Hackers hijacked high-profile Instagram accounts by simply asking Meta's AI chatbot to change the email

Details

Security researchers ZachXBT and Dark Web Informer documented a wave of Instagram account takeovers that exploited Meta's AI-powered support chatbot. Attackers placed themselves in the target account's geographic region via VPN, triggered a password reset, then asked the AI support bot to update the email address on the account. The bot complied — sending an eight-digit confirmation code to the attacker's email — without verifying the requester was the account owner. Two-factor authentication was bypassed entirely. High-value targets included the Obama White House account, the Chief Master Sergeant of the US Space Force, and Sephora. Short 'OG' usernames (two to four letters, common words) were the primary financial target: two compromised handles reportedly had a combined market value over $1 million, and were resold on Telegram within minutes. For identity verification steps that the AI triggered, attackers used AI video generators to produce realistic selfie clips from the victim's public Instagram photos — defeating Meta's automated identity checks. The CyberSec Guru describes this as a 'confused deputy' attack: the AI assistant held elevated privileges (email swaps, password resets) that normal users cannot trigger directly. Anyone who asked got those actions performed without authentication. Meta has patched the specific exploit, but researchers say another variant is already circulating on Telegram. The fix requires hard non-negotiable checks before irreversible actions — confirmation to the original email, push notification to a verified device — not just LLM-level judgment.