OpenAI fixes ChatGPT shared-link account takeover, pays $5,000 bounty

On October 31, 2026, Parameter disclosed an account takeover vulnerability affecting ChatGPT. The attack required nothing more than opening a malicious shared-conversation link: no paste, no login, no additional click. Within roughly 4.5 seconds of the victim opening the link, the attacker could execute JavaScript in the context of chatgpt.com, reading private conversations and the session token. According to the article, the issue was reported on October 18, escalated from P2 to P1 on October 21, and fixed on October 26. The related bounty was $5,000. The piece also offered a short 20-line self-check for similar postMessage listeners in web applications.

The following day, Techmeme picked up coverage from The Hacker News (by Swati Khandelwal), confirming that researchers were paid $5K by OpenAI for reporting the bug. That account described the impact in simpler terms: a malicious ChatGPT share link could run an attacker's code in the logged-in session of anyone who opened it. The two reports agree on the substance. They differ only in timing and depth. Parameter published the full technical walkthrough on October 31, while The Hacker News followed, with Techmeme archiving the item on November 1. As of now, the vulnerability has been resolved and the bounty has been paid. No separate root-cause analysis or follow-up from OpenAI appears in the available coverage.