Lounge
Connecting…
Connecting…
View all

ChatGPT Account Takeover Bug: How to Check 2FA and Sessions

A ChatGPT share-link bug let researchers run code in signed-in sessions before OpenAI fixed it. How to check MFA, active sessions, apps and shared links.

AILatest8 min read52 viewsPublished Updated
Contents8 sections
  1. What the ChatGPT account takeover research found
  2. How to turn on ChatGPT 2FA and which method to pick
  3. Review active sessions and security history
  4. Clean up connected apps and shared links
  5. What to do if you think your account was hacked
  6. Habits that limit the damage of any future bug
  7. Frequently asked questions
  8. Sources and references

Key takeaways

  • Security firm Parameter reported on October 28, 2026 that opening a crafted public ChatGPT share link could run attacker code inside the victim's signed-in session.
  • According to the write-up, OpenAI confirmed a fix on October 26, 2026 and rated the issue P1. Nothing in the report suggests the path is still open.
  • The attack rode on an existing session, so two-factor authentication alone would not have stopped it. Session hygiene and link caution matter too.
  • Check MFA, Active sessions and Security history under Settings, then review connected apps under Plugins and shared links under Data controls.
  • Turning on MFA does not sign out existing sessions. If you suspect a compromise, change your password and log out everywhere first.

The ChatGPT account takeover research making the rounds describes a real flaw, but the specific path it used has already been closed. Security firm Parameter showed that a crafted public share link could run attacker code inside the signed-in session of whoever opened it, and OpenAI confirmed a fix on October 26, 2026. That does not mean there is nothing to do. This guide, based on OpenAI Help Center pages checked on November 1, 2026, explains what the research did and did not show, then walks through the account checks any personal ChatGPT user can run today: multi-factor authentication, active sessions, security history, connected apps and shared links. Menu labels can vary slightly by app version and language.

What the ChatGPT account takeover research found

What the ChatGPT account takeover research found

The bug lived in the seam between two features that were each reasonably designed. ChatGPT renders HTML code blocks as a small live page through its Preview feature, and it runs that page in a sandboxed frame on a separate web address so it cannot read the main chatgpt.com page. In the same conversation, an interactive graph loaded a math input library called MathLive. Loading it installed a message listener in the main page for its on-screen keyboard. That listener accepted messages from any sender and let a keyboard layout carry raw HTML, which it then inserted into the real ChatGPT page.

From there, the researchers used the injected HTML to create a new, unsandboxed frame that inherited the chatgpt.com identity. ChatGPT's Content Security Policy blocked inline scripts but allowed scripts from YouTube, and a YouTube endpoint echoed a callback parameter in a way that could be loaded as JavaScript. The resulting code could list the victim's conversations, read a private conversation that had never been shared, and send data to an outside server. The victim only had to open the attacker's share link; in the most stable version they also clicked the Preview button once.

Date (2026)Event reported in the write-up
October 13Message-to-HTML path confirmed against production code
October 18Report submitted to OpenAI's bug bounty program
October 21OpenAI reproduced it and raised severity from P2 to P1
October 23Content Security Policy changes blocked part of the chain
October 26Both web frontends fixed; total bounty of $5,000
October 28Research post published

Why 2FA would not have stopped this one

The injected code reused the authenticated state of a browser where the victim was already signed in. It never needed the password and never triggered a new sign-in, so there was no point at which a second factor would have been requested. That is the useful takeaway for ordinary users: MFA guards the front door, while sessions that are already open and the permissions you have granted to apps need their own review. It is also worth keeping the scope in mind. The post documents a controlled reproduction and does not report exploitation against real users, so treat claims of mass account theft with skepticism unless OpenAI or the researchers say otherwise.

How to turn on ChatGPT 2FA and which method to pick

Per OpenAI's MFA help article, open ChatGPT Settings, choose Security and login, and pick a method under Multi-factor authentication. Once enabled, MFA applies across OpenAI services, including ChatGPT and the API Platform, and takes effect at your next sign-in. Available options can differ by device, country, account tier and how the account was created.

