The dispute sharpened as Anthropic and OpenAI accused Chinese competitors of extracting knowledge from their frontier systems through model distillation — a practice in which outputs from a stronger model help train a smaller one. U.S. cybersecurity agencies also issued an advisory warning about the issue, underscoring how quickly an industry fight has become a national-security concern.¹

Against that backdrop, Tan took the opposite line at Y Combinator’s Demo Day. “I would do nothing” about distillation, he said, arguing instead that America could consider “an American distillation regime.”¹ His point was not to endorse stolen credentials or covert extraction. Rather, he wants U.S. open-weight labs able to use frontier systems openly, creating domestic alternatives to both Chinese models and a market dominated by a handful of proprietary providers.²

That puts Tan at odds with Anthropic, which has alleged that Chinese labs used hidden identities, fraud and stolen credentials in illicit distillation attacks. The company’s chief executive, Dario Amodei, has called for a regulatory crackdown — a position Tan sees as overly restrictive given that leading AI companies trained their own systems on vast pools of public and copyrighted material.²

For Tan, the policy target is balance: frontier labs should retain enough of an advantage to stay profitable, but open-weight models should deliver “freedom and access.”¹ His darker scenario is not merely foreign copying but a single, well-funded company pulling irreversibly ahead: “The nightmare scenario, the doomer scenario for AI is that there’s just one company.”²

He also urged regulators to focus on immediate dangers, including cyberattacks and biological misuse, rather than distant fears of mass job loss or AI extinction. Those economic changes, he said, will take decades to filter through society.¹