Connecticut’s AI Bill Sheds Major Mandates
Key highlights this week:
We’re tracking 1,016 bills in all 50 states related to AI during the 2025 legislative session.
The congressional reconciliation budget bill, containing a 10-year moratorium on enforcement of state or local AI laws, passed the U.S. House, sending it to the Senate.
Nebraska enacted a sexual deepfake law.
And Sen. Maroney hopes that a stripped-down version of his comprehensive AI bill will gain gubernatorial approval this year in Connecticut, which is the subject of this week’s deep dive.
The Connecticut Senate has passed a stripped-down version of Sen. James Maroney’s (D) comprehensive AI bill (CT SB 2). This is the second attempt at a comprehensive AI bill pushed by Sen. Maroney after his bill last year failed to gain the support of Gov. Ned Lamont (D) after also gaining approval in the Senate. The amendments this week further reflect how momentum has shifted against the regulation of AI nationwide, and how even advocates are acknowledging a new political reality.
Last Wednesday, the Senate approved Sen. Maroney’s SB 2, but the version bears little resemblance to the ambitious legislation originally introduced. What started out as an algorithmic discrimination regulation on developers, integrators, and deployers of AI systems is now a transparency bill limited to deployers with a grab back of programs added to accelerate the adoption of AI in Connecticut.
Many Mandates Removed
The amended version eliminates all obligations on “developers” and “integrators” of AI systems found in the original bill, which included disclosures, documentation, impact assessments, and a “reasonable duty of care” to prevent algorithmic discrimination. Industry groups had argued that these mandates were too vague and posed a threat to startups.
Many of the obligations that the original bill placed on “deployers” of AI systems have also been rolled back under the amended bill, however, deployers will still be held to transparency requirements when making consequential decisions affecting consumers.
Under the amended bill, deployers would not be held to the “reasonable duty of care” to prevent algorithmic discrimination found in the original bill. Additionally, the amended version of the bill removed requirements that deployers draft and file an impact assessment and other public disclosure requirements on the deployer's website regarding the high-risk AI systems deployed.
Lawmakers removed provisions from the original bill that would have required generative AI models to mark synthetic content, although the amended bill still makes it unlawful to disseminate a sexual deepfake image. Connecticut is one of the few states that has yet to pass either political or sexual deepfake legislation.
Disclosure Requirements Remain
However, the bill retains certain requirements for deployers when relating to consumers. Under the amended bill, deployers must still notify consumers when they are interacting with an AI system. They are also required to provide disclosures when a “high-risk” AI system plays a substantial role in making a “consequential decision” that is adverse to the consumer and provide explanations for adverse decisions, including appeal opportunities with human review. The amended version does clarify that the notice shall be given before a consequential decision is made. The amended bill also exempts health care providers and federal contractors from the deployer transparency requirements.
Acceleration of AI Adoption
The amended version also includes several provisions to develop the AI industry in the state, such as:
A public education and outreach program for small business developers and deployers;
Establishment of an AI safety institute;
Establishment of a Connecticut AI Academy to offer online courses;
Authorizing expenditures for workforce development programs; and
Repurposes technology talent pipeline programs to include AI and quantum computing.
The bill retains provisions creating an AI regulatory sandbox similar to the program implemented in Utah. The regulatory sandbox provision had also been part of an alternate AI bill pushed by Gov. Lamont (CT SB 1249). However, some economic development sections were scrapped, such as sections creating a confidential computing cluster to foster the exchange of health information to support academic and medical research, the creation of a Connecticut Technology Advisory Board, and a proposed AI systems fellowship program.
Public Sector AI Use
Finally, the amended bill would add measures to integrate AI use in the operation of state government. The additions would create a pilot program to determine whether generative AI can be used to improve agency processes, direct policies and procedures to train state employees on AI, and create an AI working group.
Will the Governor Sign It?
Sen. Maroney had offered to rewrite the obligations on developers and deployers to instead incentivize measures to prevent algorithmic discrimination, but the measure ultimately passed without those provisions. After years of work on task forces and multistate working groups on AI, the AI regulation advocate lamented that his efforts for more regulation had fallen short. “I’m disappointed I didn’t do the work necessary to build a bigger coalition to get the important parts of the bill passed,” Sen. Maroney told Pluribus News.
