On 1 October 2026, Connecticut stopped letting software take the blame. If a tool helped decide who you hired, promoted or let go, using it is no longer a defence to a discrimination complaint. The same day, WARN layoff notices started asking whether AI is involved. The headline duty, telling people before a tool helps decide their job, doesn't start until October 2027. The part that can cost you money is already live.
What is the Connecticut AI law?
The Connecticut AI law is Public Act 26-15 (Senate Bill 5): for employers, it says that using an automated tool to help make a hiring, promotion, discipline or firing decision is not a defence to a discrimination complaint, that WARN layoff notices must say whether AI is involved, and that from 1 October 2027 you must tell applicants and employees before such a tool helps decide their job. It applies to anyone doing business in Connecticut who uses a covered tool there, with no size threshold for the notices.
The act's official title is "An Act Concerning Online Safety", and most of its 74 pages deal with other things: AI companions, social media and minors, and state programmes. Governor Lamont approved it on 27 May 2026 and announced the signing on 2 June. Some law-firm summaries call it the CART Act. This guide covers only the employment sections, for the people in HR, operations and IT who'll have to make it work. It's based on the signed act. It isn't legal advice.
What changed on 1 October 2026?
Two things took effect. Neither has a grace period.
Using an AI tool is not a defence to discrimination
Connecticut's anti-discrimination law now says that "the use of an automated employment-related decision technology ... shall not be a defense against a complaint alleging a discriminatory practice." You can't argue that the vendor's model made the call. If the outcome discriminates, it's your outcome.
The same section gives you something back. The Commission on Human Rights and Opportunities, or a court, "may consider evidence of anti-bias testing or similar proactive efforts" to avoid discrimination, including "the quality, efficacy, recency and scope" of that testing, its results "and the response thereto." Testing isn't required, but it's the evidence that counts in your favour.
This applies to employers with one or more employees. That's how Connecticut's discrimination law already defines an employer.
WARN notices must say whether AI is involved
The federal WARN Act requires employers with 100 or more employees (not counting part-time staff) to give 60 days' notice of a plant closing or mass layoff, including to the state. Since 1 October, an employer that sends that notice to the Connecticut Labor Department must also disclose "whether the layoffs that are the subject of such written notice are related to the employer's use of artificial intelligence or another technological change." The Labor Commissioner decides the form and manner.
The act doesn't define "another technological change", and it doesn't say what makes a layoff "related" to AI. We couldn't find a published form from the Labor Department as of October 2026. If a reduction in force is on your plan, decide your answer, and the evidence for it, before the notice goes out.
| Date | What happens |
|---|---|
| 27 May 2026 | Governor approves Public Act 26-15 |
| 1 October 2026 | No defence for automated tools in discrimination claims. AI question on WARN notices |
| 1 January 2027 | University of Connecticut report on AI and the state's workforce due to the legislature |
| 1 October 2027 | Disclosure and pre-decision notice duties start. Vendors must give you what you need for them |
| 31 December 2027 | Last day of violations the attorney general can offer a 60-day cure for |
Does the Connecticut AI law apply to your company?
The notice duties apply to a "deployer": a person "doing business in the state who deploys an automated employment-related decision technology in the state." There's no minimum headcount. They protect employees and job applicants in Connecticut. If you hire or employ people in Connecticut, plan as if you're in scope.
Two parts have their own thresholds. The discrimination change follows Connecticut's discrimination law, which covers employers with one or more employees. The WARN question only reaches employers who file a federal WARN notice, so in practice employers with 100 or more employees.
What counts as an automated employment decision tool?
The law's term is "automated employment-related decision technology". It means "any technology that processes personal data and uses computation to generate any output", such as a "prediction, recommendation, classification, ranking, score or other information", that is "a substantial factor used to make or materially influence an employment-related decision."
Three definitions do the work.
The decision. Employment-related decisions are decisions to hire, promote, discipline or discharge someone, renew their employment, select them for training or an apprenticeship, or set their tenure, terms, privileges or conditions of employment. Two kinds are left out: decisions that cause only a "nonmaterial change" in someone's tasks, hours or assignments, and decisions about "workplace health and safety, scheduling and planning or productivity monitoring."
