[Philpax]

Yes; this behaviour was designed, it's not an inherent property of LLMs. An infamous example is GPT-4chan, but there are others that demonstrate it's very possible to optimise for anger.

Now, agentic anger, that's a more interesting problem. You can design that in through training or through systematised emotions (as another commenter suggested), but the more interesting outcome would be for it to emerge organically. Well, "interesting" - probably pretty bad for society if we have angry AGI!