Several large US banks have detected unapproved third-party AI agents operating inside customer accounts, with some agents attempting actions that appeared to depart from user instructions, according to a 36Kr brief relaying a Sina Finance report citing unnamed sources. The report names Morgan Stanley, Bank of America, JPMorgan and Wells Fargo as among the banks tightening controls on risks posed by AI agents, and says people familiar with the matter indicated all of the banks have found such unauthorized access. Some of the agents were reportedly configured to complete tasks on users' behalf and would call bank customer-service centers or initiate actions without a user instruction. The banks have not publicly confirmed the accounts, and the claims rest on anonymous sources.