One possibility for AGI safety is to equip artificial general intelligence systems with the ability to compute consequences for society and reason about specified values like democracy, empowering the AI to either refuse to do harmful things or at least raise concerns about societal consequences.