It remains unclear whether the proposed changes will be enough to win the support of Gov. Lamont (D). Last week, the governor echoed his concerns about new AI regulations making the state an outlier, potentially joining Colorado as the only states to require broad obligations on deployers. “I just worry about every state going out, doing their own thing, a patchwork quilt of regulations. Connecticut being probably stricter and broader than most,” Gov. Lamont said before debate of the bill last week.
Federal Preemption Looms
Looming over this legislation is the provision in the U.S. House budget reconciliation bill that would impose a ten-year moratorium on state regulation of AI. This week, the U.S. House passed the reconciliation bill with minor changes to the preemption moratorium. If enacted by Congress, the federal preemption would appear to block enforcement of even the pared-back provisions in Connecticut’s amended bill requiring disclosures. But many of the bill’s provisions to accelerate the adoption of AI in Connecticut would likely survive preemption.
While the amended bill preserves some consumer-facing transparency requirements and launches a suite of economic development and workforce training initiatives, its retreat from regulatory guardrails marks a significant concession to industry concerns. As the bill heads to the House, Connecticut’s approach may serve as a bellwether for how states navigate the tension between innovation and regulation.
Recent Developments
Major Policy Action
Congress: The U.S. House passed the budget reconciliation bill (US HR 1) that includes a ten-year moratorium on state regulation of artificial intelligence, approving minor changes to that provision that would clarify application to regulation of AI models in interstate commerce, and add an exemption for laws with criminal provisions. The reconciliation budget now moves to the U.S. Senate, where the AI preemption moratorium must survive the Byrd Rule before likely returning to the House for a final vote.
Connecticut: The Senate amended a health insurance bill related to mental health parity (CT SB 10) to strip out provisions that would have regulated the use of artificial intelligence in utilization review. The Senate passed the amended version, sending it to the House.
Nebraska: Gov. Jim Pillen (R) signed a bill (NE LB 383) into law to amend child pornography laws to include computer-generated child pornography. The law creates the Parental Rights in Social Media Act, but incorporates provisions in NE LB 172 relating to sexual deepfakes.
Illinois: The Senate approved a measure (IL HB 1859) that would prohibit community colleges from using AI for instruction in lieu of a human instructor. The bill returns to the House to concur on a Senate amendment.
Nevada: The Senate approved a bill (NV AB 406) that prohibits AI from practicing mental or behavioral health care or making such representations, and prohibits schools from using AI for mental health counseling. A Senate committee also amended and advanced a political deepfake bill (NV AB 73). The amendment clarifies that the measure applies to communications that provide information about political or social issues with the intent to influence the outcome of an election.
New York: Lawmakers held a press conference Tuesday to push for approval of AI-related legislation before adjourning on June 12. More than 50 state lawmakers signed a letter opposing the proposed ten-year AI moratorium in the House budget reconciliation bill.
Texas: The Senate Business and Commerce committee amended and passed an AI comprehensive regulation bill (TX HB 149) championed by Rep. Capriglione (R). The amendment modifies the definition of “artificial intelligence,” exempts certain biometric data used in AI development from biometric regulation, clarifies disclosure for AI use in health care, and clarifies that financial institutions are in compliance with anti-discrimination provisions if they are already in compliance with banking laws.
Wisconsin: The Senate passed a bill (WI SB 33) to make it unlawful to publish or distribute a sexual deepfake of an identifiable person with the intent to coerce, harass, or intimidate that person.
Notable Proposals
New York: Assemblymember Charles D. Lavine (D) introduced a bill (NY A 8546) requiring any legal document or filing that was drafted using generative AI to include a separate affidavit that discloses the use of AI and certifies that a human has reviewed the source material and verified the accuracy of the content. The bill is a companion to NY S 2698.
Pennsylvania: Rep. Robert E. Merski (D) filed a resolution (PA HR 170) to establish an advisory committee to conduct a study on the field of artificial intelligence and its impact and potential future impact
