Connecticut enacts comprehensive AI law

Connecticut lawmakers gave final passage on May 1 to Senate Bill 5, a sweeping measure that combines artificial-intelligence rules, online-safety provisions and consumer protections into a single state law. The House approved the bill 131-17 after the Senate had passed it 32-4, according to Connecticut General Assembly records. The measure has since been posted as Public Act 26-15 and was transmitted to Gov. Ned Lamont on May 14. The law does not operate as a single broad licensing regime for AI companies. Instead, SB 5 packages together requirements for subscription-based AI services, frontier-model developers, AI companion operators, employers using automated decision tools, synthetic media providers and several state workforce and policy programs. The General Assembly’s bill analysis lists staggered effective dates beginning October 1, 2026. (cga.ct.gov) ### What exactly did Connecticut pass? Public Act 26-15 is titled “An Act Concerning Online Safety,” but the text covers a wide range of AI-related subjects. The bill status page says it establishes requirements concerning artificial intelligence systems, artificial intelligence companions and automated employment-related decision processes, while also creating an Artificial Intelligence Policy Office, an AI learning laboratory program, a Connecticut AI Academy, an AI working group and a Connecticut Technology Advisory Board. (cga.ct.gov) The Office of Legislative Research bill analysis says the law also requires consumer disclosures for subscription-based AI providers, internal reporting processes for certain frontier developers, detectability rules for synthetic digital content and state workforce programs tied to AI skills and apprenticeships. ### Which AI uses face the clearest new rules? Beginning October 1, 2026, subscription-based providers of AI technology cannot enter into or renew a subscription unless they give consumers written notice of key terms and conditions and the consumer accepts them in writing, the enacted text says. (cga.ct.gov) Those disclosures must include material limits on the service and whether the provider can reduce or eliminate functionality. (cga.ct.gov) The same October 1, 2026 date appears across several other provisions. The bill analysis says AI companion operators must provide notices to users, adopt protocols to detect certain expressions such as self-harm and are barred from offering certain high-risk companion functions to minors. It also says employers using automated employment-related decision technology must provide disclosures and written notices, and that discriminatory use of such tools can still violate state anti-discrimination rules. (cga.ct.gov) ### What does the law say about synthetic media and frontier models? Section summaries in the bill analysis say certain providers with more than 1 million monthly users must include embedded metadata designed to make covered media harder to tamper with or strip. The analysis also says developers of synthetic digital content must include labels or technical measures making that content detectable as synthetic. (cga.ct.gov) The bill analysis says frontier developers — defined there as companies training large-scale AI models with large amounts of computing power — face whistleblower protections for employees who report specified catastrophic risks. Large frontier developers with more than $500 million in annual gross revenue last year must maintain an internal anonymous reporting process for those risks, according to the analysis. (cga.ct.gov) ### How hard is this law to enforce? The Office of Legislative Research analysis says several provisions are enforced by the Connecticut attorney general, and some violations are treated as violations of the Connecticut Unfair Trade Practices Act. The enacted text also assigns roles to the insurance commissioner and consumer protection commissioner for safe-harbor applications in some parts of the law. (cga.ct.gov) Connecticut’s approach differs by section. Some parts impose disclosure duties, some create workplace and child-safety restrictions, and others set up state institutions and planning bodies rather than direct operating rules for private companies. That structure is reflected in the General Assembly’s own summary and in outside legal analyses published after passage. (cga.ct.gov) ### When do businesses and residents need to pay attention? October 1, 2026 is the first major effective date in the enacted law, covering provisions including subscription disclosures and several AI-related operating rules in the statute text and bill analysis. Another key date is October 1, 2027, when the public act says certain rules for automated employment-related decision technology apply to deployers using those tools in employment decisions. (cga.ct.gov) July 1, 2027 is another milestone in the bill analysis. By then, the analysis says the state’s Board of Regents and partner institutions must develop AI-related education and training programs through the Connecticut AI Academy framework. (cga.ct.gov 1) (cga.ct.gov 2)