The Massachusetts State Senate has passed a bill that would (among many other things) require large developers of frontier AI models to submit their models to third-party safety testing at least every 120 days, going further than previously passed state laws on AI safety. This provision is contained in the "Transparency in Frontier Artificial Intelligence Act"[1] as part of a large omnibus bill titled "An Act relative to economic development in the commonwealth."
Whether this provision will survive in the final bill is unclear. The Massachusetts House previously rejected an amendment that would have added a similar provision to its version of the bill, but that was back in July when the Overton window for AI policy was very different from today's. The House and the Senate have appointed a six-person "committee of conference" to hammer out the differences between the two versions. The bill remains eligible for passage at any time until year's end.
If you live in Massachusetts and think such a policy would reduce catastrophic risks from AI, contact your state legislators and let them know you support this. Your voice might be particularly important if you're represented by one of the six legislators on the committee of conference:
And if you have more time, consider asking for a meeting—you might just get one!
The testing requirement would only apply to frontier developers that are both (a) making more than $500M / year in gross revenue (together with affiliates) and (b) using over 10^26 FLOP to train a model, so it shouldn't harm startups. If you'd like to read the statutory language yourself to see if you support it, the "Transparency in Frontier Artificial Intelligence Act" passed by the Senate is at pages 148–169 of the PDF available for download here, as Chapter 93M. The mandatory third-party safety evaluation requirement is on pages 160–162, in Section 3A(b). In brief, the Senate summarized the amendment that added the provision as follows:
Frontier AI Risk Reports and Independent Verification (Amendment 471): Verifies that large AI developers are complying with this legislation’s new safety requirements by requiring independent, third-party safety audits. Additionally requires large AI companies to undergo in-depth, independent safety testing every four months to identify any residual catastrophic risks.
The global AI policy conversation is shifting very rapidly, so if you've been meaning to reach out to your legislators but haven't yet gotten around to it, now is a great time to act.
Confusingly, this is the exact same name as the California AI safety law enacted in 2025 as SB 53, though what Massachusetts is considering would go further.
