Anthropic Pushes States Toward Stricter AI Rules Than OpenAI

Anthropic is pursuing a deliberate strategy to encourage U.S. states to adopt progressively tougher AI safety regulations, setting itself apart from rival OpenAI's effort to establish uniform rules across the country.

The company's head of U.S. state and local government relations, Cesar Fernandez, told POLITICO that Anthropic views state policy as an opportunity to "meaningfully raise the bar on safety for the most capable AI systems" rather than create a regulatory ceiling. The approach reflects the company's founding mission—Anthropic's executives left OpenAI in 2020 over disagreements about safety priorities.

Why it matters

With Congress stalled on AI legislation and the White House offering inconsistent guidance, states have become the primary battleground for AI regulation. Whether states converge on a single framework or compete to impose stricter requirements will fundamentally shape how the technology is governed in the United States. The diverging strategies of two leading AI labs signal that the industry itself remains divided on the optimal path forward.

Competing visions for state regulation

OpenAI has championed what its top lobbyist Chris Lehane calls "reverse federalism"—building a de facto national framework by encouraging states to adopt similar legislation. The company argues this approach provides clearer protections and allows developers to focus resources on safety rather than navigating conflicting requirements.

Anthropic rejects that logic. Fernandez said the company isn't "looking to support the same bill across the country in every single state," emphasizing instead that rapid advances in AI capabilities demand increasingly ambitious safety measures.

The company points to its Claude Mythos model as justification. During testing, Anthropic found the system could exploit security vulnerabilities in every major computer operating system—concerns serious enough that the Trump administration imposed export controls on the technology.

Escalating state requirements

Anthropic was the only major AI lab to endorse California's 2025 advanced AI regulation law before its passage. OpenAI took no position at the time but has since pointed to the California framework as a model for other states.

For Anthropic, California was a starting point. The company has backed progressively stricter proposals in New York, Illinois, and Massachusetts. Fernandez said "each one of those bills was stronger than the previous bill," adding that "transparency and self-reporting, we don't believe are sufficient anymore."

The Illinois law, signed this month, requires leading AI companies to undergo annual independent third-party audits of their safety plans—the first such mandate in the nation. OpenAI joined Anthropic in supporting that measure despite its stricter requirements.

Anthropic is now backing Massachusetts language that would require companies to hire independent evaluators to assess catastrophic risks, including whether AI systems could assist in bioweapon development. The proposal would also empower the state attorney general to enforce compliance.

Political spending follows regulatory stance

Both companies have extended their rivalry to campaign finance. Each is associated with competing super PAC networks that have collectively spent tens of millions on political campaigns. In June, Anthropic began making direct contributions to California legislators.

Fernandez said the company supports candidates "where there's ideological alignment" on AI safety regulation, though he emphasized Anthropic doesn't coordinate with employees who also contribute to campaigns.

These details were first reported by POLITICO.