There have long been science fiction writers who wrote 'horror stories' about 'intelligent' computers gone berserk, and people worrying about 'dangerous' future artificial intelligence.
What is the bottom line?
Melting pots will never generate an ai that is safe, unless they can destroy all other competing melting pots.
Traditional, language and culture based societies, led always by people who have a consolidated past, will never generate an ai harmful to humanity as long as they can maintain separateness from other similar societies.
One of the biggest deceptions in future societies will be pseudo indigenous people. They will have the clothing of their ancestors, the features, the gene pool, but use the language of their conquerors, think like the melting pot and seem to serve the melting pot against the previous interests of their tribe.
Their ai will have two components. A visible public component ostensibly useful for something practical, and a shadow component, part if their extinguished culture's long term revenge against its conqueror.
Almost everybody in melting pot societies will fall into this role.