Towards Regulatable AI Systems: Technical Gaps and Policy Opportunities