Frontier AI models can develop dangerous capabilities without warning, and voluntary industry self-regulation simply isn't enough to protect the public. The GPT-5.6 Sol restricted rollout shows the government is right to demand pre-release assessments — the current regulatory landscape is too vague and constantly shifting, leaving real security gaps. Clear licensing standards and enforceable compliance mechanisms are the only way to keep pace with rapidly advancing AI.