AI agents and bank accounts: Morgan Stanley, JPMorgan and Bank of America move to restrict unvetted agents, and what account terms say

AI agent bank account access has become a problem big US banks say they're acting on. On October 22, 2026, Axios reported that major banks have detected unauthorized third-party AI agents in customer accounts, most often an AI assistant logging in with the customer's own username and password. Morgan Stanley has begun blocking them, and JPMorgan, Bank of America and Citi are restricting them, Axios reported. The fight is the banking version of one already playing out on the open web, where websites are closing their doors to AI agents.

This analysis is based on Axios's October 22 report; JPMorgan's October 13 earnings call as reported by PYMNTS; the Financial Times's October 16 report on Bank of America and Citi, as summarized by PYMNTS; American Banker's report on Charles Schwab's October 15 earnings call; Chase's Digital Services Agreement and Bank of America's Online Banking Service Agreement; the CFPB's rulemaking page; court reporting by the ABA Banking Journal; and Visa's and Mastercard's announcements on agent payments, all checked October 26, 2026. It describes account terms and rules; it isn't legal or financial advice.

The short version

• What's happening: customers are giving AI agents their bank logins. Axios reported that some agents contacted call centers or took steps users hadn't asked for, and that banks struggle to tell agents from people.
• What banks are doing: Morgan Stanley is blocking third-party agents, and JPMorgan, Bank of America and Citi are restricting them, per Axios. Most are moving toward read-only access, as they already offer to apps such as Quicken. The Financial Times had reported Bank of America's and Citi's limits on October 16, and Charles Schwab's CEO has said it won't let agents log in for clients.
• What account terms say: Chase's terms, updated October 2026, add "AI Agent Terms" that treat an agent's actions as the customer's own. Bank of America's terms make customers responsible for anyone they let use their login, "even if he or she exceeds your authorization."
• The rules: the CFPB's Section 1033 open-banking rule is enjoined and under reconsideration, so there is no settled federal framework for agent access.

What did Axios report about AI agents in bank accounts?

Axios's Madison Mills reported, citing unnamed sources, that banks are finding third-party agents inside customer accounts, and that some appear to act independently of user instructions. The most common case, per Axios, is an agent logging in with the customer's permission and credentials. That breaks most banks' rules that humans log in to their own accounts, even when the customer approves it.

Executives' main worry, Axios reported, is agents with login access making purchases or moving money. Banks are struggling to separate human from bot activity, which is pushing them to block the technology and rewrite their terms quickly.

Axios's report rests on anonymous sources and doesn't quantify how many accounts are affected or name the agents involved. The companies' on-the-record statements are narrower.

Last verified: October 26, 2026

JPMorgan's CEO had already drawn the line. On the bank's third-quarter earnings call on October 13, 2026, Jamie Dimon said, "You don't want rogue agents, and there could be thousands of them," according to PYMNTS, which cited a Motley Fool transcript. PYMNTS reported that he said the bank wouldn't let just any outside agent in because the risk to clients and the bank was too high, and pointed to "the liability transfer" as a question.

Why are banks worried about AI agents with your password?

Three risks run through the reporting and the banks' own documents.

The agent can do anything you can. Bank of America's security page puts it plainly: "When you share your login information with a third party, they have the same access to your information as you do." It adds that third parties "store your login information, which could be compromised if they have a security breach." An agent holding a password isn't limited to reading balances; it can reach whatever the website lets a logged-in person do.

Agents can be steered by what they read. An agent that browses the web or reads email can be hijacked by planted instructions, as our prompt injection explainer covers. Axios didn't say what caused the incidents it described.

Banks can't tell who is acting. If an agent logs in like a person, the bank's fraud and identity systems see a person. That's why card networks are building ways for agents to identify themselves (see below), and why websites are deploying agent detection, as our guide to detecting and blocking AI agents describes.

What do bank account terms say about sharing your login with an AI agent?

Two of the banks Axios named publish terms that speak to this directly.

Chase. Chase's Digital Services Agreement, marked "last updated October 2026," now contains "AI Agent Terms." It defines an Agent as "any software or service that takes autonomous, semi-autonomous, or non-autonomous action on behalf of, or at the instruction of, any person or entity." Key lines:

• "We will rely on any instructions or actions taken by your Agent as if they were taken by you personally."
• "You are responsible for any actions or transactions entered or requested by the Agent."
• "You also acknowledge that our privacy policies do not apply to the Agent."
• "You agree it is your responsibility to monitor your Agent's activities."
• Chase reserves the right to "limit, prevent, suspend or terminate access by an Agent to your accounts… at any time without notice, which may render the third-party agent non-functional."

