Connecticut just passed the broadest state AI law on the books
Connecticut's SB 5 layers chatbot disclosure, synthetic media provenance, frontier model rules, and AI hiring transparency into a single statute with staggered 2026 to 2027 effective dates.
Connecticut’s legislature passed SB 5 on May 1, and Governor Ned Lamont has said he plans to sign it. The bill, now styled the Artificial Intelligence Responsibility and Transparency Act, is the broadest state AI law on the books once enacted. It packages chatbot disclosure, synthetic media provenance, frontier model whistleblower protection, and AI hiring transparency into a single statute, with effective dates rolling out from October 2026 through January 2027.
For a SaaS company that ships a customer-facing chatbot, the most immediate concern is the AI companion section. Starting January 1, 2027, an operator must tell the user at the start of each interaction, and at least once an hour during a continuous session, that they are talking to AI rather than a person. That sounds simple, but it is a real product change. The interstitial has to fire on the first message of a session and again on the sixty-minute mark, and the disclosure has to be clear, not a tooltip buried under a settings menu. The same section adds detection requirements for user messages that suggest self-harm, plus stricter rules for any companion-style product available to minors. If your product positions itself as a friend, coach, tutor, or companion, that is the section to read first.
The synthetic content provenance section is the piece that reaches most AI vendors regardless of geography. Effective October 1, 2026, large generative AI providers must embed provenance data into generated or materially altered audio, image, or video output. Connecticut is following the same general approach other states have taken with synthetic content, pushing the burden upstream to the model provider rather than the platform that hosts the output. If you are running inference on someone else’s foundation model, your contract diligence should include whether the provider has implemented provenance signing and whether the API surfaces a way for you to pass that signal through to your end product. If you are the model provider, the engineering work needs to be staffed now, not in September.
The hiring section is the one most teams will hit fastest in operational terms. The statute requires employers to disclose when automated tools are used in hiring, and it removes the ability to rely on those tools as a defense against discrimination claims. A small SaaS company hiring engineers in Hartford will be on the hook just like a Fortune 500 employer. If you bought an AI sourcing or screening tool last year, ask the vendor for documentation of bias testing and an audit right in the contract. You will need both.
The single concrete action for this quarter is an inventory. Walk through every product surface and every internal workflow where an AI model produces output that a customer, candidate, or employee will see, and write down what the model does, who provided it, and what the user is told about it. That inventory is the foundation for compliance with Connecticut, with the chatbot disclosure laws that Nebraska, Idaho, and Oregon enacted last month, and with whatever the federal layer eventually adds on top.