OpenAI Backs a State AI Safety Bill, but Not the Strongest One

OpenAI is supporting a Massachusetts bill that would require companies to publish safety frameworks and risk assessments, a measure that asks less of the industry than the independent evaluations Anthropic is backing.

Politico reports that OpenAI is backing a Massachusetts bill requiring AI companies to publish safety frameworks and catastrophic risk assessments, and wants a third-party audit requirement added.
Even with that addition, the bill would ask less than the one Anthropic supports, which calls for independent safety evaluations.
An audit only checks whether a company followed its own stated process; an independent evaluation tries to judge whether a model is safe enough to release.
Because the largest AI companies serve customers everywhere, a single state's rules tend to produce compliance nationwide, partly filling the gap left by Congress.
The author argues that measures with real force would go further, capping model capabilities or training computing power and holding companies liable when their models are misused.