Connecticut’s AI Act Takes Effect

Connecticut has joined states on the forefront of AI regulation with the passage of its Connecticut Artificial Intelligence Responsibility and Transparency Act (the “CART Act”) earlier in 2026, with the law coming online earlier this month on October 1. The CART Act is distinct from other states with laws governing AI with its breadth and comprehensive nature. Artificial Intelligence is defined specifically, but broadly, to capture “any machine-based system that, for any explicit or implicit objective, infers from the inputs such system receives how to generate outputs, including, but not limited to, content, decisions, predictions or recommendations, that can influence physical or virtual environments.” The CART Act regulates Artificial Intelligence in a number of ways, including the following:

Frontier developers, those that build, train, and deploy “foundational models” with specified levels of computing power, must refrain from enacting policies that discipline or discharge covered employees for reporting concerns about the frontier developer engaging in activity that poses a specific and substantial danger to public health or safety due to “catastrophic risk.” Further, by October 1, 2027, among other structural requirements, these frontier developers must establish processes for covered employees to anonymously report these catastrophic risk concerns.
Employers that utilize automated employment-related decision technologies must, by October 1, 2027, disclose to applicants the use of such technologies (unless such use is obvious) and prior to using these technologies as a substantial factor in an employment decision provide enumerated notices to the employee or applicant. Automated employment-related decision technologies are those that process personal data and use computation to generate any output, including, but not limited to, any prediction, recommendation, classification, ranking, score, or other information, that is a substantial factor used to make or materially influence an employment-related decision. Specifically carved out from this definition are traditional technologies, like word processing, traditional databases, and internet activities.
Effective January 1, 2027, operators of artificial intelligence companions, those artificial intelligence technologies that have a natural language interface and provide human-like responses via anthropomorphic features and can sustain relationships across multiple interactions, must include protocols that detect risks of suicide, self-harm, or physical violence and take enumerated measures when these risks are detected. Additionally, for users of artificial intelligence companions that are under 18, operators must meet or exceed industry standards for preventing the companion from encouraging certain dangerous behavior. It is important to note that there are numerous exclusions from technologies that constitute artificial intelligence companions, where the definition is designed to capture technologies that are specifically designed for such relationships.

While the law is in effect presently, many of its provisions have delayed effective dates to allow providers some time to formulate their compliance postures. That said, many of these compliance obligations will require holistic shifts in technology or internal business operations, so tackling compliance should not be delayed. A copy of the law can be found here: https://www.cga.ct.gov/2026/act/pa/pdf/2026PA-00015-R00SB-00005-PA.pdf