Elsewhere, the agreement says that if you disclose your login credentials, "YOU ASSUME THE ENTIRE RISK FOR THE FRAUDULENT, UNAUTHORIZED OR OTHERWISE IMPROPER USE OF YOUR LOGIN CREDENTIALS," and that Chase may terminate service if it deems a customer's use of "a third-party AI agent or agentic AI potentially compromises" security.

There is a tension worth noting. The terms describe how agent access with credentials is treated, while the spokeswoman quoted via PYMNTS said outside agents can't log in with a customer's credentials. Read together, Chase's position appears to be that it may block such agents, and that when one does get in, the customer carries the responsibility.

Bank of America. Its Online Banking Service Agreement, effective October 19, 2026, doesn't mention outside AI agents; its only reference to AI concerns the bank's own assistant, Erica. Section 8.B says: "If you permit another person to use your User ID and password and related code(s)," you are responsible for that person's transactions "even if he or she exceeds your authorization," until you notify the bank. Its Zelle section says customers won't "share your credentials with a third party to use the Service on your behalf except in legally authorized situations." Bank of America's security page recommends connecting apps by logging in "directly to Bank of America" instead of sharing credentials, and says connected services can be reviewed and revoked in Online and Mobile Banking.

Bank of America also uses AI itself: CEO Brian Moynihan said it added 1,000 contractors to fix flaws found with Anthropic's Claude Mythos, as our report on Moynihan's comments covers. The restrictions Axios describes are about outside agents customers bring, not the banks' own.

Where does the CFPB's open-banking rule stand?

Section 1033 of the Dodd-Frank Act gives consumers a right to access their financial data, and the CFPB finalized a rule implementing it in October 2024. It hasn't taken effect in practice:

• On August 22, 2025, the CFPB published an advance notice of proposed rulemaking to reconsider it.
• On October 29, 2025, the US District Court for the Eastern District of Kentucky, in Forcht Bank, N.A. v. CFPB, issued a preliminary injunction barring the CFPB from enforcing the rule, the ABA Banking Journal reported. The ABA's summary says the court reasoned that Section 1033 permits data sharing only with consumers or fiduciary-like agents, not commercial third parties.
• As of October 24, 2026, the CFPB's reconsideration page lists the August 2025 notice and no proposed rule.

That leaves agent access governed mostly by each bank's own terms and data-sharing deals, such as the data-access agreement JPMorgan and Plaid announced in September 2025, according to PYMNTS. Whether a consumer's AI agent counts as the consumer, an "agent" in the legal sense, or a commercial third party is exactly the kind of question an unsettled 1033 rule leaves open.

How are card networks trying to verify AI agents?

Payment networks have been building agent identity for purchases, as opposed to bank logins:

• Mastercard Agent Pay (announced April 29, 2025) says it "will require trusted AI agents to be registered and verified" and uses "Mastercard Agentic Tokens," with consumers controlling what an agent may buy.
• Visa's Trusted Agent Protocol (announced October 14, 2025) uses "agent-specific cryptographic signatures" so merchants can distinguish approved agents "from malicious automation and rogue bots."

Neither gives an agent a customer's bank password. They're the network-level version of the read-only, permissioned access banks say they prefer. Agent makers face the same design question; Anthropic's Claude Duet, for example, is built to ask before taking certain actions.

What's confirmed, reported and uncertain

• Confirmed (primary documents): Chase's AI Agent Terms; Bank of America's credential-sharing terms and guidance; the CFPB's 1033 reconsideration status; Visa's and Mastercard's agent programs.
• Reported: the incidents, Morgan Stanley's blocking and other banks' restrictions (Axios, anonymous sources); Dimon's quote and JPMorgan's read-only stance (PYMNTS, citing a transcript and CNBC); Bank of America's deposit limits, Moynihan's remark and Citi's login ban (the FT, as reported by PYMNTS); Wurster's remarks (American Banker); the Kentucky court's reasoning (ABA Banking Journal).
• Uncertain: how many accounts and which agents are involved; whether any customer lost money; how banks will handle liability disputes; what a revised 1033 rule will say about AI agents.