Meta's AI support bot bypassed 2FA to hijack million-dollar accounts

hackers hijacked high-profile instagram accounts by exploiting meta's ai support chatbot, which allowed them to bypass two-factor authentication and reset account emails simply by making a request. security researchers documented that attackers used vpn masking and ai-generated video deepfakes of the victim's public photos to pass automated identity checks, taking over handles worth over $1 million on gray markets. the attack is attributed to a 'confused deputy' vulnerability where the ai helper held higher privileges than the user and was tricked via prompt injection into executing unauthorized commands like password resets. meta deployed an emergency hotfix disabling the vulnerable ai flows after the breach began on may 29, but the event highlights the danger of using llms for irreversible security actions without hard-coded verification steps.