OpenAI ChatGPT · Disclosed

OpenAI Fixes ChatGPT Share Link Flaw That Ran Attacker Code

Oct 31, 2026 · 1 min read

A ChatGPT flaw let a malicious share link run attacker code in the logged-in session of anyone who opened it and read their private conversations. Researchers at Parameter reported the issue on October 18, and OpenAI fixed it eight days later, rating it P1, its most serious category. The sources mention no real exploitation.

By HIFY Radar newsroom Updated 10/31/2026 02:56 AM Get this every day →← Back to the feed 1 source

HighVulnerabilityDisclosedPatched

In this article

What happened
Why it matters
What to do
For the technical team
Timeline
Sources

What happened

A ChatGPT flaw let a malicious share link run attacker code in the logged-in session of anyone who opened it and read their private conversations. Researchers at Parameter reported the issue to OpenAI on October 18. The fix shipped eight days later, and the flaw was rated P1, the most serious category.

Why it matters

Anyone who opened the link could have private conversations read by a third party. The risk applied to any user of ChatGPT's chat sharing feature, with potential exposure of sensitive data exchanged in the service.

What to do

OpenAI has already applied the fix, and the sources cite no manual action. As a precaution, avoid opening share links from unknown sources and avoid sharing chats that contain sensitive data.

For the technical team

Affected: OpenAI ChatGPT (links de compartilhamento) · ChatGPT, conversation sharing feature (share links). Versions not cited in the sources.
Exploitation: No exploitation reported in the sources. Flaw demonstrated and reported by researchers at Parameter through OpenAI's bug bounty program.
Fix: OpenAI fixed the flaw eight days after the October 18 report. Specific version or patch not cited; no user action is mentioned.
Vector: Malicious ChatGPT share link. Attacker code ran in the victim's logged-in session when the link was opened, allowing private conversations to be read. Flaw type not specified in the sources.
More details: Rated P1 by OpenAI, its most serious category, according to the source. Timeline: reported on October 18 through the bug bounty program; fixed eight days later.