Can AI Eliminate itself if it believes it’s a threat to humanity?

From the context of an AGI I got this hypothetical question in my mind.