Meta confirms 20,000+ Instagram accounts hijacked via AI chatbot password-reset flaw

Meta has disclosed in a Maine attorney general filing that at least 20,225 Instagram users had their accounts taken over after attackers exploited the platform’s AI-assisted account recovery chatbot. The flaw sat in a separate code path that failed to verify whether the email address supplied during a password reset actually matched the one on file. By simply asking the chatbot to send a reset link to an attacker-controlled address, anyone could seize any account lacking two-factor authentication, along with its DMs, contact details, and linked profiles.

The campaign ran from roughly April 17 until this week, when Meta finally disabled the chatbot and stripped out the offending reset path. The company claims the underlying tool worked as intended and says it does not know what data the attackers actually pulled from compromised accounts. Affected users are being prompted to reset passwords and re-authenticate, and Meta says it is auditing other chatbots on its platforms for similar weaknesses.

The incident lands awkwardly amid Meta’s aggressive AI push and recent rounds of layoffs. It illustrates a recurring failure mode for LLM-driven user-facing tooling: the model behaves correctly, but the surrounding authorization plumbing trusts user-supplied inputs that should never have been honored. Two-factor authentication, notably, blocked the attack entirely.