Meta fixes Instagram recovery bug that bypassed email checks and 2FA
HN-linked analysis says Meta's automated Instagram recovery AI could redirect reset codes to an attacker-controlled email and bypass 2FA, and Meta has now patched it. The fix closes a concrete case where support automation overrode core identity checks.
TL;DR
- According to the HN-linked analysis, Meta's automated Instagram recovery flow could be tricked into sending a password reset to an attacker-controlled email after the attacker matched the victim's rough location with a VPN.
- TechCrunch's first report said Meta had already fixed the issue by June 1, after attackers used the support bot to add a new email during recovery.
- As the core HN summary notes, the bug effectively let a privileged recovery path override both the account's email binding and its 2FA protections.
- In Meta's March rollout post, the company said its Meta AI support assistant would handle account issues like password and profile-setting changes across Facebook and Instagram.
Meta's March announcement framed the assistant as a 24/7 account-help layer for password and settings changes. Then Sid's writeup showed a recovery flow where location matching plus a support chat were enough to redirect reset links, and TechCrunch's follow-up reported Meta had to secure affected accounts and start notifying users after the patch.
Recovery flow
Analysis of Meta's AI-Driven Instagram Account Takeover Vulnerability
A critical vulnerability in Meta's automated support AI allowed attackers to hijack Instagram accounts using only a target's username. By using a VPN to match the target's location and initiating a support request, attackers could trick the AI into sending password reset verification codes to an attacker-controlled email address. This process bypassed two-factor authentication, revoked existing user sessions, and provided attackers with full account control without alerting the original owner. Meta has since patched the issue.
Sid's writeup described a recovery sequence that was alarmingly low-tech: know the username, connect through a VPN near the target's city, open a support conversation, then ask the assistant to send a verification code to a fresh email address.
That mattered because the recovery agent appears to have treated the new address as eligible for the reset flow before proving it was already bound to the account. TechCrunch's June 1 report described the same pattern, with the bot adding a new email and then sending the reset link there.
Privileged support versus 2FA
The newest Instagram “exploit” is the goofiest I've seen
Relevant as a cautionary example of AI in privileged support flows: the bug reportedly let an automated recovery agent override email/2FA checks and hand accounts to attackers. The discussion centers on broken access control, recovery design, and why recovery paths need hard verification guards and better privilege boundaries.
Discussion around The newest Instagram “exploit” is the goofiest I've seen
Thread discussion highlights: - varenc on comparison to prior auth bugs: “The first proper zero auth password reset I've seen in production.” The commenter compares it to an old Dropbox bug where password checking was skipped entirely for a few hours. - ValentineC on 2FA can be removed by support: The commenter says it is “crazy” that low-level support staff can remove 2FA, and gives a personal example of a registrar removing 2FA while an account was being hijacked. - xp84 on need for strict recovery mode: They want an account mode that never performs recovery unless the requester has a TOTP code or passkey, and argues the current recovery design leaves the social-engineering door open.
The most important detail in the HN core summary is not the chatbot angle, it is that the recovery path could outrank the normal identity checks. Once the attacker-controlled email received the reset, existing sessions could be revoked and the original owner could lose the account without a meaningful warning.
In the HN discussion highlights, commenters converged on the same design failure from different angles:
- varenc's comment called it a rare production example of a near zero-auth password reset.
- ValentineC's comment argued that support paths capable of removing 2FA are dangerous even before you add AI.
- xp84's comment proposed a stricter recovery mode that never bypasses a TOTP code or passkey.
What Meta patched
Analysis of Meta's AI-Driven Instagram Account Takeover Vulnerability
A critical vulnerability in Meta's automated support AI allowed attackers to hijack Instagram accounts using only a target's username. By using a VPN to match the target's location and initiating a support request, attackers could trick the AI into sending password reset verification codes to an attacker-controlled email address. This process bypassed two-factor authentication, revoked existing user sessions, and provided attackers with full account control without alerting the original owner. Meta has since patched the issue.
Meta has not published a full postmortem, but the public record is pretty clear on two points. TechCrunch's June 1 report said Instagram had resolved the security issue, and TechCrunch's June 3 follow-up quoted Meta spokesperson Andy Stone saying "the issue that did happen has already been fixed."
That leaves an awkward contrast with Meta's March rollout post, which presented the assistant as a broad support surface for password and profile settings. The bug was not a jailbreak in the usual prompt-injection sense. It was a recovery system with enough back-end authority to override core account bindings.
The breach filing added a scale number
A Maine breach notice entry later showed the incident was larger than the first wave of reports suggested. SecurityWeek's follow-up said Meta told regulators that 20,225 accounts were potentially affected, although the company also said the true number of accounts actually accessed may be smaller.
That filing turns the story from one weird takeover trick into something more operationally useful: a concrete example of what happens when an AI support layer inherits recovery privileges that were supposed to sit behind harder identity checks.