You must log in or # to comment.
This exploit was discovered by two security guys at the Brave web browser.
Heartbreaking.jpg
I have yet to see any roving Brave infosec mall ninjas react to the fact that they plan to do this agentic AI shit too:
Brave gets credit for finding this exploit in Comet, so well done! But how did they find it? Well, Brave is trying to do its own AI agent web browser. Even though this is a fundamentally stupid idea that cannot be secured.
It would be extremely funny if your prompt injection near the end actually caught someone using the Comet Assistant.
(It probably won’t, given the Venn diagram of “Comet assistant users” and “Pivot to AI readers” is two circles places a mile apart)
