Lately I've been trying to detect/mitigate prompt injection attacks. Wrote a blog post about why it's hard: https://alexcbecker.net/blog/prompt-injection.html