AI needs risk-based rules: prove safety pre/post release, curb dangerous capability, harden security, protect truth, rights & privacy, certify reliability, enforce accountability