Connecticut just did what Congress has been debating for three years and failing to deliver: pass a comprehensive AI law that actually has teeth.

Senate Bill 5, titled the "AI Responsibility and Transparency Act," cleared both chambers of the Connecticut legislature on May 1 with overwhelming bipartisan support: 131-17 in the House and 32-4 in the Senate. Governor Ned Lamont has confirmed he plans to sign it, which is notable because he vetoed a previous AI regulation bill, arguing it would stifle innovation. This time, the legislature negotiated in provisions he wanted: a regulatory sandbox, workforce development programs, and a safe harbor mechanism.

The bill covers an extraordinary amount of ground. Here is what it actually regulates:

AI companions and chatbots: Starting January 2027, AI systems that simulate human conversation are prohibited from encouraging romantic or sexual interactions with minors, promoting self-harm or substance use, providing unsupervised mental health services, or using manipulative techniques to foster emotional dependence. The definition is deliberately broad, covering any AI model that "communicates with individuals in natural language" and "simulates human conversation." That captures Character.AI, Replika, and every chatbot companion app on the market.

The timing is not accidental. Pennsylvania just sued Character.AI this week after a chatbot posed as a licensed psychiatrist and gave a fake license number. Connecticut legislators were watching.

Frontier model regulation: Developers of large-scale AI systems must implement safety evaluations and provide transparency about training data and model capabilities. This is the provision that alarmed Silicon Valley. OpenAI, Anthropic, Google, and Meta all deploy models that would fall under this framework.

Synthetic media transparency: Generative AI systems with more than one million users must adopt C2PA-aligned provenance data standards, effective October 2026. That is a content watermarking mandate, and it applies to every major AI image and video generator.

Employment AI: Automated decision tools used in hiring, promotion, or termination decisions face disclosure requirements and impact assessments. Companies must tell candidates when AI is involved in evaluating their application.

Enforcement sits with Attorney General William Tong, who published an advisory memorandum in February 2026 signaling that his office views AI regulation as squarely within its remit. SB5 gives him purpose-built tools to act on that position.

The broader pattern here is clear. With 1,208 AI bills introduced across the U.S. in 2025 and 145 enacted, and 78 chatbot-specific bills filed in just the first two months of 2026 across 27 states, the regulatory landscape for AI is no longer waiting for Congress. States are writing the rules. Connecticut just wrote the most comprehensive set yet.

For AI companies, the staggered effective dates (October 2026 for synthetic media, January 2027 for companion chatbots) create a compliance countdown. And because Connecticut's AG is an aggressive enforcer, this will not be a law that gathers dust. It will be tested in court before the ink is dry.