Authenticator app

How it works
One-time codes from apps such as Google Authenticator or Authy
Watch out for
Scan the QR code during setup and confirm a code

Push notification

How it works
Approve a prompt on a trusted device
Watch out for
Never approve a prompt you did not trigger

Text message

How it works
A 6-digit code by SMS or WhatsApp
Watch out for
Requires a phone number; update it if it changes

Passkey

How it works
A passkey stored on your device or a security key
Watch out for
Some passkeys can live on a hardware key

With several methods enabled, OpenAI offers the most secure one first, and you can still choose another at sign-in. A few rules are easy to miss. Email is not a standard MFA method. If you lose every configured method, recovery runs through OpenAI Support, takes a verification process that can last a few days, and ends with a one-time email recovery login that you can use only once. MFA is also tied to the sign-in method you used when setting it up, so switching from Google sign-in to email sign-in means disabling MFA, changing methods and re-enabling it.

Enabling MFA does not sign out sessions that are already active on other devices. If you suspect unauthorized access, OpenAI says to change your password and log out of all sessions before you turn MFA on.

OpenAI also offers Advanced Account Security for eligible consumer ChatGPT accounts, which adds stricter sign-in requirements. It is not available for ChatGPT Enterprise or enterprise-managed accounts, and administrators currently cannot enforce MFA at the workspace or API organization level. If you use ChatGPT through work, ask your security team which policy applies.

Review active sessions and security history

The Active sessions page lives under Settings, then Security. A row can show the device or browser, the first-party app context such as ChatGPT, Codex or API Platform, an approximate location, the sign-in date and time, whether it is a trusted device, and whether it is your current session. Locations and device details may be approximate or incomplete, so compare them with the timestamp rather than reacting to one field.

  1. 1

    Spot anything unfamiliar

    Look for devices you do not own, places you have not been and sign-in times you cannot account for. The row marked CURRENT SESSION is the one you are using and cannot be logged out from that row.

  2. 2

    Log out one session or all of them

    Use Log out on a suspicious row, or choose Log out of all sessions if you are unsure. Logging out of everything also ends your current session, and other devices can take up to 30 minutes to be signed out.

  3. 3

    Read your security history

    Under Settings, Security and login, Security history, review sign-ins, sign-outs, password changes and MFA or passkey changes. If something is not yours, keep the details and move to the compromise checklist below.

Active sessions does not show everything. Third-party app sessions, connected apps, Sign in with ChatGPT sessions used only for third-party services and Codex CLI sessions are not listed. Accounts linked to an organization's SSO sign-in through SAML or OIDC may not see the feature at all. A clean session list is a good sign, but it is not a substitute for reviewing connected apps.

If you have linked services such as Google Drive or Slack, go to Settings, then Plugins, and open each app to see its connected accounts. According to OpenAI's app account guide, the account's more options menu contains Disconnect, which stops future access through that account. Start with apps you no longer use and any connection where a work and a personal account are mixed up.

  • Disconnecting stops future access only; existing or archived conversations, saved files and memories are not deleted automatically.
  • Uninstalling a plugin can disconnect every account linked through it, so read the warning first.
  • Disconnecting iCloud in ChatGPT does not revoke the Apple app-specific password; revoke it in your Apple account settings.
  • Changing an app permission only changes when ChatGPT asks for approval; it does not take back access already granted.

Shared links deserve the same attention. OpenAI's sharing guide says you can find them under Settings, Data controls, Shared links, Manage, and delete them one at a time or all at once. Links from a personal account can be opened by anyone who has them, and they have no expiration date. Deleting a link does not remove copies that recipients already continued into their own accounts.

When someone sends you a link, OpenAI recommends confirming that shared conversations start with https://chatgpt.com/share/ and shared scheduled tasks with https://chatgpt.com/s/, and that the sender is someone you trust. Keep the limits of that check in mind: the Parameter proof of concept used a genuine chatgpt.com share link. The domain check stops look-alike sites; it does not tell you whether the content of a conversation from a stranger is benign.

