Massachusetts may be on the verge of passing one of the country’s most stringent AI safety laws, putting the state on a collision course with President Donald Trump’s hands-off approach to rapidly evolving technology. Trump, who has embraced industry self-regulation, recently declared on social media that “the only control or ‘guardrails’ that AI needs is a STRONG AND SMART (High IQ!) PRESIDENT.”
Gov. Maura Healey and some state legislators think otherwise.
“The companies that are building out these AI technologies have a responsibility to build them safely, and government has a responsibility to set reasonable guardrails,” Healey said in an interview with MassLive last month.
She acknowledged that 50 states putting in place different regulations “isn’t the most optimal thing,” but said she didn’t expect action at the federal level under the current administration. Healey has asked state legislators to add language on AI safety and transparency to the Mass Wins economic development bill, now being negotiated on Beacon Hill.
So what exactly is happening?
States such as California, Illinois and New York have already put in place laws that apply to the biggest AI developers — companies like OpenAI, Anthropic, Meta and Microsoft — that are building what are sometimes called “frontier models.” The concern is that as AI technology advances, it could either help a bad actor execute cyberattacks or build chemical, nuclear or biological weapons — or carry out nefarious plans on its own.
California, Illinois and New York laws require major AI developers to publish their own plan for managing catastrophic AI risks, report serious safety incidents to the state and protect employees who blow the whistle on non-compliance or other serious concerns. Illinois was the first to require an annual third-party audit to verify whether AI developers were following their own safety protocols.
However, none of the laws allow outsiders to look under the hood to see how the technology works — and whether the potential for truly catastrophic events is increasing. And yes, the law defines catastrophic: the death or injury of 50 or more people, or causing $1 billion or more in damage.
This deeper look into how the technologies are evolving is what could make Massachusetts different: the law would give independent evaluators the job of testing the core software (or models) for catastrophic risk potential at least three times a year.
Tech companies would cover the costs of being evaluated, and the hope is that these evaluation firms — an industry still in its infancy — would be like corporate financial auditors, not overly dependent on a single client. While there are thousands of large companies that must produce audited financials each year, there may not be nearly that number of big AI developers — companies generating more than $500 million in annual revenues and using massive computing power to train their AI systems.
But the first-in-the-nation concept of independent evaluators examining AI models isn’t a done deal. As of Friday, state Sen. Barry Finegold, Second Essex and Middlesex District, who chairs the economic development committee, said the issue was “still being discussed.”
Among the companies actively lobbying the Legislature are Google, Anthropic, OpenAI and Meta, the parent company of Facebook, Instagram and WhatsApp, Finegold said.
Anthropic, the maker of Claude, and OpenAI, the developer of ChatGPT, both support independent evaluations of their AI tech.
“We’re endorsing this bill because it would give Massachusetts the strongest AI safeguards in the country — and that’s something we need more urgently with each passing day, as these systems grow more capable,” Cesar Fernandez, head of U.S. State and Local Government Relations at Anthropic, said in a statement. “Real credit goes to the Joint Committee on Economic Development and Emerging Technologies, especially Co-Chairs Finegold and (Rep. Carole) Fiola, (6th Bristol District), who understand that we can’t wait to take AI risks seriously.”
Last month, Anthropic CEO Dario Amodei published an essay suggesting the industry should slow the pace of AI development to control the risks of AI going rogue and doing bad things. Anthropic also recently announced its own plan to bring safety evaluators into the company, in partnership with the consulting firm Accenture.
OpenAI also supports independent evaluators.
“As AI systems become more capable, the safeguards and oversight around them need to keep pace,” an OpenAI spokesman wrote via email. “People deserve confidence that companies’ safety commitments stand up to scrutiny. We support independent evaluations, including embedded evaluators, and want to help make them effective — with qualified experts, meaningful access and strong security protections for the sensitive systems they assess.”
An embedded evaluator would be an outside evaluator who could work inside the AI company and be given access similar to an employee’s, although the Massachusetts bill doesn’t use that term.
Meta, Google and Amazon — a major player in AI, with thousands of Massachusetts employees — didn’t respond to requests for comment.
It appears few people in the Massachusetts AI scene have been lobbying elected officials on this issue. Paul Baier, Paul English, Kara Peterson and Judah Phillips, who have been active on things like creating Massachusetts AI Week or bringing AI training into Boston Public Schools, all told me they haven’t been involved.
But they do have concerns.
English, an entrepreneur who funded the Paul English Applied AI Institute at UMass Boston, said via email “There is always the ‘if we legislate and China does not’ argument that I don’t know what to do about.”
Developers of AI tech in the U.S. would likely fall under the jurisdiction of the proposed Massachusetts legislation, but the state would have little ability to enforce the rules against Chinese companies without a U.S. presence. And bad actors could download publicly released Chinese AI models — so-called “open-weight” models — and modify them, beyond the reach of any federal or state regulations.
Paul Baier, CEO of the Boston research firm GAI Insights, said allowing U.S. companies to compete effectively with the Chinese was also a worry for him. But he also pointed out that the state laws — and the Massachusetts bill — speak loudly but carry a small stick. The penalties for flouting the proposed Massachusetts rules would be up to $1 million for a first violation and up to $3 million for subsequent violations.
“The best thing the regulations would have is high financial penalties and criminal charges for AI companies (like) OpenAI, Anthropic and Meta releasing products that are not tested enough, (and) allow users to hack into other companies,” or promote suicidal behavior, Baier wrote via email.
“Anthropic and OpenAI want minor penalties,” he wrote. “The penalty should be one to five percent of the market capitalization of a company.”
What specifics will end up in the Massachusetts bill that becomes law? We should know within the next few weeks.
But if Massachusetts does get the country’s toughest AI safety law, it could set up a showdown with the Trump administration. In December 2025, Trump signed an executive order discouraging “excessive state regulation” of AI and directed the Department of Justice to create an AI Litigation Task Force to sue states over laws the administration regards as “onerous.”
©2026 Advance Local Media LLC. Visit masslive.com. Distributed by Tribune Content Agency, LLC.