Banks report unauthorized AI agents in customer accounts, and nobody says who pays when one moves money

Large US banks are detecting unauthorized third-party AI agents inside customer accounts, according to an Axios report published October 22, 2026, and the federal regulator in charge of national banks did not dispute it. The open question is not whether banks can block these agents. It is who pays when one gets through.

Everything below is attributed, because the central claim rests on unnamed sources in a story we could not read in full.

What was reported

Axios, citing unnamed sources, reported that unauthorized third-party AI agents had breached customer accounts at several large US banks. American Banker's first version of its story named Morgan Stanley, Bank of America, JPMorgan Chase and Wells Fargo, then corrected it: not all of those banks are aware of unauthorized access, and Morgan Stanley told the outlet it has "not observed any instances of unauthorized or rogue AI agents."

PYMNTS, relaying Reuters, adds that some agents appeared to attempt actions the account holder had not asked for. No outlet we opened reports a confirmed, named case of stolen funds.

What the regulator said

Comptroller of the Currency Jonathan Gould spoke Thursday on Fox Business. American Banker reports he said banks are using their own AI systems, which he called "SI" for superintelligence, to block bad actors from moving funds. His words: "The banks have been all over this for months now." He also said: "I'm here to assure you that SI is a powerful defensive weapon as well."

Host Stuart Varney asked twice whether customers could get their money back after an unauthorized AI agent transfer. According to American Banker, Gould did not say how such cases would be resolved. He called banks "some of the most fortified parts of our system" and said federal agencies are helping them get the latest tools.

What the banks say

Bank leaders had addressed the issue before the Axios story. PYMNTS reported on October 19 that JPMorganChase, Citigroup, Bank of America and Charles Schwab each said AI agents cannot move customer money. Per media reports it cites:

• Schwab CEO Rick Wurster said the brokerage blocks unauthorized AI agents a double-digit number of times each week.
• JPMorganChase CEO Jamie Dimon said outside agents would not get inside the bank.
• Bank of America CEO Brian Moynihan said a human should stay in the loop to approve money leaving an account.
• Citigroup said it does not let third-party agents log in or transact.

All four, per the report, still allow read-only functions such as viewing balances. That detail matters, because read-only access to a balance, a transaction history or a payee list is itself valuable to an attacker, and it is the one thing an agent acting for a customer legitimately needs.

On earnings calls on October 15, U.S. Bancorp's Gunjan Kedia said the bank had not encountered third-party AI cash-sorting agents, while PNC's William Demchak called agents transacting inside accounts "incredibly unsafe." He added: "For that to ultimately take place, there's going to have to be laws passed on liability shifting."

A possible technical link

American Banker also points to a separate flaw called PHP2Shell, found by Wiz Research and Calif in August, which affects core PHP, used by roughly 70% of websites. A patch landed Wednesday, October 21, but some bad actors may have exploited it earlier. The outlet describes reports this week that unauthorized agents had gained access to many sites and platforms, including those linking bank customers to accounts. It does not establish that this flaw explains the bank incidents, and neither do we.

The liability gap

Demchak's remark is the sharpest statement of the problem. Existing consumer protection rules were written for a human, a card or a bank-run channel. When a customer's own assistant misreads an instruction, or a stranger's agent impersonates one, the rules do not clearly say whether the customer, the bank or the agent's maker carries the loss.

Politicians are reaching for answers. At a Fox News live event the same day, Florida Gov. Ron DeSantis said: "Imagine a rogue AI agent liquidating your bank account and investment portfolio." He suggested criminal liability for AI executives as a deterrent. Our tracker of AI safety bills in Congress follows the federal proposals on agent accountability.

Why this is a safety story, not only a fraud story

Banks are the first place where agents with delegated authority meet money and a regulator. Consumer products such as Claude Duet already promise to act for you, with approvals. Whether an agent is authorized is a question of identity and consent, and the industry has no settled way to prove either to a bank.

What would settle it

• Named incidents with dollar figures, from a bank, a regulator or a court filing.
• A rule that says who bears the loss when an authorized or unauthorized agent moves money.
• A standard way for a bank to tell a customer's sanctioned agent from an intruder.

What customers can check now

None of the sources we opened give customer guidance, so this is our reading of the reports, not advice from a regulator. The executives quoted say money movement by outside agents is blocked, and the open exposure is read-only access. Customers who connect any assistant to a bank login should ask their bank whether it permits agent access, review which apps hold their credentials and revoke any they no longer use.

Our position

Treat the report as credible but unproven. The strongest evidence is not the anonymous sourcing but the pattern of denials: executives say agents cannot move money, and nobody says agents cannot look. If you use any agent with account access, prefer read-only permissions, keep the approval step on, and watch for statements from your bank on agent access. We will update this piece when named incidents or a liability rule appear.