State governments are forming new laws and policies that set "guardrails" to limit AI's potential harms. But officials know their reach is limited.