Research revealed a vulnerability in ChatGPT where a shared link containing a math library payload could execute code in a victim's active session.
OpenAI patched the P1 flaw server-side within eight days.
Security firm Parameter discovered the issue and reported it to OpenAI through its bug bounty program on October 18. The link led to a shared conversation with an interactive graph, which used MathLive, a tool for mathematical expressions.
Initially, the user only had to open the link and click “Preview” to trigger the attack. In a later test, the preview opened automatically. This allowed commands to be executed within the user’s connected session.
Parameter tested the issue in a controlled environment, across different researcher accounts. It did not use the attack on real users. OpenAI fixed the issue in two versions of the ChatGPT website by October 26, blocking messages that allowed the attack. It called it a P1 issue, its most severe category.
↳
READ ALSO
HACKS.GR / RELATED
AI Unlocked
How the rise of OpenAI and Anthropic is changing the pace of funding
↗
OpenAI paid a total of $5.000 for Parameter's report, which was published on October 28. As the fix was made on OpenAI's side, users do not need to change any settings or install anything.
>Stay connected with Hacks.gr
Set us as a preferred source on Google and find us on social media.
Preferred source
↗
end of article
Top comments (0)