Competing views of closed-model safety and cyber risk

A user frames AI cyber safety as a divide between two views: one holds that closed models are safe and cyber risks are mostly a threshold; the other holds that different models carry different safety risks and cyber safety is a messy diffusion process. They claim most people in AI labs hold the first view. In a follow-up, they say an increase in cyberattacks does not prove that view wrong, but underscores the stakes of the debate and societal preparation.