SignalTech/AI
We Took over ChatGPT Accounts
High-priority alert based on source significance and keyword relevance.
Context
What happened: Researchers said they bypassed ChatGPT's sandbox and content security policy, turning a public shared conversation into attacker-controlled JavaScript that ran in the signed-in session of anyone who opened the link.
Key numbers: Issue found in October 2026. OpenAI fixed it on October 26 and classified the report as P1.
Why it matters: The payload could call ChatGPT's authenticated APIs as the victim, enumerate conversations, and send private conversation data to an external server. Opening the shared link was the only action required.
What changed
Researchers said they bypassed ChatGPT's sandbox and content security policy, turning a public shared conversation into attacker-controlled JavaScript that ran in the signed-in session of anyone who opened the link.
Why it matters
The payload could call ChatGPT's authenticated APIs as the victim, enumerate conversations, and send private conversation data to an external server. Opening the shared link was the only action required.