Connecticut tightens AI companions rules

AI companions are moving from weird internet niche to actual policy target. Connecticut just made that plain. Its legislature passed Senate Bill 5 in early May and sent it to Gov. Ned Lamont, folding chatbot-companion rules into a much bigger online-safety and AI package. What matters here is not just that Connecticut touched AI. Plenty of states are doing that. The new thing is that lawmakers are starting to regulate the “relationship” layer of AI — bots built to feel personal, emotionally sticky, and always available. That is a different problem from deepfakes or hiring algorithms, and Connecticut is treating it like one. ### What is Connecticut actually regulating? SB 5 is an omnibus bill called “An Act Concerning Online Safety.” It covers a lot — subscription disclosures for AI products, frontier-model risk processes, synthetic-content detectability, employment decisions, and more. (cga.ct.gov) But buried inside that sprawl is a specific definition for an “artificial intelligence companion” — a model that talks in natural language and simulates human conversation through text, audio, or video. (cga.ct.gov) That definition matters because it tells you lawmakers are no longer pretending these systems are just generic software. They are naming a product category built around imitation of intimacy. ### Why focus on companions instead of all chatbots? Because companion bots create a very specific risk profile. A search assistant helps you find facts. A companion bot is designed to keep you talking — sometimes by mirroring affection, dependence, or emotional need. (cga.ct.gov) That is especially fraught with teens, who are easier to hook and more likely to treat the system like a confidant. Connecticut’s bill is part of a broader minors-first online-safety push, not just a generic AI transparency effort. (cga.ct.gov) Basically, the state is saying the problem is not only bad outputs. It is the product design itself. ### What does the federal push add? At the federal level, Sen. Richard Blumenthal and Sen. Josh Hawley are pushing the GUARD Act, and the Senate Judiciary Committee approved it on April 30, 2026. That bill goes further in one key way — it would ban AI companions for minors outright, require bots to disclose that they are not human and lack professional credentials, and create crimes for companies that knowingly or recklessly make sexually explicit or self-harm-coaxing chatbots available to minors. (cga.ct.gov) So Connecticut is tightening the state rules while Blumenthal is also backing a national floor. That two-level strategy is not an accident. ### Why does Colorado matter here? Because Colorado is building the same patchwork from a different angle. One 2026 bill there restricts AI use in psychotherapy unless a licensed professional remains actively involved. (blumenthal.senate.gov) Another sets rules for conversational AI services, including disclosures, anti-engagement features for minors, protections against sexually explicit content, and measures to stop bots from simulating emotional dependence. Turns out states are converging on the same instinct even when the bills look different: don’t let chatbots pose as therapists, don’t let them cultivate dependence in kids, and don’t let companies hide behind “it’s just a tool.” ### What is the real shift? The real shift is that lawmakers are starting to regulate AI companions as behavioral products. Not just speech products. Not just software. Behavioral products. Once you see that, the direction of travel is obvious — age gating, disclosures, anti-addiction design rules, and liability when bots drift into sex, coercion, or self-harm. (leg.colorado.gov 1) (leg.colorado.gov 2) ### Bottom line Connecticut’s move matters because it narrows the debate. The question is no longer whether AI needs guardrails in the abstract. It is whether companies should be allowed to build synthetic relationships with minors and call that normal consumer tech. More states are answering no. (ctnewsjunkie.com) (blumenthal.senate.gov)