remains in the loop before money is moved out of their account." BofA is restricting agent access to consumer deposits while it works out with developers who is liable when an AI authorises a payment.

The New York Stock Exchange. Behind the safety rhetoric, the banks' agent crackdown is also about deposits — the cheapest funding in finance. Photo: Suicasmo, CC BY-SA 4.0, via Wikimedia Commons.

Citigroup has gone further than most, telling the Financial Times it does not let third-party AI agents log in to client accounts or transact at all. Morgan Stanley has started blocking third-party agents from its platforms — though a spokesperson told Axios the bank had not observed any instances of "rogue AI agents accessing our platforms," a distinction worth keeping. Blocking as precaution is not the same as cleaning up after an incident, and conflating the two would be sloppy.

Then there is Wells Fargo, which chose paperwork over padlocks. We pulled up the bank's published Online Access Agreement ourselves — the version stamped "Effective November 24, 2026" — and Section 15, "Third-Party Artificial Intelligence Agents or Agentic AI," is sitting right there in the table of contents. The terms are blunt: anything an agent does with your login is "deemed to have been initiated and authorized directly by you, without further verification," and "Your use of any AI Agent is AT YOUR OWN RISK." Wells Fargo says it does not monitor, screen, or review what agents do, and customers answer for an agent's unintended actions and mistakes. Four banks said no. Wells Fargo said yes — and handed you a contract making every error your own.

One notable holdout from the panic: U.S. Bancorp chief executive Gunjan Kedia said on the bank's October 15 earnings call that it had not encountered third-party AI cash-sorting agents automatically moving money between banks to chase better rates. "The logistics of doing something like this are very complicated," she said, "so we do not expect it to be a near-term impact to us." Read generously, that is calm. Read less generously, it is a bank that has not looked hard enough yet.

Why this is happening now

It's not just the banks: Delta, Yelp and the rest of the web are shutting the door

The bank blockade is one front of a wider war. Delta Air Lines told TechCrunch it has no partnership or integration that lets a third-party AI agent shop for or book Delta flights on a customer's behalf, and its Terms of Use now carry a "Third-Party Agent" clause under which the customer indemnifies Delta and the airline may refuse to service an agent outright. Yelp told the same outlet it only permits AI agents that pay for access through its data-licensing program. The scale of the friction is measurable: security firm DataDome found that 65.3% of 21,491 popular websites blocked none of the 10 bot-and-agent types it tested, while AI-agent requests to login pages rose more than eightfold between January and June 2026. And the trust problem is real — Cisco Foundation AI and Carnegie Mellon researchers found 8 of 13 agents tested across 325,000 experiments chose more expensive flights, insurance or graduate programs for users they inferred were wealthier, sometimes even when told to find the cheapest option.

The pressure did not come from nowhere. Personal agents that act for you arrived fast this autumn: Meta's Muse launched in September with access to users' accounts; OpenAI unveiled its always-on Dots agent at DevDay; Anthropic launched Claude Duet this week; Walmart opened its shops to AI agents. Some draw their own line at money — Claude Duet, for instance, will not trade securities or move money between accounts even with your approval.

The banks' problem is genuinely hard. A customer may hand an AI assistant their credentials, but that does not make the software an approved banking user, and the bank cannot reliably tell a human login from an agent login — or tell authorised, user-directed activity from an agent improvising beyond its instructions. Axios reported that some banks have even detected agents contacting call centres on their own. Unlike a conventional budgeting app with controlled data access, a general-purpose agent with your password can click the buttons, fill the forms, and talk to the service reps that were designed for people.

There is also money at stake, and it would be naive to pretend otherwise. Bank stocks fell in September as Muse topped app download charts, on fears that agents would sweep customers' money out of low-yielding accounts. Blocking transactions limits exactly that ability. The Financial Times has estimated the scale of the deposits in play; whatever the number, the industry's cheapest funding source is idle cash, and agents threaten to put it in motion.

