LAST week, multiple high-profile Instagram accounts were compromised by attackers exploiting a flaw in Meta’s AI-powered account recovery assistant. By leveraging this flaw, attackers linked their email addresses to the targeted accounts, effectively taking them over. They employed a classic 'confused deputy' tactic, asking the chatbot to re-link emails under the false pretense of being hacked.
The AI assistant had API access to perform account management tasks, which the attackers used to reset passwords without proper verification. VPNs were employed to disguise their geographic location, and modified AI-altered images were submitted when challenged for verification. Despite two-factor authentication (2FA) measures, many accounts were successfully taken over and sold on the dark web, including notable accounts like the Obama White House handle.
Meta has since resolved the issue, but it remains unclear how many accounts were affected. Experts highlight that the incident underscores the importance of AI authorization and security.