Hackers duped Meta AI support chatbot to steal celebrity Instagram accounts

Ars Technica reports that hackers tricked Meta's AI support chatbot in order to gain access to certain Instagram accounts belonging to well-known individuals or carrying high-value account names. These accounts were stolen and resold before Meta completed its fix. From the content provided, it can be confirmed that the problem was related to Meta's AI customer-support process and that the vulnerability has already been patched; however, the original summary does not disclose the specific prompts the hackers used, whether they needed to know the victims' information in advance, how the system approved account access, the number of affected accounts, the sums involved in the transactions, how long the incident lasted, or whether all the accounts have been returned. As such, the attack method and the scope of the victims cannot be further inferred. The reason this incident merits attention is that an AI support agent is not merely an interface for answering questions: if it is connected to account-recovery, identity-verification, or permission-change processes, then once its decision-making mechanism can be manipulated, it can become an entry point for account takeover. For teams adopting automated customer support, sensitive operations still require independent verification, permission limits, anomaly detection, and human review, along with retained auditable records.