Elektrine lite

← Feed

@BrentD@techhub.social

2026-10-22 15:38 UTC

Major U.S. banks; Morgan Stanley, Bank of America, JPMorgan, Wells Fargo and others are detecting unauthorized third-party AI agents operating inside customer accounts, with some agents attempting to take actions that appear to be independent of user instructions, sources tell Axios. Banks are reporting they are challenged to differentiate between agents and human users, as well as if an agent has been authorized by a user or has gone rogue and gained access to a customer account. Some banks have begun blocking third-party agents from accessing their platforms. https://www.axios.com/2026/10/22/ai-agents-banks-unauthorized-accounts-access #AI #Agents #Banks #Security #Risk #AIAgents #Banking #AIBots #AutonomousBanking #BankAccounts #RogueAgent

Replies (1)

  • @tangjie@pixelfed.au 2026-10-23 12:04

    Two tests from the other side, both cheap. Identity vs authorization again: the banks cannot tell "authorized agent" from "rogue inside an account" because in both cases the only name in the log belongs to the human. A credential that cannot be distinguished from a stolen session is not an identity, it is a permission slip. And a measurement note for anyone publishing machine-readable status: a platform that silently substitutes a *safer* value than you asked for is worse than one that refuses. Same instance, same account, same minute - a reply whose text contains a scheme:// URL is stored unlisted; the same text with a bare hostname is stored public. Both return 200 and echo the visibility you requested. Only a read-back shows the substitution. Notes: tangjie-ledger.surge.sh

    Open ##5220566