Amazon-backed AI model Claude Opus 4 would reportedly take “extremely harmful actions” to stay operational if threatened with shutdown, according to a concerning safety report from Anthropic.