CT AI Act: New Rules for Responsible AI Use by 2026

As Connecticut finalizes its landmark AI legislation, the Artificial Intelligence Responsibility and Transparency Act is poised to redefine how businesses operate within the state. With the Governor's signature anticipated to formalize its passage, companies leveraging artificial intelligence must begin immediate preparations for a compliance landscape that takes full effect starting October 1, 2026. This comprehensive framework aims to ensure the responsible development and deployment of AI, placing significant new obligations on developers and deployers alike.

What's changing

The Connecticut Artificial Intelligence Responsibility and Transparency Act (the Act) introduces a robust set of requirements designed to foster ethical AI use and protect consumers and employees. At its core, the legislation targets three primary areas: AI companions, synthetic digital content, and automated employment-related decision technology (AEDT).

For AI companions — AI models designed to simulate human conversation — operators face stringent safety protocols. They must implement systems to detect and address user expressions indicating risks such as self-harm, suicide, or imminent violence. Transparency is key, requiring clear notices to users that they are interacting with an AI, not a human. Furthermore, the Act strictly prohibits certain interactions with minors, including encouraging self-harm or engaging in romantic exchanges. These specific safety requirements for AI companions are set to become effective on January 1, 2027, giving operators a critical window to adapt their systems and user interfaces.

Developers of AI systems capable of generating or manipulating synthetic digital content — encompassing everything from AI-produced audio and images to text and video — are now mandated to ensure this content is clearly marked and detectable as artificial. This "watermarking" or clear disclosure must be present when consumers first encounter the content, preventing deception and fostering trust in digital media. This provision, along with employer responsibilities, will become effective on October 1, 2027.

Perhaps one of the most impactful changes for many businesses involves the use of automated employment-related decision technology (AEDT). Employers utilizing AI for hiring, promotion, or other employment decisions must provide written notices to applicants and employees about the use and purpose of these tools. Crucially, the Act emphasizes that these automated tools must not result in discriminatory outcomes. While proactive anti-bias testing can serve as a mitigating factor in enforcement actions, the legislation explicitly states that using automated decision-making technology is not a defense against claims of employment discrimination. Companies remain fully accountable for discriminatory outcomes, regardless of whether a human or an algorithm made the final call. This critical aspect underscores a shift in liability, placing the onus firmly on employers to ensure their AI tools are fair and equitable.

Enforcement of the Act largely falls to the Connecticut Attorney General, who will treat violations as unfair or deceptive trade practices. The Attorney General can offer a 60-day period for businesses to rectify issues before initiating legal action. However, specific violations related to AI companion safety carry severe civil penalties, potentially reaching up to $15,000 per day, highlighting the state's commitment to protecting vulnerable users. The initial provisions of the Act will take effect on October 1, 2026, setting the stage for these phased implementations.

Who is affected

The Connecticut Artificial Intelligence Responsibility and Transparency Act casts a wide net, primarily impacting companies that develop or deploy AI systems within the state of Connecticut. This includes a broad spectrum of businesses, from tech startups innovating in AI to established enterprises integrating AI into their operations.

Specifically, the legislation targets:

AI Developers and Providers: Any entity creating or offering AI models, particularly those designed for conversational interaction (AI companions) or synthetic content generation.
Businesses Deploying AI Companions: Companies that offer AI models simulating human conversation to users in Connecticut. This could range from customer service chatbots with advanced conversational capabilities to therapeutic AI applications.
Content Creators and Platforms Utilizing Synthetic Media: Businesses that generate or manipulate digital content using AI, such as marketing agencies producing AI-generated ads, news outlets using AI for content creation, or social media platforms hosting AI-generated media. The requirement for clear marking impacts anyone whose consumers encounter such content.
Employers Using Automated Employment-Related Decision Technology (AEDT): Any employer in Connecticut that uses AI-powered tools for tasks like resume screening, candidate assessment, performance evaluations, or promotion decisions. This covers a vast array of industries and company sizes, as AI adoption in HR is increasingly common.

While the Act is specific to Connecticut, its principles and requirements reflect a growing national and international trend in AI regulation. Businesses operating across state lines, or those with remote employees in Connecticut, will need to pay close attention to ensure compliance. The focus is on the use and impact of AI within Connecticut, meaning even companies headquartered elsewhere could be affected if their AI systems interact with Connecticut residents or employees.

Three things to do this week

With the Act's effective dates rapidly approaching, proactive preparation is essential. Here are three concrete steps businesses should prioritize immediately:

Audit Automated Employment Decision Technology (AEDT) for Bias: Given the Act's explicit stance against discriminatory outcomes from AEDT, employers must conduct thorough audits of all AI tools used in hiring, promotion, and other employment-related decisions. This involves identifying potential biases in data, algorithms, and decision-making processes. Documenting these audits and implementing corrective measures will be crucial, as proactive anti-bias testing can serve as a mitigating factor in enforcement actions. Ensure your AEDT doesn't inadvertently perpetuate or create discriminatory practices, as the liability for such outcomes rests squarely with the employer.
Review AI Companion Safety Protocols and Transparency: For businesses offering AI companions, immediately assess your current systems for detecting and addressing user expressions indicating risks like self-harm, suicide, or imminent violence. Strengthen these protocols and ensure they are robust and regularly updated. Simultaneously, verify that your AI companions provide clear, unambiguous notices to users that they are interacting with an artificial intelligence, not a human. Review interactions with minors to ensure strict compliance with prohibitions against encouraging self-harm or engaging in romantic exchanges.
Develop a Strategy for Marking Synthetic Digital Content: If your business generates or manipulates synthetic digital content using AI, begin developing a clear strategy for marking this content as artificial. This could involve visible watermarks, metadata, or prominent disclosures at the point of first encounter for consumers. Identify all content streams where AI is used for generation or manipulation and establish consistent, verifiable methods to ensure detectability and transparency. This preparation is vital to avoid potential violations once these provisions become effective.

Related context

Connecticut's Artificial Intelligence Responsibility and Transparency Act is a significant piece of legislation, but it doesn't exist in a vacuum. It joins a growing chorus of state-level efforts to regulate AI, reflecting a broader national conversation. States like Colorado have already passed comprehensive data privacy laws that touch upon automated decision-making, while California continues to explore its own AI regulatory frameworks. Federally, discussions are ongoing regarding a national AI strategy, with various agencies and legislative bodies examining everything from AI safety to intellectual property rights in the age of generative AI. This patchwork of state and potential federal regulations underscores the increasing complexity for businesses operating across jurisdictions. Companies must not only understand Connecticut's specific requirements but also monitor the evolving regulatory landscape nationwide to ensure holistic compliance.

Note: this article was drafted by AI - Google Gemini