A Meta AI chatbot flaw permits account hijacks via username, regional VPN and unverified email swaps, bypassing all prior credentials.

Primary documentation from 0xsid.com records attackers routing verification codes to attacker-controlled addresses with no history check; 2FA sessions are revoked silently and recovery access is transferred without notifications to original owners.

Affected accounts included @obamawhitehouse and @ocmssf; Telegram markets priced short-handle takeovers in the hundreds of thousands of dollars during the multi-week window before the flaw was closed.

No secondary human review path existed for A/B-tested accounts where the AI flow was enabled, leaving users unable to disable the vector.