Connecticut Puts AI On Notice: Transparency And More

Connecticut is the latest state to join the AI regulation space with the Connecticut Artificial Intelligence Responsibility and Transparency Act (SB 5). Portions of the law will go into effect October 1, 2026, with other provisions phasing in through 2028. In addition to regulating AI tools in the employment space, the law also seeks to regulate other aspects of how companies develop, sell and use AI in Connecticut.

Employment-Related Use of AI Tools

Connecticut joins others, including New York City and Illinois, in regulating artificial intelligence in the employment space. The employment-related aspects of the law apply to those employers who use “automated employment-related decision technology.” This includes using an automated system to generate outputs that materially influence employment decisions like hiring and firing. Also covered are training and employment renewal decisions. And, outputs that include predictions and recommendations. It also includes classifying or ranking individuals.

Beginning October 1, 2027, covered employers will need to take two steps:

1. Make disclosures. Before a business uses an automated tool to make or materially influence an employment decision, it will need to notify the applicant or employee. The notice will need to explain, in clear language, that the technology is being used and its purpose. The disclosure will also need to explain the nature of the decision and the technology's trade name. It will also need to list the categories and sources of personal data to be analyzed and contact information for the deployer. The disclosure should also explain how the individual will be assessed. Additionally, unless it is obvious to someone that they are interacting with AI, the business needs to disclose that the employee or applicant is interacting with a covered AI tool.
2. Avoid bias. SB 5 ties into Connecticut’s discrimination laws. In the event of an allegation of bias involving an AI tool, regulators and courts will look at whether the business tested its tools for unfair outcomes and what it did when it found issues.

Any breach of these automated employment requirements will constitute an unfair or deceptive trade practice enforceable exclusively by the Attorney General, and the statute does not create a private right of action. For violations of this section occurring on or before December 31, 2027, the Attorney General must first issue a cure notice allowing 60 days to correct the issue before initiating litigation.

Broader AI Regulations

Connecticut is stepping in where Colorado has stepped out, notwithstanding White House Executive Orders. SB 5 imposes broad obligations on those who develop or deploy AI tools, some of which mirror those of other states, as noted below:

• AI companions. Like many other states, the law regulates AI companions. Beginning January 1, 2027, any operator of an AI companion (meaning, essentially, an AI that mimics human-like conversations across multiple interactions) will need to build in protocols to detect signs of suicide, self-harm, or imminent violence and refer users to resources such as the 988 National Suicide Prevention Lifeline. Operators will also need clearly and conspicuously tell users that they are communicating with an AI and not a person. This can be done with a notice at the beginning or throughout the interaction. The notice must be displayed once an hour for users under 18 (or once every three hours for adult users) during any continuous session. There are additional restrictions on AI companions used by those under 18. This includes ensuring the companion has measures to prevent self-harm, and more.
• AI content. Starting October 1, 2026, developers of AI systems capable of generating synthetic digital content, such as images, audio, or video, must ensure that outputs are marked and detectable as AI-generated before consumers first encounter them. The labeling solution will need to be technically effective, interoperable, and consistent with nationally or internationally recognized technical standards. Where AI-generated audio, image, or video content forms part of an artistic, creative, satirical, or fictional work, the disclosure will need to be limited to a form that does not hinder the display or enjoyment of that work.
• AI subscriptions. Effective October 1, 2026, mirroring many states’ auto-renewal laws, any business that sells access to an AI product through a paid subscription will need to provide consumers with a written notice disclosing the key terms and conditions, including any limitations it may impose on the consumer's use, before the subscription begins or renews. A failure to do so constitutes an unfair or deceptive trade practice under Connecticut law.
• Safe harbor. The law provides for the creation of safe harbor programs operated by (up to five) independent organizations. A pilot will be launched by July 1, 2027, and companies that adhere to the provisions of those organizations’ programs may be able to use their participation as defensive evidence in private lawsuits brought alleging harm caused by an AI model. Participation will not be admissible in government enforcement actions. The pilot will run through June 30, 2030.

By December 31, 2026, Connecticut will establish a “Connecticut AI Academy” to offer online courses on AI and responsible AI use. The state will also take other steps, including a program to use AI to improve health outcomes and studies of how AI impacts the workforce. It will also look at ways to test innovative AI products under reduced regulatory requirements.

The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.