The substantial factor. A factor is substantial if it "meaningfully alters the outcome" of the decision. The law's examples are "a constraint, ranking, score, recommendation or classification."
What's excluded. Ordinary software is out "insofar as" it doesn't make or materially influence the decision: word processors, spreadsheets, databases, spam filters, spellcheckers and the like. So is a system used in a way that's "incidental" to the decision, and information that's "purely descriptive, diagnostic or statistical" and not relied on.
Ranking is in, summarising is out
That gives you a working test for hiring.
Likely covered:
- Your applicant tracking system scores or ranks applicants, and recruiters work from that order.
- An assessment or video-interview platform scores candidates.
- A recruiter asks a chatbot which applicants to shortlist.
- Performance scores from a tool feed decisions about promotion, pay or termination.
Likely not covered:
- AI drafts the job ad, the interview questions or the rejection email.
- AI schedules interviews or plans shifts.
- A dashboard reports statistics that nobody uses to decide about a person.
- Productivity monitoring on its own (though if its scores later drive a firing, the firing decision is in scope).
The definition doesn't depend on the word AI. A rules-based scoring system can be covered, and a general-purpose chatbot can be covered too, depending on what someone asks it to do. Put the line in writing: your AI usage policy should say that chatbot output isn't used to rank or choose people.
What must employers tell people from October 2027?
From 1 October 2027, a deployer has two duties.
1. Say when people are dealing with a tool
If a covered tool is "intended to interact with" an employee or applicant, such as an interview chatbot or an assessment, you must tell each person who uses it, in plain language, that they're interacting with it. There's no need to say so where "a reasonable person would deem it obvious."
2. Give written notice before the decision
If a covered tool helps make an employment decision about someone in Connecticut, you must give them written notice before the decision is made. It must disclose:
- that you've deployed an automated employment-related decision technology;
- the purpose of the technology and the nature of the decision;
- the tool's trade name;
- the categories of personal data it will analyse, and how that data will be assessed in reaching a decision;
- the sources of that personal data;
- your contact information.
Note what isn't there. Unlike Colorado, Connecticut doesn't require an explanation after a rejection, a right to correct data or a human review. The duty is to tell people before. But item 4 means you must be able to say how the tool assesses people's data, and you can't write that sentence without your vendor.
What your vendor owes you
From 1 October 2027, the developer of a covered tool must give you "all information" you need to meet both duties. That applies only if the tool was "advertised, marketed, configured, contracted for, sold or licensed" to materially influence employment decisions. A vendor can also agree in a contract to take on your notice duties, as long as the contract says clearly which ones.
Nobody has to reveal trade secrets. But if you or your vendor hold information back for that reason, you must send a notice saying what's being withheld and why.
Penalties and enforcement
Only the attorney general enforces the notice duties. A violation is an unfair or deceptive trade practice under the Connecticut Unfair Trade Practices Act, which allows a civil penalty of up to $5,000 for each wilful violation. For violations on or before 31 December 2027, the attorney general can first issue a notice of violation and give you 60 days to cure it. People can't sue you over the notices themselves.
The bigger exposure is the change that's already live. Discrimination complaints go through the existing route, and employees and applicants can bring them. From 1 October 2026, "the tool did it" doesn't answer one.
Anti-bias testing is a standing date, not a purchase check
Read the mitigation clause again: "quality, efficacy, recency and scope" of the testing, its results, "and the response thereto." Recency means a vendor's bias report from the year you signed the contract won't carry much weight in the year a complaint arrives. Response means the record has to show what you did when a test found a problem. That's not a one-off audit. It's a review that comes round on a date, with an owner who acts on it.
Our framework gives every AI use in production an owner, a backup and a review date, and asks three questions each time: is it still used, still safe and still worth it? Review dates, in module 7 of our free course, sets that up in fifteen minutes a tool. For a hiring tool, add one standing item to "still safe": the latest test of its outcomes by group, what it showed, and what changed because of it. That record is the evidence Connecticut's law tells a court to weigh.
