Model distillation, the process of training smaller AI models using outputs from larger ones, has become a central regulatory battleground in Washington this week, with Silicon Valley and DC lawmakers debating whether and how to regulate the practice. The technique, which involves using outputs from models like OpenAI's GPT-4 or Google's Gemini to train smaller, cheaper models, has been standard practice for years but is now being questioned as potential intellectual property theft. The debate intensified after reports surfaced that several startups have used distillation to create models rivaling those from OpenAI, Anthropic, and Google at a fraction of the cost. AI giants argue they need protection for their research investments, while smaller players warn heavy-handed regulation could cement the dominance of well-funded incumbents. Meta has emerged as an unexpected wildcard, with internal tensions over whether to support restrictions that could limit use of its publicly released Llama models. Lawmakers admit they are struggling to understand the technology, and the debate reveals how reactive AI policy remains. Restrictions on distillation could fundamentally reshape the AI industry's economics by dramatically raising the barrier to entry for AI development.