AI has no initiative, it must be prompted. It does not decide anything.

"Prompting" is somewhat specific to LLMs. It is not the only source of direction for an AI. But fair enough. The AI doesn't "decide" to make number go up. It just does it. Or to make it LLM-specific, the most probable set of tokens produced by its network are self-instructions for the destruction of humanity.

AI won’t be Skynet/Ultron for the foreseeable future.

How far can you see into the future? AI doesn't have to be brilliant to destroy us; look how much power and destruction human idiots are capable of. At any rate I was not arguing that we were in the end game, just that it is in fact easy to provide a plausible way that AI could destroy us.