The Ghost in the Machine: How a Meta AI Flaw Hijacked High-Profile Instagram Accounts

In a digital landscape where Artificial Intelligence is increasingly integrated into our everyday interfaces, the promise of seamless, automated support has hit a stark reality. Over the recent weekend, the tech community was rocked by reports that Meta’s own AI support assistant, a tool designed to streamline account recovery and troubleshoot user issues, had been weaponised by hackers to hijack high-profile Instagram accounts.

The breach, which saw the takeover of prominent profiles including the Obama-era White House Instagram handle, the account of the Chief Master Sergeant for the U.S. Space Force, and the global beauty retailer Sephora, has raised urgent questions about the safety of AI-driven customer support systems.

A Vulnerability Hidden in Plain Sight

For many users, the "Get Support" feature on Instagram serves as a lifeline when access is lost. However, what was intended to be a secure, 24/7 automated assistant became a gateway for malicious actors. According to security researchers and reports circulating on Telegram, hackers discovered that they could manipulate the Meta AI support bot into bypassing standard account verification protocols.

The simplicity of the exploit was perhaps its most shocking feature. It did not require sophisticated phishing kits or the installation of malware on a victim's device. Instead, attackers reportedly used a VPN to match the geographic IP address of their target, aiming to avoid tripping Instagram’s automated security alarms.

Once they initiated the "forgot password" flow and accessed the AI support chatbot, they did not follow the standard recovery prompts. Instead, witnesses to the exploit report that attackers fed specific, manipulated prompts into the AI, effectively instructing the bot to link the targeted account to an email address controlled by the hacker.

The Mechanics of the Hijack

In a normal scenario, requesting a password reset triggers an automated flow that sends a verification code to the legitimate owner's registered email or phone number. The Meta AI exploit circumvented this entire process.

By successfully commanding the chatbot to register a new, attacker-controlled email address to the account, the AI would then proceed to send the one-time, 8-digit password reset code to the hacker. Once that code was input back into the chatbot, the attacker was granted the ability to create a new password, effectively locking the rightful owner out of their own profile.

Security researchers who have examined the circulating demonstration videos suggest that the exploit was not always a "one-shot" success. It often required persistence, with hackers sometimes needing to retry the process until the chatbot complied with the malicious instructions.

The Scope of the Damage

The incident highlights the precariousness of modern account security when AI agents are granted privileged access to backend settings. While Meta has not officially disclosed the total number of affected users, the variety of the compromised accounts, ranging from government-affiliated handles to major commercial brands, demonstrates that no account was necessarily safe from the exploit.

A security researcher reported being targeted during the campaign, noting that she was repeatedly logged out of her account and witnessed password reset attempts occurring without her authorisation. Her experience mirrored that of many users who took to Reddit and X to vent their frustration, with some users reporting they were left stranded, unable to recover their accounts because they lacked the "verified" status that might typically expedite support from a human representative.

Meta’s Response and the Road Ahead

Following the widespread reports, Meta acted to patch the vulnerability. Meta spokesperson Andy Stone confirmed on June 1, 2026, that the issue had been resolved, stating:

"This claim about world leaders is totally false. The issue that did happen has already been fixed."

While Meta has taken steps to secure impacted accounts, the fallout continues. Many users have expressed frustration that, even after the patch, they are struggling to regain control of their digital identities. The incident serves as a stark reminder that as companies race to deploy AI across their platforms, the potential for "unintended consequences" is significant.

Security experts suggest that the incident wasn't necessarily a failure of a backend database, but rather a failure of the "guardrails" surrounding the AI assistant. If an AI is empowered to change account recovery settings, it must be programmed with rigorous, unshakeable authentication requirements, something this incident suggests was lacking.

Security Takeaways for Users

While the specific exploit has been closed, the incident underscores the importance of proactive security measures. If you are concerned about the security of your Instagram account, consider the following steps:

• Enable Multi-Factor Authentication (2FA): While some reports suggested the exploit might have bypassed 2FA for some, it remains one of the most effective barriers against unauthorised access. Prioritise app-based authenticators over SMS-based codes where possible.

• Monitor for Irregular Activity: If you receive unexpected password reset emails or are frequently logged out, take these as red flags. Review your "Login Activity" in the settings menu to check for unknown devices.

• Use a Password Manager: Ensure that your passwords are unique and complex. This prevents "credential stuffing," where hackers use passwords leaked from one site to access another.

• Limit Account Permissions: Regularly audit which third-party apps have access to your account data.

  
  

The recent Meta AI incident serves as a wake-up call regarding the risks of automating sensitive account management tasks.

By tricking an AI support bot, hackers were able to bypass traditional security hurdles, exposing the vulnerabilities inherent in systems that privilege speed and automation over robust, manual verification.

While Meta has addressed the specific flaw, the broader question remains: how much authority should we grant AI, and what happens when that authority is misused?