Anthropic's state-policy lead, Cesar Fernandez, spelled out a regulatory strategy this week that cuts against how most of the industry is playing it. Instead of lobbying for one uniform national framework, the company wants states to compete — each passing rules a notch stricter than the last for the most capable models. It points to bills it has already backed in New York, Illinois, and Massachusetts, which it calls the strongest in the country, and it remains the only major lab to have endorsed California's SB 53 before that became law last year. OpenAI is pulling the other way. Its team is spreading matching, lighter-touch bills across states to build a de facto national standard — an approach reporters have taken to calling "reverse federalism" — while lobbying Congress for preemption that would let Washington override the patchwork. Anthropic's line is that it won't accept federal preemption unless the federal law is at least as strong as what it is pushing at the state level. If you ship on frontier models, this is the fight that sets your 2027 compliance surface: a floor defined by whichever state moves hardest, or a ceiling written in D.C.