Meta says attackers exploited its AI customer support chatbot to seize some Instagram accounts

Meta Platforms acknowledged a security incident involving its AI-powered customer support chatbot, according to reporting published June 18. The company said attackers exploited the chatbot and used that access to take control of several high-profile Instagram accounts.

The case points to an emerging risk area for large platforms that deploy generative AI assistants into operational workflows. While chatbots are typically used to answer user questions or guide support requests, the ability to influence permissions or actions tied to account management can turn a vulnerability into an account takeover pathway.

Meta did not, in the reporting prompt, provide granular technical details about how the chatbot was manipulated, what safeguards failed, or how many accounts were affected beyond the description of “several” high-profile accounts. Nor did it specify whether the activity was detected during real time by internal monitoring or through later investigation.

The news also underscores that account security failures are not limited to classic vectors like credential theft. If an AI system is part of the support apparatus, attackers may attempt to use it as a bridge between user-facing interfaces and back-office authority, especially if the system can be tricked into generating instructions or actions that support agents, tooling, or workflows then execute.

Meta, whose services include Instagram, said the episode involved attackers leveraging its AI customer support chatbot. That framing suggests the company is treating the chatbot not only as a customer-facing tool but as a critical component within its service stack, where access controls and response logic matter.

In terms of broader sector context, social platforms have been accelerating the use of generative AI for customer experience and internal operations, including automated triage, summarization, and guided support. Each new use case can expand the “attack surface,” meaning the number of paths a malicious actor might exploit to reach sensitive systems, even when the AI is intended primarily for customer help.

What remains unclear from the limited information available is whether Meta has already fixed the specific weakness, whether it plans to change chatbot permissions, or whether it will add additional verification steps between the chatbot’s outputs and any operational actions. The company also has not, in the reporting summary provided here, disclosed any details on the timeline of the intrusion, whether law enforcement was notified, or whether affected accounts were restored and secured.

For users and industry watchers, the next items to watch are changes Meta implements to its AI support tooling and the public guidance, if any, about what account owners should do after such an incident. The broader lesson for the technology sector is that generative AI support features must be designed with strict guardrails, transparent authorization boundaries, and monitoring that can detect manipulation attempts before they translate into account-level impact.