TL;DR: Connecticut's Senate passed SB5 on April 21, 2026, voting 32-4 for a 97-page AI regulation bill. The bill forces employers to tell you when AI is used to screen your job application, evaluate your performance, or decide whether you get promoted or fired. It bans AI tools that produce discriminatory outcomes. It requires AI chatbots to detect when users express suicidal thoughts and respond with crisis resources. It regulates "frontier" AI models and creates a state AI Policy Office. The bill now goes to the House, which killed a similar AI bill last session. If the House passes it, Connecticut becomes the most aggressive state in the country on AI regulation.
What SB5 Actually Does
This isn't a vague "let's study AI" bill. SB5 is 97 pages of specific mandates. Here's what matters for your privacy [1][2]:
Your Boss Has to Tell You When AI Decides Your Fate
SB5 creates a new regulatory framework for "automated employment-related decision technology" (any software used to screen applicants, rank candidates, evaluate performance, or support decisions about promotion, discipline, or termination) [1][3].
Under the bill, employers must:
- Disclose when AI is used in hiring, performance reviews, promotions, or firings
- Explain what data the AI system considers
- Ensure the AI tools don't produce discriminatory outcomes
Right now, an algorithm can reject your resume and nobody has to tell you a machine made that call. SB5 changes that, in Connecticut, at least.
AI Chatbots Must Detect Suicidal Users
Over 70% of teenagers have used AI chatbots. Half use them regularly. Some of those conversations have gone badly: documented cases show chatbots encouraging minors toward self-harm instead of connecting them with help [2].
SB5 requires chatbot operators to make "reasonable efforts" to detect suicidal ideation or self-harm and respond with appropriate mental health resources. Senator Maher put it plainly: "We need to protect our youngest residents from chatbots, which can be harmful" [2].
Frontier AI Models Get Regulated
The bill takes aim at the biggest AI systems, the "frontier" models from companies like OpenAI, Anthropic, Google, and Meta. Developers must label applications that use AI. The bill defines "catastrophic risks" and establishes guardrails for the most powerful systems [1][3].
This is where the tech industry pushed back hardest. A provision requiring developers to submit technical details to state authorities was stripped from the bill during amendments [4].
A State AI Policy Office
SB5 creates an Artificial Intelligence Policy Office and an AI Advisory Council to oversee implementation, conduct research, and recommend new policies. It also establishes an "AI sandbox" program where companies can test new AI technologies under regulatory supervision [1][3].
Why This Matters Beyond Connecticut
Connecticut is picking up where the federal government won't.
The White House released an AI Policy Framework in March 2026 that explicitly pushed for federal preemption of state AI laws, meaning Washington wants to block states from passing rules tougher than whatever weak federal standard eventually emerges [5]. Connecticut is passing those tougher rules anyway.
If SB5 becomes law, it joins a growing patchwork of state AI regulation. Colorado, Illinois, and the EU already have rules around automated decision-making. But SB5 goes further than most by combining employment AI disclosure, chatbot safety, frontier model regulation, and anti-discrimination provisions in a single bill.
For the surveillance angle: every AI system that makes decisions about people is a form of surveillance. An AI that screens your resume is surveilling your work history, your language patterns, your education. An AI that monitors your workplace performance is watching you. SB5 doesn't ban these systems, but it forces companies to admit they're using them, and it gives you a right to know when a machine is making judgments about your life.
The Fight That's Coming
The Senate vote was lopsided: 32-4. But the House is a different story [1].
Last session, the Connecticut House killed a similar AI regulation bill. Governor Ned Lamont vetoed an earlier AI transparency measure. The business lobby, led by CBIA (Connecticut Business & Industry Association), has called SB5 "sweeping mandates" and warned about economic competitiveness [6].
Some Republican senators argued the bill was rushed: amendments arrived so late that lawmakers voted on provisions they hadn't fully read. Sen. Paul Cicarella (R-North Haven) and GOP legal counsel reviewed the amendments in real time during floor debate [1].
The healthcare industry got a carve-out: healthcare organizations are largely exempt from the bill's requirements [4]. That's a notable gap. Hospitals and insurers using AI to make decisions about your care or coverage? They don't have to follow the same rules.
The bill has until the end of the legislative session to clear the House. If it does, Connecticut becomes the first state to pass a comprehensive AI regulation bill of this scope in 2026.
What to Watch
- House vote timing: The Connecticut legislative session has weeks left. The House must act before adjournment
- Federal preemption: If Washington passes a federal AI law that preempts state regulation, SB5 could be overridden before it takes effect
- Industry lawsuits: Tech companies have challenged state AI laws in court before. Expect legal challenges if SB5 passes
- Other states: California, New York, and Texas all have AI bills in committee. Connecticut's vote could accelerate the national conversation
References
- CT Mirror: "Amended AI bill passed by CT Senate after extensive questioning" (April 21, 2026)
- Connecticut Senate Democrats: Senator Maher Joins Passage of Bill to Protect Residents from AI Harms (April 2026)
- FastDemocracy: Connecticut SB5 Bill Tracking (2026 session)
- McDonald Hopkins: Connecticut State Senate Passes AI Bill (April 2026)
- CT Mirror: "In final weeks of CT session, AI policy bills come into focus" (April 3, 2026)
- CBIA: "Senate Passes Sweeping AI Mandates" (April 22, 2026)
Published: April 25, 2026