The problem with these conversations is people discuss it as singular, AI.

It should be discussed how will AIs interact with each other?

What are the goals of someone who is intelligent?

It's likely that wiping out anything would first mean it would be threatened. AIs in these scenarios would sure decentralize their points of failure - rendering any supposed threat as entertainment value.

Also there could be negating AIs which tamper with each other - thus spending resources to combat other AIs is more likely than silly humans.