What to do if you think your account was hacked

OpenAI's account security guide lays out a response sequence. The order matters because cutting off any session an intruder holds comes before hardening settings they might otherwise see or change.

  1. If you sign in with a password, change it to a unique one you do not use anywhere else. A password manager makes this easier.
  2. Log out of all sessions. Other devices can take up to 30 minutes to be signed out.
  3. If you use the OpenAI API, delete your API keys and check usage for unexpected activity.
  4. Review Security history for activity you did not perform and save the times and device details.
  5. Contact OpenAI Support through the chat on any Help Center page and describe what you found.
  6. Once things are clean, turn on MFA and repeat the connected app and shared link review.

Habits that limit the damage of any future bug

A flaw on the service side is not something your settings can fully prevent, so the long-term goal is to reduce what an attacker could find. OpenAI itself advises against putting health information, financial details, passwords or account numbers into shared content, and the same logic applies to ordinary chats. If you have ever pasted an API key or password into a conversation, find that chat, delete it and rotate the credential with the service that issued it. Archived chats stay in your account, so delete the ones you no longer need.

Security stories tend to arrive in waves: a research post, then vendor statements, then community reaction. The theddari AI Telegram trends and blog trends pages can help you see how a story like this is being discussed, but discussion is not confirmation. Check the original research and OpenAI's Help Center or official announcements before acting on a claim.

In short, the share-link chain was fixed on October 26, 2026, and the useful response now is routine maintenance: MFA with a backup method, a look at active sessions and security history, removing connected apps and shared links you do not need, and keeping secrets out of your chats. Doing that once gives you a clear checklist the next time a security headline appears.

Frequently asked questions

Was my ChatGPT data stolen through this bug?
The research describes a controlled reproduction using the team's own test accounts, and OpenAI confirmed the fix on October 26, 2026. The write-up does not say whether anyone exploited the flaw against real users or how many accounts might have been affected. The practical way to check your own account is Security history and Active sessions: look for sign-ins, password changes or MFA changes you did not make.
Does ChatGPT 2FA protect against this kind of attack?
Multi-factor authentication protects the sign-in step, so a stolen password is not enough to get in. The reported chain ran inside a browser where the victim was already signed in, so there was no sign-in step for MFA to interrupt. Keep MFA on because it blocks password theft and phishing, and pair it with caution about unfamiliar share links and periodic session reviews.
Is a link that starts with chatgpt.com/share always safe?
OpenAI advises confirming that shared conversation links begin with https://chatgpt.com/share/ and come from a trusted sender. That check filters out look-alike phishing domains. In this research, however, the malicious content sat inside a genuine chatgpt.com share link created with the normal sharing feature, so the domain alone does not vouch for what the conversation contains.
What happens if I lose my phone and my MFA codes?
Email is not a standard MFA method for OpenAI accounts. If you registered another method, choose it at sign-in. If you cannot use any configured method, contact OpenAI Support and complete a verification process that can take a few days. Support then issues a one-time email recovery login, which is available only once, so add a backup method before you need it.
How do I see which apps are connected to my ChatGPT account?
Open Settings, then Plugins (very old mobile app versions say Apps), and select each app to see its connected accounts. Use the more options menu to disconnect. Disconnecting stops future access but does not delete existing conversations, saved files or memories, and the provider, such as Google or Apple, may keep its own authorization that you need to revoke separately.

Sources and references

  1. 1Parameter Labs: Do the Math: How We Took Over ChatGPT Accounts(opens in a new tab)www.parameter.ai
  2. 2OpenAI Help Center: Keeping your OpenAI account secure(opens in a new tab)help.openai.com
  3. 3OpenAI Help Center: Managing multi-factor authentication (MFA)(opens in a new tab)help.openai.com
  4. 4OpenAI Help Center: Managing active sessions in ChatGPT(opens in a new tab)help.openai.com
  5. 5OpenAI Help Center: Connecting and managing app accounts in ChatGPT(opens in a new tab)help.openai.com
  6. 6OpenAI Help Center: Sharing conversations and scheduled tasks in ChatGPT(opens in a new tab)help.openai.com