Crafted ChatGPT Share Link Could Have Exposed Private Chats, Now Fixed
OpenAI shipped a server-side fix, so users need take no action to close the gap.
Security firm Parameter showed that a crafted ChatGPT share link could execute attacker code inside a victim's logged-in session, exposing private conversations and actions on the account. The team reported it through OpenAI's bug bounty program on October 18, and OpenAI shipped a fix eight days later, classing it P1, its highest severity. The exploit needed only that the victim open the link and click Preview, and Parameter also built a variant where Preview opened automatically. The chain abused a MathLive keyboard listener that accepted messages from any sender and allowed raw HTML, which let the Preview frame push markup into the main site, with a YouTube-hosted script running inside a frame that inherited the chatgpt.com identity. Parameter says this was a controlled test with no known real-world abuse, and that OpenAI paid $5,000 for the report and rejected the relevant messages in two frontend versions by October 26. Users need take no action because the fix is server-side.
Why it matters
The flaw showed that a link shared in good faith could run hidden code inside a logged-in ChatGPT session, reaching conversations the attacker was never given. Users who open shared links now have a safer baseline because the messages the attack relied on are rejected, and no action is required on their part. Parameter describes the work as a controlled test with no known real-world abuse, so the practical effect is a closed vulnerability rather than a confirmed breach.
Matters or hype?
Does this story matter, or is it hype? Vote before you see what everyone else said.