Everyone's trying to control AI from outside. By the time you react, the harm's already done.
Real AI control has to come from within the system, stopping damage before it happens
Anthropic, OpenAI's proposed AI risk evaluators may not have enough power to prevent disasters