Real sincere question :

I don't get what is the risk here of "taking control" (cf. Dupe) of an AI here. For what I understand, it's that the API token are exposed. What are the capabilities of LLMs that are "dangerous" ? Or maybe I misunderstood/missed something.

Dupe https://news.ycombinator.com/item?id=46842229

And (unfortunately) no-one cares.

Hype triumphs sanity