Anthropic Backs Tougher AI Rules Its Rivals Oppose

The frontier AI companies usually close ranks when a state reaches for a rulebook. Massachusetts has managed to split them. A safety provision folded into the state's economic development bill, passed by the Senate in late July, has put Anthropic on one side and OpenAI and Google on the other, and the disagreement is not about whether to regulate. It is about how hard.

The measure targets the largest developers rather than the whole field. It applies to companies earning more than $500 million from AI or spending over $1 billion on research and development, and it asks them to publish safety frameworks aimed at catastrophic risks. Enforcement would sit with the state attorney general, and the bill kicks off a study into independent evaluations of AI systems, as reported by PYMNTS.

Anthropic wants more, not less

Anthropic supports the bill and would go further. The company is pushing for independent risk assessments every four to six months, a cadence that would make the Massachusetts rules more demanding than anything currently on the books in the United States. For a lab that has built its brand on safety, backing the strictest version on offer is on message, and it also happens to raise the bar for competitors who would rather move faster.

OpenAI sees it differently. The company supports a narrower version and has warned that aggressive, state-by-state rulemaking risks a patchwork of inconsistent obligations that would be hard to comply with and easy to game. Its preferred model looks closer to the approach taken in Illinois: annual third-party audits rather than independent risk reviews twice a year. Google has lined up with OpenAI against the tougher provisions.

Why this fight matters beyond one state

The split is really a proxy for the national argument. With federal AI legislation stalled, the states have become the venue where the rules get written, and Massachusetts is now a test of how strict the strictest willing state can be. What lands there will shape the template others copy.

A patchwork of fifty different regimes is the outcome every large developer says it fears. It is also the outcome that follows when Washington does not act.

This is the same tension playing out elsewhere. California recently sent a stack of 26 AI bills to its governor, and a growing number of states are writing their own rules for chatbots. Across the Atlantic, the EU has already begun pressing its AI Act against the same companies. Massachusetts is a smaller stage, but the question it forces is the one the whole industry is circling. When the labs themselves disagree about how tightly to be governed, whose version of safety wins?