BOSTON — The last few weeks have been “a tipping point” in the world of artificial intelligence, U.S. Rep. Lori Trahan said Tuesday as warnings mount in Massachusetts and lawmakers juggle various viewpoints in the midst of regulatory negotiations.
“We should be in Washington right now. We should not be on a seven-week hiatus reconvening after the election,” Trahan said of Congress. “The imminent risks that have been posed by artificial intelligence just these last few weeks are cause for, you know, a national conversation around guardrails.”
Following recent AI security incidents, “the conversation has just completely changed in Washington. It is on us to keep that momentum up during these seven weeks while we’re home, but then also post-election, using November and December to pass these safety guardrails,” she said, sitting next to Sen. Barry Finegold in his State House office.
Trahan said she’s “hopeful” a bill she filed with California Republican Rep. Jay Obernolte, known as the FRONTIER Act, will move after midterms. The bill would create a national, risk-based framework that mandates catastrophic risk transparency, incident reporting and independent evaluations, and has a “federal kill switch” so the government can step in if there’s a model that poses an imminent catastrophic risk.
Gov. Maura Healey last week called for federal oversight and protections “before it’s too late,” and requested that legislative economic development negotiators require independent safety evaluations of frontier AI models and earlier disclosures of critical safety incidents.
“Ideally, the federal government would be providing a central standard, because all the different technology companies ran across 50 different standards is a problem,” Technology Services and Security Secretary Jason Snyder said Tuesday at the Massachusetts Digital Government Summit.
States need overarching federal regulation instead of a smattering of state-level laws, Finegold said. The Andover Democrat is one of six economic development conferees reconciling House and Senate proposals, the latter of which (S 3178) includes AI safety regulations. Senators in July adopted a Sen. Mike Rush amendment that would require large AI developers to comply with independent, third-party safety audits and undergo independent safety testing every four months.
Anthropic has publicly backed the Senate proposal. Tech industry coalition Chamber of Progress on Sept. 10 sent a letter to conferees urging them to oppose the Senate proposal “unless amended.”
The letter calls on lawmakers to remove the requirement from its frontier AI provisions that developers undergo independent model evaluations and align more closely with an Illinois AI law (SB 315). The letter also says that the Senate bill as drafted “would require covered developers to submit their frontier models to independent evaluations for catastrophic capabilities and risks at least once every 120 days. Yet the bill imposes this mandate without an established accreditation system for determining who is qualified to perform these evaluations, a broadly accepted testing methodology, or objective benchmarks for determining whether a model has passed.” It also takes issue with the 120-day schedule, as “frontier models and their safeguards are continuously updated.”
“Massachusetts can advance meaningful frontier AI oversight without imposing requirements that lack the standards, capacity, and infrastructure necessary for reliable implementation,” the chamber’s director of state and local government relations for the Northeast, Jabari Cooper, wrote.
Finegold said Tuesday that he’s “concerned about the outside lobbying that’s going on” around state regulation of AI, calling it “pretty intense.”
Leaders of Anthropic, OpenAI and x.AI agreed recently that the advancement of frontier AI models should be slowed down for risk prevention purposes. OpenAI said in July that its AI system autonomously breached its testing environment and hacked into another AI company, startup Hugging Face, according to the Associated Press. An Anthropic researcher announced in early September he was resigning from the company due to concerns about company responsibility, per PBS.
“Today we’re literally working on the honor system,” Trahan said. “The only reason why we know about the Hugging Face incident is because OpenAI chose to tell us.”
Asked about the juxtaposition of state investment in AI and warnings about the technology, Finegold said, “I think we can walk and chew gum.”
“We want to have an AI economy that can do amazing things like cure cancer and things like that. But at the same time, we want to make sure we understand what’s going on there,” he added.
Speaking about the United Nations General Assembly’s meeting this week, Trahan said she wished “we had done the hard work going into those meetings of regulating AI here at home, because I think it dramatically changes our leverage in these conversations.” U.S. House Speaker Mike Johnson has argued that China would have an advantage over the U.S. if AI development is paused.
Finegold weighed in on a different realm of the AI sector Tuesday, noting that he and others “plan to file legislation to make sure that media companies are adequately compensated for the work they do” if a lawsuit between OpenAI and the New York Times isn’t settled. The New York Times filed a federal lawsuit against OpenAI and Microsoft to end their practice of using published stories to train chatbots, per the Associated Press.
“I think it’s a great thing that AI companies learn from all the journalism all of you do, but there has to be a price for it,” he told reporters in his office.
Ella Adams is a reporter for State House News Service and State Affairs Pro Massachusetts. Reach her at [email protected].
Original article
Comments