Four of the largest United States banks block AI agents from logging in to customer accounts. JPMorgan Chase, Bank of America, Citigroup and Charles Schwab each confirmed the restriction between 13 and 19 October 2026, naming third-party tools that sign in using a customer’s own credentials. Read-only data feeds through approved open-banking providers remain open.
How the Week Unfolded
The restrictions were not announced together. They emerged across seven days, in earnings calls, television interviews and a Financial Times report, which is why the picture has looked inconsistent.
- Tuesday 13 October 2026: Jamie Dimon, chairman and chief executive of JPMorgan Chase, told the bank’s third-quarter results call that outside agents would not be allowed inside the company’s systems.
- Wednesday 14 October 2026: Brian Moynihan, chief executive of Bank of America, told CNBC that a human should authorise money leaving an account, and that the bank is still working out who is liable for AI-authorised payments.
- Friday 16 October 2026: Bloomberg reported that Charles Schwab is blocking unauthorised agents attempting to move client money, including Meta Platforms’ Muse assistant.
- Monday 19 October 2026: the Financial Times reported that Citigroup and Bank of America had joined JPMorgan and Schwab, and that Wall Street executives had raised the issue with the Federal Reserve.
An AI agent, in this context, is software that acts on a person’s behalf without a human approving each step. Meta launched its Muse consumer assistant in September 2026; CNBC’s report on 13 October also named a tool called Instinct. Our earlier explainer on what Meta Muse can access sets out how that permission model works.
What Is Blocked and What Is Still Permitted
The common thread is the login, not the data. Every institution named below bars an agent from authenticating as the customer; most still allow an agent to read balances through a sanctioned pipe.
| Institution | What it bars | What it still permits |
|---|---|---|
| JPMorgan Chase | Outside agents signing in with customer banking credentials; credential sharing and screen scraping | Limited data such as balances through intermediaries including Plaid, under the bank’s account-access policy |
| Citigroup | Third-party AI agents or assistants logging into an account or transacting directly on its digital platforms | Limited read-only access to account information for certain open-banking providers |
| Bank of America | Agent access to consumer deposits, per a person familiar with the matter cited by the Financial Times | Not specified; the bank says it is building tracing tools with agent providers |
| Charles Schwab | Unauthorised agents logging in on a client’s behalf; password sharing with third-party bots | Balance checks and research through Schwab’s own Model Context Protocol servers, with no money movement |
Rick Wurster, chief executive of Charles Schwab, said the firm turns away unauthorised agents between 10 and 20 times a week. Schwab is steering clients towards Charley, its own assistant, which it announced on 30 September 2026.
None of this is a ban on AI inside the banks. On the same 13 October call, Dimon said JPMorgan already uses agents internally and intends to keep doing so, adding that the bank is “not against agents or for agents”. The line being drawn runs between the bank’s own automation and software it does not control.
Why $1.6 Trillion in Idle Deposits Is the Commercial Stake
Security is the stated reason. Deposits are the other one, and the figures are large enough to explain the speed of the response.
Torsten Slok, chief economist at the asset manager Apollo, estimates that JPMorgan Chase, Bank of America and Wells Fargo together hold about $1.6 trillion in non-interest-bearing deposits, around 16% of their liabilities. The Financial Times reported, and both Techstrong.ai and The Paypers repeated, that paying roughly 3% on that money would cost those banks about $47 billion a year.
The mechanism is simple. An agent with payment authority could move idle cash from a checking account paying a United States national average of about 0.1% into products paying 3.3% to 5%. Those low-cost deposits fund commercial lending, so a tool that sweeps them automatically changes a bank’s funding cost rather than merely its app experience.
On the 13 October call, analysts from Bank of America, Wells Fargo and Evercore ISI each pressed JPMorgan on exactly that risk. Chief financial officer Jeremy Barnum answered that personal agents are a new form of existing yield-seeking and fintech competition rather than a wholly new threat. Dimon said regulators should consider the liquidity question.
Who Carries the Liability When an Agent Moves Money
This is the unresolved part, and no bank or regulator has settled it.
Dimon told the third-quarter earnings call that misuse at scale was the concern: “you do not want rogue agents, and there could be thousands of them.” He also argued that customers should not bear a liability transfer when an agent is misused. Barnum’s position on the same call was narrower: a customer who hands over login credentials to an agent bears the consequences.
Wurster put the question plainly, telling Bloomberg: “We can’t control it, the client can’t control it, something could go wrong and who’s liable?” Moynihan said Bank of America is developing protocols to establish which party is responsible for a payment an AI tool authorises.
Those three positions are not the same, and United States law does not currently resolve them. Rules written for a person entering a password, or for a company acting under a written mandate, do not obviously cover software acting on a standing instruction. Anyone affected should check the account-access terms published by their own bank rather than assume a common standard.
What Regulators Have Said
As of 19 October 2026, no United States regulator has issued a rule, guidance or enforcement action on agent access to bank accounts.
The Financial Times reported that Wall Street executives contacted the Federal Reserve about the growth in consumers using AI agents for financial transactions, and that regulators have privately welcomed the restrictions. The Federal Reserve declined to comment. Officials quoted in that reporting believe the law may need to clarify how a customer authorises autonomous software and who bears a loss when a transaction fails. That account rests on unnamed sources and has not been confirmed on the record by any agency.
The pattern is not confined to banking. Amazon blocked Meta’s Muse from making purchases in September 2026 and sent enforcement demands over Perplexity’s Comet browser, as we reported in Amazon’s block on Muse purchases. The term “rogue agent” has also surfaced outside finance, including the Wikimedia case involving OpenAI agents.
Frequently Asked Questions
Why Do Banks Block AI Agents?
The institutions cite two reasons: the security risk of software signing in with a customer’s credentials, and unresolved liability for payments an agent authorises. Reporting by the Financial Times adds a commercial motive, the risk of agents moving idle deposits to higher-yielding products.
Can an AI Assistant Still See a Bank Balance?
In most cases yes. JPMorgan Chase permits limited data such as balances through intermediaries including Plaid, Citigroup permits read-only access for certain open-banking providers, and Charles Schwab allows balance checks through its own servers without money movement. These policies apply to United States consumer accounts.
Which AI Agents Have Been Named?
Meta Platforms’ Muse has been named by Charles Schwab and in Amazon’s earlier block. CNBC’s 13 October 2026 report also named a tool called Instinct. Most bank statements describe categories of third-party agent rather than specific products.
Has Any Regulator Ruled on This?
No. As of 19 October 2026 there is no United States rule, guidance or enforcement action on the question. The Federal Reserve declined to comment on reports that executives had raised it.
Does This Mean Banks Are Rejecting AI?
No. Jamie Dimon said on JPMorgan’s 13 October 2026 results call that the bank uses agents internally and will continue to, and Charles Schwab is promoting its own assistant, Charley, announced on 30 September 2026. The restriction applies to third-party software, not to the banks’ own tools.
