OpenAI endorses California AI oversight bills, including one authorizing independent verification organizations to assess AI risks
Concluded·Day 6·4 sources·updated 7 Sep 2026
Massachusetts Senate AI lab audit bill
The gist
Anthropic Backs, OpenAI Opposes Massachusetts Frontier AI Audit Bill
Two of the largest AI labs are publicly lobbying against each other on state-level frontier model regulation, with Anthropic supporting stronger independent audits and OpenAI seeking to limit them. The outcome could shape how other states approach AI oversight requirements.
The full picture
Anthropic and OpenAI are lobbying on opposite sides of a Massachusetts bill that would require leading AI labs to submit frontier models to independent third-party reviews for catastrophic risks. The Massachusetts Senate passed the bill as part of an economic development package in July 2026. Anthropic has lobbied in favor and is pushing lawmakers to strengthen the measures further, while OpenAI seeks to narrow the bill to align with existing AI laws in California, New York, and Illinois. Anthropic also endorsed a separate Massachusetts House bill that would require independent safety evaluations and grant the state attorney general more powers than the Illinois law provides; even if third-party audits were added to the Senate proposal, it would remain weaker than that House bill. The bill has been described as the strongest frontier model legislation in the country. Rep. Casar called both companies' responses to his congressional oversight letters 'insufficient.'
How it developed
Anthropic and OpenAI have taken opposing positions on a Massachusetts economic development bill passed in July 2026 that would require AI labs to submit frontier models to independent third-party reviews for catastrophic risks.
OpenAI wants the proposal aligned with existing AI laws in California, New York, and Illinois, while Anthropic pushes for stronger measures including independent safety evaluations and expanded attorney general powers. Rep. Casar called both companies' responses to his oversight letters 'insufficient'.
Averi launches evaluation methodology where the eval company does not interact with model weights and the model creator receives no information about the evaluation.
Dean Ball argues the false neuralese viral claim illustrates the need for legislated independent auditing of frontier AI labs.
Sources
bostonglobe.com
The daily email
Want this in your inbox?
I send one email each morning with the stories that moved. If you would rather just read here, that works too.
Subscribe free