I thought policy was supposed to mitigate the risks of AI, not its benefits...?