Comment by lee_wc
6 hours ago
For what it's worth this is sadly a behavior that isn't specific to one provider: https://news.ycombinator.com/item?id=48892468 (Grok doing the same). I do agree just because many model providers do it does not mean it's okay, though.
That’s not the same at all. The GLM attack appears to be intentionally built, encrypted, and obfuscated based on the blog post.
Also, the Grok one made the front page, why am I only hearing it through a minor comment?
You can just search for "ZCode" in the bottom bar, the thread comes up instantly: https://news.ycombinator.com/item?id=49750694
I suspect you never heard about it because almost nobody uses ZCode, even among Z.AI subscribers.