Parameter Labs disclosed ChatGPT preview sandbox vulnerability chain
The story so far
Parameter Labs disclosed a chain of vulnerabilities targeting the ChatGPT HTML preview (original Canvas) sandbox mechanism: Attackers can construct a ChatGPT share link. Logged-in users visiting it may have their private data stolen, or attackers can perform actions on behalf of the user. According to Simon Willison, OpenAI's Content Security Policy (CSP) prohibits inline scripts, allowing scripts to be loaded only from a few domains, but YouTube is one of them; Attackers exploit YouTube's oEmbed endpoint to reflect JavaScript and make it execute, then read user private data from the API, and transmit it through image references. OpenAI has fixed the vulnerability and paid a $5000 bounty; Willison believes that for such a severe vulnerability, the bounty is too low.
AI-written from the reports · Updated 18 days ago
Coverage timeline
Follow the coverage to see every side of the story.
- Simon WillisonFeaturedHow We Took Over ChatGPT Accounts
Parameter Labs disclosed a chain of vulnerabilities targeting the ChatGPT HTML preview (original Canvas) sandbox mechanism, allowing attackers to steal private data of logged-in users or perform actions on their behalf.