The Connecticut Senate passed a scaled back artificial intelligence transparency and accountability bill Wednesday that was reworked in an attempt to win Republican votes and, ultimately, the support of the House and Gov. Ned Lamont (D).
In the hours before the late night vote, Sen. James Maroney (D), the prime sponsor, removed two key sections of the bill dealing with algorithmic discrimination. Maroney had announced earlier in the day that those sections were rewritten to incentivize, rather than mandate, AI developers and deployers to take proactive steps to ensure their systems don’t discriminate.
With those provisions gone, the bill now focuses on requiring disclosure to consumers when AI is being used to make a consequential decision about them. It also makes it a crime to disseminate unauthorized AI-generated intimate images and includes provisions focused on AI innovation and job training.
“I’m disappointed I didn’t do the work necessary to build a bigger coalition to get the important parts of the bill passed,” Maroney told Pluribus News on Thursday.
He said he made the judgment that it was most important for the bill to be passed with a strong bipartisan vote.
Maroney is hoping to avoid a redux of what happened last year when his first attempt at passing an AI algorithmic discrimination bill died in the House after a veto threat from Lamont.
“The stronger support coming out of the Senate hopefully gives it more legs in the House,” he said.
Maroney’s 11th-hour pivot is the latest indication of an increasingly fraught AI regulatory landscape in the states, as pushback to regulations gains steam. Republicans in the U.S. House are now considering a 10-year moratorium on state AI regulations.
But it is still not clear the changes made to the bill are enough to win over Lamont, a skeptic of AI regulations. Earlier Wednesday, before the latest amendments, Lamont told reporters he was “really cautious about the bill.”
“I just worry about every state going out and doing their own thing; a patchwork quilt of regulations,” Lamont said.
On Thursday, Lamont signaled more openness to the bill. In video posted to X by John Craven of News 12 Connecticut, the governor praised Maroney’s efforts to build broader support for the legislation saying, “It’s constructive, so we’re going to take a look at it.”
Maroney, a national leader in AI policy, initially set out to pass a law to prevent discrimination when AI systems are deployed to make high-stakes decisions about people’s lives in areas such as employment, housing and lending. Maroney’s 2024 bill became a model for other states, including Colorado, which passed the nation’s first sweeping AI algorithmic discrimination law last year.
But along the way Maroney has faced a significant roadblock: Lamont’s skepticism. The governor has argued that existing anti-discrimination laws already apply to AI and has expressed concern that regulations will stifle innovation.
Maroney’s original bill included a mandate that AI developers and deployers conduct impact assessments and take other steps to avoid algorithmic discrimination. On Wednesday morning, he announced a revised version that would shield AI companies from punitive damages if they adhered to national or internationally recognized responsible governance frameworks. But even that proved politically fraught and Maroney ultimately determined to remove those sections altogether. He said he would try again next year.
“We are at risk of perpetuating and amplifying biases and to not act early is really abdicating our responsibility as leaders,” Maroney said in the interview. “I pledge to continue that work but unfortunately I wasn’t able to get to that this year.”
Nonetheless, Maroney said the bill would still protect Connecticut consumers with its focus on transparency, while also promoting innovation and empowering residents to acquire the skills they need to compete as AI becomes more ubiquitous. It would also create an AI Safety Institute in Connecticut to promote “responsible, ethical and trustworthy AI.”
“This bill positions us as a national leader in responsible innovation — ensuring that AI works for everyone and that no one is left behind in our rapidly evolving economy,” Maroney said in a statement after the vote.
Daniel O’Keefe, Lamont’s chief innovation officer, previously warned that the Senate’s efforts to regulate AI could have a “chilling effect” and suppress innovation.
Lamont introduced an alternative bill that would create an investment fund and a “regulatory sandbox program” to encourage AI innovation in Connecticut. It would make it explicit that existing anti-discrimination laws apply to AI. Maroney’s bill also includes a sandbox program.
Senate Majority Leader Bob Duff (D) said Wednesday it is even more important that states erect AI guardrails since President Trump repealed Biden-era AI safety measures. He said “it’s up to us to make sure that we’re providing those parameters and regulations.”
Maroney said 8 in 10 companies now use AI to help make employment decisions, and that more than half of large landlords use AI to screen tenants. He pointed to a recent study by University of Washington researchers that found AI resumé screening tools showed a strong bias in favor of white-sounding names and male names. And he highlighted Pew research that found most Americans are more wary than excited about AI.
Citing lawmakers’ slow response to the advent of social media, Maroney said that as AI rapidly takes hold, “it would be legislative malpractice to not do something.”
This story has been updated to add additional comments from Connecticut’s governor.