logoalt Hacker News

phyzometoday at 6:55 PM0 repliesview on HN

Because the author is wrong, and LLMs don't actually work that way. Prompt injection cannot be fixed. Role boundaries are a bandaid you can apply, but attackers can work around it.