AI bill targeting chatbot risks, child safety passes CT state Senate | #childsafety | #kids | #chldern | #parents | #schoolsafey

HARTFORD — A bill regulating these early stages of artificial intelligence, including protections for children and users of chatbots, passed the state Senate Tuesday in a bipartisan vote that sends it to the state House of Representatives.  The bill also aims to encourage workforce development and require transparency for the AI industry.

“We’re looking at becoming the most AI-literate workforce in the nation,” said Sen. James Maroney, D-Milford at the beginning of the four-and-a-half-hour debate. “We make sure we’re empowering our state government to utilize AI in a responsible way to better serve our citizens.” The legislation passed 32-4, with opposition from Republicans Sens. Rob Sampson of Wolcott, Tony Hwang of Fairfield, Eric Berthel of Watertown and Ryan Fazio of Greenwich.

Gov. Ned Lamont, an obstacle to the legislation in recent years, said he favors this year’s bill, which includes some legislation he proposed in February.

In a mid-afternoon statement, before the legislation reached the Senate, Cathryn Vaulman, Lamont’s communications director, said the governor’s office worked closely with Maroney on the bill.

“The governor appreciates that this bill is focused on empowering Connecticut residents to use AI responsibly,” she said. “It boils down to: AI is an important part of our – and everyone’s – economic growth story. This bill facilitates that, while making sure that in specific-use cases, folks have safety and clarity in how to work with these systems.”

In recent years, Maroney advocated wide-ranging legislation that Lamont warned could stifle the nascent industry of artificial intelligence. In 2024, the bill passed the Senate and died in the House after Lamont said he would veto it. A similar event occurred last year. This legislative session, Maroney said he was in regular communication with the governor’s office.

It was a years-in-the-making victory for Maroney, the co-chairman of the General Law Committee, who has become a nationally recognized expert on the still-emerging AI industry, which Sen. Paul Cicarella, R-North Haven, said can be “scary.” There are also expanded whistleblower protections for people working in the industry who report risks described as “catastrophic” that could threaten lives or millions of dollars.

“I believe that that’s important, that we don’t want to deter people within these companies from coming forward to stop something that could be so catastrophic,” Cicarella said, warning that AI’s potential for damage might be unknown now. “Honestly, if a company was going to hurt our residents because they could make a lot of money, I think that this would be one of those times when it was appropriate.”

Maroney, and Cicarella, a ranking Republican on the General Law Committee, agreed that the new, fast-moving AI industry already adheres to many of the proposed industry guidelines.

A so-called sandbox provision, would give some regulatory flexibility for new use of AI. The bill would order various state departments, including the Department of Economic and Community Development, to oversee the testing of various AI models. “We are giving them regulatory certainty and relief,” Maroney said. “The regulatory sandbox is telling them what they can and can’t do. We are looking at talking to other states to build sandboxes.”

Cicarella said that since there are no national rules, states may have to take the issues head-on without deterring businesses from moving here. “We haven’t seen enough movement,” Cicarella said.

In cases of companion chatbots, if a user starts displaying suicidal ideations, the unit would have to shut down. Other protections would shield children from sexual content and self harm.

Some of the chatbot legislation was, in part, inspired by a CT Insider investigation last year that showed shortcomings when it came to creating protections for children with sexually explicit AI chatbots.

In cases where companies with chatbots violate the industry-best standards, the state attorney general would be authorized to file unfair business practice claims. “We are not by any means, going to be the first state to have this requirement,” said Maroney, adding that many of the children’s safeguards have already been adopted in several states around the country.

“This is an example of where ‘move fast and break things,’ which is the big tech mantra, has had real life implications on real people,” Maroney said. “Unfortunately, we have seen in other states where children have died by suicide, often having been encouraged or at least validated for their delusions by their companion chatbot.” He pointed out a recent murder-suicide in Greenwich that is alleged to have been encouraged by AI. 

“When they’re talking about the issues they may have, some of the models may validate those concerns,” Cicarella said. “Some of the models may even transition, I would hope accidentally, to how-to kill yourself. If we could do something here to save one person I think that that’s truly important.”

Maroney said that evidence-based methods would require makers of chatbots to detect expressions of suicide, self-harm or imminent physical violence and take measures to generate any output to encourage such expressions. “The key is in using industry-best standards or better,” Maroney said. “We all know that a good friend is the friend who tells you what you need to hear not what you want to hear.” 

Another section of the 71-page bill would require industry disclosure, including the notification of job seekers to be informed if AI would be involved in their employment screening, including the types of information asked of job candidates. Using current law, AI would be banned from discriminating on a variety of levels, including race, gender and age.

State companies using the technology would not be liable if the tech was flawed and was prejudiced against certain candidates. “There are laws on the books being used for the enforcement,” Maroney said. So-called deep fakes would have to contain meta data revealing the authenticity of an image.

The free Connecticut AI Academy would also be expanded to children if the bill were to become law on October 1. Maroney warned that a recent study by Tufts University found that 105,000 Connecticut jobs could be lost due to the growth of artificial intelligence over the next several years.

“It’s amazing what AI can do and how it can streamline projects,” Cicarella said. “This is really a good piece of legislation that talks about workforce development. This will tell you how to learn this and get a job.” He noted that there were some concerns on the use of AI in health care, to which Maroney replied that privacy would be assured in compliance with state and federal law. “The goal behind this is to incentive this for improving health outcomes,” Maroney said, hoping that it would encourage companies to move to Connecticut. “There would be protections to make sure no one’s personal health information would be leaked.”

A whole section on child protections would address addictive social media feeds. Notifications to minors would be curtailed overnight. Cicarella said that in previous years, he was concerned about negative ramifications for the industry. This year’s bill, he said, seems to have clarified the issue. “I believe it will do a lot of good,” he concluded.

“It’s government over reach, plain and simple,” said Sen. Rob Samson, R-Wolcott, a conservative who voted against the bill. “This is a massive takeover. There are not-clear legal standards anywhere in this bill. This is just a cobbled together mess of policies that the majority finally came up with to bring us today.” 

“We need to catch up to technology,” said Senate Majority Leader Bob Duff, D-Norwalk, at the end of the debate. “It would be a grave mistake for us to wait any longer for this bill.”