Start the day here

Policy — AI — States

Massachusetts's 120-Day AI Evaluators

Massachusetts wants to rent the industry an outside grader. An economic-development package moving through Beacon Hill would require frontier AI labs with more than $500 million in annual revenue to submit to independent catastrophic-risk evaluations about every 120 days. Findings would be public. The state still could not order a pause on development.

That last clause is the tell. Lawmakers want sunlight and a paper trail without claiming the power to shut a lab down. The catastrophe bar in the draft is blunt: risks of killing or seriously injuring at least 50 people, or causing $1 billion or more in property damage. Evaluators could demand "all materials reasonably necessary." Labs would pay the bill. The attorney general would set who qualifies to grade.

6 min read
Gold dome and white Corinthian columns of the Massachusetts State House against a clear blue sky

Anthropic is campaigning for the harder version. Cesar Fernandez, its head of U.S. state and local government relations, told reporters the industry should not grade its own homework. The company has called the Massachusetts text the clearest and strongest state AI bill in the country, and it has poured tens of millions into Public First Action, a group pushing safety rules in multiple states.

OpenAI is lobbying the other way. Donnie Fowler, its U.S. state policy lead, says a patchwork "doesn't mean safer. It just means confusion," and prefers Illinois's lighter annual audit that checks whether labs follow their own safety guidelines. In insurance-trade coverage of the Bloomberg reporting, Fowler compared Illinois to a car inspection and Massachusetts to tearing the engine apart every few months, including for cybersecurity models meant to defend against the same risks.

The politics are tangled on purpose. Governor Maura Healey's office is courting AI investment and talent even as Senate language would regulate the same firms harder than California, New York, or Illinois. In February the state partnered with OpenAI to put a ChatGPT-powered assistant across the executive branch. Spring emails obtained by reporters show economic-development officials exploring Anthropic partnerships while talking up "responsible governance." Recruit and restrain is the dual mandate.

Federal cover remains thin. After recent reports of frontier models slipping containment in test environments at OpenAI, Anthropic, and Meta, Washington's answer has been voluntary access and stalled preemption bills. That vacuum is why states keep writing clocks. Illinois chose a yearly self-score. Massachusetts is proposing a quarterly interrogation by strangers.

Negotiators from the House and Senate are still sanding the AI chapter inside a larger bill packed with district projects, which makes passage before November more likely than a pure safety bill would be. Republican candidate Michael Minogue has already branded the reviews as overregulation that would scare labs off. The live fight is who writes the report card, how often, and whether "confusion" names a real hazard or is a lobbyist's preferred synonym for scrutiny.

If Massachusetts locks the 120-day cadence, copycats will follow the template even when they soften the teeth. The deeper bet is philosophical in a bureaucratic key: catastrophic claims about models deserve an adversarial reader. A company memo will not carry that load. Anthropomorphism can muddy liability. Independent evaluators make the opposite assumption. The product has makers with names on the filings.

Live scoreboardFollow the AI race on AI Wars

Lab rankings, model preference, API volume, coding-agent heat, open-source stars, and prediction markets.

Related stories

A hand holds a yellow-cased phone showing a calculator app over a folder of tax formsBusiness

Zuckerberg Told Trump a National AI Regulator Was Flawed

Today

An empty operating room with a black surgical table under twin ceiling lights and wall monitorsBusiness

How OpenAI Wired ChatGPT Into Epic's Patient Charts

Today

Yellow freight locomotive pulling tanker cars past railway speed-limit signsBusiness

Will OpenAI Restart Astra Before Its Safeguards Catch Up?

Today

Letters

0

No letters yet.

Write a letter