While Washington has spent 2026 debating whether to regulate artificial intelligence at all, Connecticut just started enforcing a law that does. On October 1, 2026, the core provisions of Senate Bill 5, formally titled the Connecticut Artificial Intelligence Responsibility and Transparency Act, phased into effect, putting new legal obligations on companies that build or deploy AI systems used in hiring, companion chatbots, and large-scale layoffs. Governor Ned Lamont signed the 39-section bill on May 27, 2026, after it was developed jointly with Attorney General William Tong and state Senator James Maroney, who has spent several legislative sessions trying to get an AI accountability law through the General Assembly.
What Changed on October 1
The provisions that became operative this month are specific rather than sweeping. They establish how responsibility is divided between the companies that build frontier AI systems and the companies that deploy them, add anti-discrimination language making clear that using an algorithm is not a legal defense against a discrimination claim, extend whistleblower protections to employees of large AI developers, and — in one of the law’s more granular requirements — add an AI-specific disclosure to the state’s WARN Act filings. Employers that notify Connecticut’s Department of Labor of a mass layoff must now also state whether the job cuts are connected to the employer’s adoption of AI or other new technology.
Inside the Connecticut Artificial Intelligence Responsibility and Transparency Act
SB 5 is built to phase in over roughly two and a half years rather than all at once, a structure its drafters say reflects how unevenly prepared different industries are to comply. Beyond October 2026, the law has three more milestones already on the calendar: by January 1, 2027, large frontier AI developers must have a functioning, confidential channel for employees to report safety concerns; by July 1, 2027, the state opens an application window for an independent verification organization pilot program meant to audit AI systems from outside the companies that build them; and by October 1, 2027, companies that deploy automated decision tools must give people a disclosure at the moment of interaction and written notice before an AI-influenced decision affects them.
Companion Chatbots Get New Guardrails
One of the law’s more closely watched sections targets companion chatbots — AI products designed to simulate ongoing relationships rather than answer discrete questions. Connecticut’s law was drafted and negotiated well before the wave of lawsuits elsewhere in the country accusing chatbot makers of failing to protect minors, but it lands in a policy environment shaped by exactly those cases, and lawmakers in Hartford cited the broader national pattern of harm allegations against companion-AI products as part of the rationale for including the category at all rather than leaving it for a future bill.
Employers Face New Disclosure Duties
For businesses, the automated employment decision tool provisions are likely to generate the most immediate compliance work. Any Connecticut employer using software to screen résumés, rank candidates, or make hiring, promotion, or termination recommendations now falls within scope of the developer-deployer framework that took effect October 1, even though the point-of-interaction notice requirements for workers don’t arrive until October 2027. Employment lawyers have been advising clients since the law’s passage in May to start auditing which vendors and tools qualify as automated employment decision tools well before the later deadlines arrive, rather than waiting until the final phase-in forces the issue.
Industry Pushback vs. Advocate Praise
Privacy and AI-safety advocates have broadly welcomed the law, pointing to the companion chatbot provisions and whistleblower protections as filling gaps that federal policy has left open amid congressional gridlock on AI legislation. Business groups have taken a more cautious view, warning that Connecticut is now one of a growing patchwork of states — alongside California, Illinois, Colorado, and New York — each writing its own definitions of what counts as a “high-risk” AI system or an “automated employment decision tool,” which raises compliance costs for any company operating across state lines. Attorney General Tong has defended the law’s design as proportionate, noting the multi-year phase-in was specifically built to give developers and employers time to adjust rather than face all obligations simultaneously.
The Next Deadlines on the Calendar
Connecticut is not finished. The whistleblower reporting-channel requirement arrives in less than three months, and the independent verification pilot program is scheduled to open for applications next summer — a mechanism few other states have attempted, and one that will test whether a government-sanctioned outside auditor can meaningfully check AI developers’ own safety claims. With SB 5’s full timeline stretching to October 2027, Connecticut has set itself up as a multi-year test case for whether state-level AI law can be specific enough to enforce without becoming obsolete before its last provision even takes effect — and other legislatures weighing their own AI bills in 2027 sessions will be watching to see whether Hartford’s staggered approach works better than a single effective date would have.
Photo: James St. John / BY via flickr