If you also hire in Colorado or California
Colorado and California also passed laws on AI in employment decisions this year. The three overlap but don't match.
| Connecticut (PA 26-15) | Colorado (SB 26-189) | California (SB 947) | |
|---|---|---|---|
| Signed | 27 May 2026 | 14 May 2026 | 30 September 2026 |
| Main duties start | 1 October 2027 (some parts 1 October 2026) | January 2027 | 1 July 2027, according to SHRM |
| Focus | Notice before decisions; no "the tool did it" defence | Notice, explanation after rejection, correction, human review | Human oversight of AI-assisted discipline and firing |
Colorado's law asks for the most after a decision; see our guide to the Colorado AI law. California's "No Robo Bosses Act" bars employers from relying solely on an automated decision system to discipline or fire someone, requires human oversight and verification when one assists, and requires employers to tell workers when one was used. If you operate in all three, build to the strictest version of each duty once, rather than three times.
What to do now
Six steps, in order. Each one produces what the next one needs.
01 · List every tool that touches a decision about a person
Start with what you bought as HR tech: applicant tracking, assessments, video interviews, background screening and performance management. Then look for AI features vendors have switched on inside software you already pay for, and ask recruiters and managers what they use chatbots for. The free Shadow AI Audit covers those built-in features. Your AI list, in module 3 of our course, turns what you find into a list with an owner on every row.
02 · Sort each use: does it rank, or only describe?
For each tool, ask whether it produces a score, ranking, recommendation or classification that someone sees before deciding about a person. If it does, treat it as covered. If it only drafts, schedules or reports statistics, write down why, so you can show your reasoning later.
03 · Test the covered tools, and diarise the next test
The no-defence rule is already live. For each covered tool, get the most recent test of its outcomes across groups, from the vendor or your own data. Record what it showed and what you did about it. Set the next date now.
04 · Decide your WARN answer before you need it
If you have 100 or more employees and any reduction in force is planned, agree who decides whether it's "related to" AI or another technological change, and what evidence they'll use. Check the Labor Department's form before filing.
05 · Ask vendors for what the October 2027 notice needs
You'll need each covered tool's trade name, the categories of personal data it analyses, how it assesses them and where the data comes from. Ask now, while you can still change vendors. Our vendor checklist has the other questions to ask at the same time. If a vendor offers to take on your notice duties, get it in the contract, naming each duty.
06 · Draft the two notices
Write the plain-language disclosure for tools that interact with people, and the six-part written notice for decisions. Decide where each one appears (careers page, application form, review process) and when it's sent, so it arrives before the decision. Have counsel review both before October 2027.
What's still open
- The WARN form. The Labor Commissioner decides its form and manner, and the act doesn't define "another technological change".
- How the attorney general reads "substantial factor". The notice duties turn on whether a tool "meaningfully alters the outcome", and there's no guidance yet on where that line falls.
- The workforce report. The University of Connecticut must report to the legislature by 1 January 2027 on tracking AI-related layoffs and job losses, which may shape what comes next.
We'll update this article when the Labor Department publishes its form or the attorney general issues guidance. The date at the top of the page shows when we last checked.
Sources
- Public Act 26-15 (Substitute Senate Bill 5), Connecticut General Assembly, May 2026: the signed act. Sections 7 to 14 (employment), 26 (WARN) and 29 (workforce study).
- Governor Lamont signs legislation establishing youth online safety protections, regulations over artificial intelligence, Office of the Governor, June 2026: the signing announcement.
- Chapter 814c, Human Rights and Opportunities, Connecticut General Statutes: the definition of employer in section 46a-51.
- Chapter 735a, Unfair Trade Practices, Connecticut General Statutes: civil penalties in section 42-110o.
- 29 U.S. Code § 2101 and § 2102, Legal Information Institute: which employers the federal WARN Act covers and the 60-day notice.
- SB26-189, Automated Decision-Making Technology, Colorado General Assembly: the Colorado law's signing and start dates.
- Newsom signs McNerney's No Robo Bosses Act of 2026, Office of Senator McNerney, September 2026: what California's SB 947 requires.
- Newsom signs revamped 'No Robo Bosses Act' into law, SHRM, October 2026: secondary reporting, used for SB 947's start date.
- New Connecticut law restricts employer AI use, mandates notice for AI-caused RIFs, Ogletree Deakins, May 2026: secondary commentary, used for context only.