The U.S. government has begun implementing a de facto licensing system for frontier AI models, with officials restricting access to advanced systems through formal approvals and informal pressure on developers. Anthropic secured permission to release its Mythos-5 model to selected companies and government agencies following a significant standoff, while OpenAI's GPT-5.6 "Sol" model is rolling out with initial access limited to approximately 20 pre-approved organizations. The arrangement signals a major shift in how frontier AI capabilities are distributed, with geopolitical considerations increasingly shaping regulatory decisions. Meta has faced pressure from U.S. officials to agree to voluntary AI model reviews ahead of public releases, reinforcing what amounts to a government-controlled distribution system for advanced AI. The restrictions reflect broader concerns about frontier AI capabilities and geopolitical implications, though they operate largely through executive pressure and informal agreements rather than formal legislation. This creates uncertainty for developers about approval timelines and requirements. The emerging control regime reveals U.S. government alarm about AI development outpacing regulatory frameworks amid intensifying geopolitical competition. While restrictions currently apply primarily to the most advanced models, their implementation may signal broader regulatory intentions as AI capabilities advance further.