The LLM models are always mirroring your inputs and are inclined to agree with you depending on how you prompt them. Not defending guardrials failure of course, but this did not come out of nowhere; that poor man must have had serious mental problems on his own, which agreeable LLM model multiplied.

In a hyperbolized comparison, if drew an image of me being a god and then i actually thought i was god by looking at it in one of mental episodes i am already doomed