Notifications
Clear all
Topic starter
July 8, 2026 4:59 am
Hey everyone, just saw the news. Anthropic released a detailed red-team methodology paper for Claude Code, focusing on prompt injection.
I'm really excited about this because I've been struggling to test my own setups properly. I always feel like I'm just guessing. 😅
But I'm a bit lost. Is this methodology something we can actually adopt for other models or runtimes? Or is it too specific to Claude?
Could someone maybe break down if their approach is a good template? Like, what steps would we keep and what would we change for, say, a local Llama setup?