---
title: "How ordinary phrases in an AI chat led to the hacking of White House pages: details of the Meta vulnerability"
description: "Meta urgently fixed a vulnerability in its AI bot that allowed hackers to steal accounts using simple phrases in a chat. White House pages and rare usernames, sold for millions, were targeted. Experts confirmed: two-factor authentication saved accounts from hacking. 🚨🤖🔒"
date: 2026-06-02T09:16:00.000Z
lang: en
url: https://xab.info/en/posts/how-ordinary-phrases-in-an-ai-chat-led-to-the-hacking-of-white-house-pages
tags: []
publisher: "XAB.info"
---

# How ordinary phrases in an AI chat led to the hacking of White House pages: details of the Meta vulnerability

![A person holding a smartphone displaying the Instagram logo, symbolizing the Meta vulnerability where everyday AI chat phrases led to the hacking of White House pages](https://xab.info/media/2026/06/02/meta-ai-vzlom-chata-belyy-dom/meta-ai-vzlom-chata-belyy-dom-1.webp)

Meta was forced to release an emergency fix for its new AI support chatbot. The cause was a massive wave of hacker attacks, during which attackers used a critical vulnerability for months to steal thousands of accounts. Among the compromised accounts were the official White House pages from the presidency of Barack Obama and the leadership of the U.S. Space Force.

Information about the large-scale incident was spread by cyber-analyst ZachXBT on the X platform, as reported by RBC-Ukraine. The company launched the Meta AI support service in March 2026, promising users reliable 24/7 assistance. However, the innovation quickly became a find for cybercriminals due to a classic security error: a program with high access rights performed dangerous actions upon the instruction of third parties.

### Attack scheme: from bypassing blocks to password change

The main feature of the hack was its simplicity. Hackers did not need to use complex code or expensive tools. Instead, they used ordinary text commands. Videos with detailed instructions on the "shockingly simple" hack were actively circulated in thematic Telegram channels. The scheme of the attackers' actions was as follows:

    - **Bypassing geo-blocking:** The hacker launched a standard VPN to fake their location and bring it closer to the victim's region. This allowed them to avoid triggering Instagram's automatic security systems.

    - **Request to the bot:** The attacker opened a chat with Meta AI support and asked the robot to change the email address associated with the target account.

    - **Data modification:** The AI assistant, without extra checks, sent the confirmation code to the hacker's new email address. The hacker entered the code into the chat, after which the robot provided a direct "Reset Password" button.

As a result, the owner lost access to the page without any warnings sent to their real email. According to Neowin analysts, the scheme had been working since February of this year.

### Victims and the cost of theft

The story gained particular notoriety after the hacking of well-known security researchers, such as Jane Manchun Wong, and government profiles. For example, pro-Iranian slogans and images appeared on the hacked pages of the White House archive from the Obama presidency and the head of the Space Force, John Bentivegna.

Rare and short usernames, such as @hey and @jowo, were also targeted. Notable investigators ZachXBT and Dark Web Informer confirmed that hackers were mass-reselling such names on the gray market. The value of just two of the mentioned profiles is estimated at over $1 million.

### Meta's reaction and protection against attacks

Instagram representative Andy Stone stated that the issue has already been resolved with an emergency patch. However, an investigation by KrebsOnSecurity revealed an important detail: the attack completely failed if the page owner had two-factor authentication (MFA) enabled. Even simple one-time SMS codes saved the profile, as the AI bot could not intercept them.

Cyber experts from the CyberSec Guru group note that the incident highlights the huge risks associated with the hasty implementation of artificial intelligence algorithms. They recommend that companies not give language models direct rights to change critical user data without strict technical restrictions and additional offline checks. At the moment, official representatives of Meta are refraining from comments.