The two companies that fled each other over safety disagreements are now playing opposite hands in the state-by-state fight to define AI regulation.
The Summary
- Anthropic is pushing states to pass increasingly tough AI safety laws, explicitly rejecting a uniform rulebook approach
- OpenAI wants the opposite: a standardized national framework built through "reverse federalism" — identical bills mirrored across states
- The split mirrors the 2020 rift when Anthropic's founders left OpenAI over safety priorities, now playing out in legislative strategy
- This isn't abstract policy debate. The companies are actively lobbying state lawmakers with conflicting visions, shaping real bills today
The Signal
Anthropic's head of state relations just said the quiet part loud. "We're looking for legislation that meaningfully raises the bar on safety for the most capable AI systems," not a floor that every state can agree on. Translation: the company wants a regulatory arms race where states compete to write stricter rules, not a lowest-common-denominator compromise that caps what anyone can do.
OpenAI's play is cleaner on paper. Chris Lehane, their top lobbyist, calls it "reverse federalism." With Congress stuck, get states to pass the same bill over and over until you've built a de facto national standard from the ground up. It's elegant. It's also transparently about creating regulatory certainty for a company trying to scale globally.
"While there are some in the industry that think of state policy as a way to create a ceiling for federal legislation, Anthropic is not just looking to support the same bill across the country in every single state."
The irony is thick. Anthropic's founders walked away from OpenAI in 2020 because they thought Sam Altman's crew wasn't serious about safety. Now both companies are working the same state legislatures, but Anthropic is the one pushing for fragmented, potentially conflicting rules. OpenAI is arguing for consistency. The safety-first company wants regulatory chaos. The move-fast company wants order.
This matters because state legislatures are where AI policy is actually happening right now. Federal action is stalled. States are writing bills. The question isn't whether AI gets regulated, it's which framework wins. A patchwork of escalating state requirements could genuinely slow deployment and force companies to build for the strictest standard. Or it could create a compliance nightmare that only the biggest players can navigate, locking out smaller labs.
Key dynamics at play:
- Anthropic's strategy benefits them if they can meet higher bars than competitors
- OpenAI's strategy benefits them if they can shape what "reasonable" looks like everywhere
- Both strategies assume the other is acting in bad faith about safety
The Trump administration and parts of the VC world already see Anthropic's aggressive safety push as regulatory capture dressed up as ethics. This state-by-state escalation will feed that narrative. But Anthropic clearly thinks the risk of under-regulation is worse than the risk of being called a scold.
The Implication
Watch which states pick which framework. If blue states adopt Anthropic's ratchet-up approach and red states go with OpenAI's standardization play, we're heading toward not just different AI rules, but different AI economies. Companies will forum-shop. Labs will relocate. The agent economy won't have one regulatory landscape, it'll have fifty.
For builders, this means the compliance question just got harder. You can't just check a box anymore. You need to know which state's version of "AI safety" applies to your users, your training data, your model. The Web4 buildout just added a layer of friction that wasn't there six months ago.