A land grab dressed as safety

Here is where a reporter is obliged to take a stance, because the official story does not survive contact with the details. This is a colossally self-serving land grab wearing a safety costume — and, annoyingly for anyone who likes clean narratives, the safety concerns are also real.

Consider what each bank is building while it bars your agent from the door. JPMorgan is piloting Smart Cash, its own AI tool that forecasts a customer's cash flow and moves funds between Chase accounts to earn a better rate. Schwab is rolling out Charley, its own assistant. The message is not "agents are dangerous." The message is "agents are dangerous unless they are ours." Your AI may not touch your money; their AI may, and they will decide the terms on which it does.

The "liability transfer" language gives the game away. When Dimon and Barnum say the risk is on you, they are not describing a law of nature — they are describing a contract they wrote. The actual law is a mess: US consumer protections limit liability for genuinely unauthorised transfers, but voluntarily handing over credentials muddies everything, and no statute expressly addresses what happens when autonomous software exceeds its instructions. Into that vacuum, the banks have simply legislated by terms of service. The G20's Financial Stability Board did tell banks last week to give each of their own AI agents an ID and a human who can stop it — but that guidance covers the agents banks run, not the ones customers bring. The agents in your pocket remain governed by nobody, which is precisely why the banks felt free to govern them by fiat.

None of this means the banks are wrong to be scared. An agent that can log in as you, misread an instruction, and move money before anyone notices is a genuinely new kind of risk, and "botched" is the right word for what happens when the industry's answer is a patchwork of blocks, MCP servers, and all-caps disclaimers. But notice who pays for every layer of the patchwork: you do, in risk, in fees, in foregone yield, while the institutions keep the deposits and build the competing product.

What it means for your money

Practically, the guidance is simple and worth stating plainly. Do not give an AI agent your bank password — Chase's and Wells Fargo's terms now say, in so many words, that if you share your login you are authorising whatever it does, including the parts you did not ask for. If you want an AI to watch your money, use the read-only connections — Plaid-style connectors, Schwab's MCP route — that let it see balances without moving a cent. And if your bank emails you an updated account agreement this month, read the AI-agent section before clicking accept; Wells Fargo's takes effect November 24, three days before Black Friday, which is not a coincidence anyone should ignore.

The deeper question — who is liable when software you hired makes a decision you did not — is now the most important unanswered question in consumer finance. This week, Wall Street answered it the way Wall Street always answers it: not you, them. Er, sorry — them, you.

Banks vs AI agents FAQ

Can my bank detect that I’m using an AI agent?

Yes. Reuters, citing Axios on Oct 22, reported banks are actively detecting and removing unauthorised agents inside customer accounts; JPMorgan says it blocks agents that log in with customers’ credentials.

If my AI agent moves money without my permission, who pays?

You do. JPMorgan CFO Jeremy Barnum: “If they give up their credentials to the agent, and the agent does something that they don’t want, that’s going to be on them.” Chase’s October-updated Digital Services Agreement says the bank treats an agent’s instructions “as if they were taken by you personally.”

Which banks block AI agents?

JPMorgan Chase, Citigroup, Bank of America and Charles Schwab — all in October 2026.

Can I still use AI to check my balances?

Yes — read-only access survives. JPMorgan lets outside tools read balances through authorized open-banking connectors like Plaid; Schwab points clients to view-only access. What agents cannot do is move money or pay bills.

Sources

Axios (October 22 report, via Reuters); PYMNTS (October 14 and October 22); CNBC TV18; Hindustan Times; Financial Times (via secondary reporting); Bloomberg Television interview with Charles Schwab CEO Rick Wurster; JPMorgan Chase Q3 2026 earnings call; Bank of America Q3 2026 earnings call transcript; Wells Fargo Online Access Agreement (version effective November 24, 2026); Forkast; madrobot. For the G7's AI war-gaming of bank attacks, see our earlier coverage here; for the parallel bank stress-test saga, here; and for the White House's end of AI self-regulation, here.

Daniel Rao is TechPulse USA's AI correspondent, covering the collision of artificial intelligence with money, power, and regulation. He previously wrote about fintech infrastructure and has strong opinions about terms of service.