Anthropic Backs Massachusetts AI Safety Bill, Breaking Ranks with Google and OpenAI
nashnova research
Anthropic publicly supports a Massachusetts bill requiring AI developers to undergo independent risk assessments every four months, while Google and OpenAI oppose it. This means → the AI industry's internal rift over who sets safety rules is now out in the open.
What does this bill actually require?
The Massachusetts Senate embedded AI safety provisions into a broader economic development bill. It requires large AI developers to hire independent assessors to evaluate catastrophic risks in their models every four months.
That is significantly stricter than the Illinois law passed in May, which only requires annual third-party compliance audits. In plain terms = Illinois demands a yearly check-up; Massachusetts wants a quarterly one.
The bill is now with a joint conference committee for final text reconciliation. Senator Barry Finegold says the committee's work will resume after Labor Day, and whether the AI provisions survive remains uncertain.
Why is Anthropic backing it?
Cesar Fernandez, Anthropic's head of U.S. state and local policy, posted publicly after the Senate passed its version on July 24, calling the bill one that "raises the bar on AI safety while ensuring innovation continues."
AI safety groups Encode and the Secure AI Project co-signed a letter of support to conference committee members. This reflects Anthropic aligning openly with safety advocacy organizations — positioning "we welcome tougher oversight" as a brand differentiator.
Why do OpenAI and Google oppose it?
OpenAI policy executive Donnie Fowler and VP of global affairs Ann O'Leary wrote separately to the conference committee. Their core argument: "regulatory fragmentation." A patchwork of state rules, they say, would delay advanced models from reaching state and local governments.
OpenAI's letter warned: "A chaotic patchwork of rules across the country will unnecessarily delay the most advanced network models from reaching state and local officials, at a time when bad actors with powerful AI are already probing for vulnerabilities."
Google's concern centers on states stacking additional testing requirements on top of the Illinois audit baseline. The company argues model-testing standards do not yet exist, making compliance outcomes unpredictable. Google previously delayed rolling out AI features in several European countries over regulatory issues — observers worry similar delays could play out state by state in the U.S.
What is this fight really about?
On the surface it is one state, one bill. Underneath, it is a fundamental dispute within the AI industry over two paths: a uniform federal standard versus state-by-state differentiated regulation.
This means → OpenAI and Google are betting on pushing a relatively permissive federal framework first, pre-empting 50 separate state rulebooks. Anthropic's strategy is the opposite — embrace stricter state standards and build trust by saying "we're not afraid of scrutiny."
In plain terms = if the Massachusetts bill becomes law, it could serve as a template for other states, ratcheting up the regulatory bar one state at a time. That is exactly why the leading AI companies are lobbying so hard on this legislation.
市场有风险，内容仅供研究参考，不构成投资建议。