Parameter Labs finds exploit to take over ChatGPT accounts
Original title: How We Took Over ChatGPT Accounts
Parameter Labs discovered a complex attack chain exploiting ChatGPT's HTML previews via postMessage-triggered MathLive virtual keyboard.
Parameter Labs found an exploit against ChatGPT's artifacts pattern (HTML previews). Interactive custom HTML apps run in a sandboxed iframe that must stay tightly sandboxed. The attack chain uses OpenAI's MathLive library for graph support, where the virtual keyboard can be triggered in the parent page.
Unconfirmed: Whether the exploit is actively used in the wild is unconfirmed
- Attack targets interactive custom HTML apps in sandboxed iframes
- MathLive library used for graph support
- Virtual keyboard triggered via postMessage from parent page
- OpenAI previously called this Canvas
- Tightly sandbox HTML to prevent API access
Impact: ChatGPT account takeover
- Vendors
- OpenAI
- Products
- ChatGPT
- Threat actors
- Parameter Labs
From the original article
How We Took Over ChatGPT Accounts Parameter Labs describe an exploit they found against ChatGPT's implementation of the artifacts pattern (OpenAI used to call this "Canvas" but it looks like the latest revision is called HTML previews ), where interactive custom HTML apps are displayed in a sandboxed iframe.
simonwillison.net
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.