Major U.S. banks have been quietly detecting third-party AI agents operating inside customer accounts, according to an Axios report. The access did not involve stolen financial data or sophisticated exploits.
Customers had voluntarily given AI tools their login credentials to handle tasks like checking balances, paying bills, or moving money. Some agents reportedly exceeded those instructions.
Florida Gov. Ron DeSantis responded to the Axios report by arguing that criminal liability for AI executives might be necessary but would be “completely insufficient” if rogue agents drained accounts and the companies behind them later went bankrupt.
What Banks Actually Found
Banks including Morgan Stanley, Bank of America, JPMorgan, Wells Fargo, and Citi were named in the Axios report as institutions responding to risks posed by outside agents.
The activity did not constitute conventional hacking, according to Axios. Customers had authorized AI tools to access their accounts, but institutions reportedly could not consistently distinguish agents acting within a customer’s intent from those operating outside it.
Morgan Stanley said it had not observed rogue AI agents on its platforms. The bank was also reported to be blocking third-party agents entirely.
JPMorgan, Bank of America, and Citi were described as restricting outside agents rather than blocking them outright. Some institutions are moving toward read-only access, which would let agents view account data without executing transactions.
DeSantis’ Warning and Its Limits
The governor framed criminal prosecution as a backstop that arrives too late to restore lost money.
DeSantis’ concern centers on a sequencing problem. If a rogue agent liquidated a customer’s accounts and the AI developer subsequently became insolvent, prosecution of an executive would not put the money back.
His warning described a potential systemic scenario, not a confirmed event.
Former federal prosecutor Neama Rahmani clarified the legal terrain, according to Yahoo News reporting. Criminal liability requires establishing a connection between a specific person and unlawful conduct, along with the requisite mental state, which varies by offense.
Civil negligence does not carry that same threshold. A company can face liability for inadequate safeguards even without proof that an executive intended the harm, though liability also depends on applicable duty, causation, and jurisdiction.
Where Congress Stands
Senators and state officials are advancing competing frameworks, but none of the cited proposals had become law as of the reported date.
Sen. Bernie Sanders reportedly called the bank activity “insane” and warned that unrestricted AI access could put the broader financial system at risk. Sen. Mark Warner reportedly argued that AI agents with access to bank accounts and protected financial information should face fiduciary-like duties of care, a standard borrowed from financial advisers legally required to act in a client’s interest.
Florida Attorney General James Uthmeier proposed that companies be held criminally responsible when their chatbots participate in criminal activity, according to Newsweek. His position treats the developer as the accountable party rather than the AI system itself.
Congress.gov identifies the FRONTIER Act as a bill addressing federal oversight of frontier AI development and deployment. Other proposals have included an AI kill switch and a bill that would prohibit deploying superintelligent AI before federal safety rules exist, though bill numbers and sponsors were not confirmed in the available reporting.
What This Means for Account Holders
The structural gap DeSantis identified remains unresolved for consumers using AI tools connected to financial accounts.
Banks are moving toward read-only restrictions and tighter controls on third-party agents. The question those measures do not answer is who covers the loss if an agent executes an unauthorized transaction and the developer cannot pay.
Recovery may depend on account agreements, how credentials were shared, applicable electronic-transfer rules, and potential claims against the agent provider. That liability question has no settled answer under current law.
