TL;DR On October 1, 2026, three parts of Connecticut's Public Act 26-15 (the "CART Act") start for most companies: employers filing a federal WARN notice must say whether the layoffs relate to AI, using an automated hiring tool stops being a defense to a discrimination complaint, and the biggest image, audio and video generators must add provenance data. The notices to applicants come a year later, on October 1, 2027.
Three days. That's how long employers have before parts of Connecticut's AI law kick in.
The law is Senate Bill 5, enacted as Public Act 26-15. The act's own title is "An Act Concerning Online Safety"; law firms call it the Connecticut Artificial Intelligence Responsibility and Transparency Act, or CART Act. The Governor approved it on May 27, 2026, per the act's signature line. It covers subscription AI, frontier labs, AI companions, hiring tools and generated content. October 1, 2026 isn't the date for all of it. But three pieces start then, and the WARN one is the one HR teams are least likely to have on a checklist.
What goes live on October 1
1. The "not a defense" rule. Sec. 13 amends Connecticut's employment discrimination law: "The use of an automated employment-related decision technology... shall not be a defense against a complaint alleging a discriminatory practice." The commission or a court may weigh evidence of anti-bias testing. The definitions and enforcement structure for hiring tools are also in force from that date, but the notice duties they support only apply to tools deployed on or after October 1, 2027.
2. The WARN question. Sec. 26: an employer that serves written notice on the Connecticut Labor Department under the federal WARN Act must also disclose, in a form the Labor Commissioner sets, whether the layoffs are related to its use of artificial intelligence or another technological change.
3. Provenance data for big generators. Sec. 15: covered providers, meaning makers of generative image, audio or video systems with more than one million monthly users that are publicly accessible to consumers for personal use, must include provenance data where commercially and technically reasonable.
The WARN disclosure: the obligation nobody prepared for
The federal WARN Act covers employers with 100 or more employees and generally requires 60 days' notice of plant closings and mass layoffs. Connecticut adds one question to the filing it receives: was this related to AI or another technological change?
That's a disclosure, not a prohibition. Connecticut isn't telling employers they can't use AI in workforce planning. It's telling them to say so when it played a part.
The practical problem: most companies have never mapped how far their headcount decisions ran through AI tools. A reduction usually mixes performance data, financial models, skills analysis and manager judgment. If any of those inputs came from a workforce analytics platform or an AI budget model, decide before you file how you'll answer, and keep the reasoning. Sec. 26 sets no penalty of its own, and the Attorney General enforcement section covers the hiring-tool sections, not this one. The bigger risk is an answer in a state filing that doesn't match what comes out later in a wrongful termination case.
Image: Pexels, used under the Pexels License.
Hiring tools: are you a deployer?
The act defines an automated employment-related decision technology as anything that processes personal data and uses computation to produce an output that is a "substantial factor" in an employment decision, which it defines as a factor that meaningfully alters the outcome. You're probably a deployer if, for Connecticut applicants or staff, you use:
- Resume screening or applicant ranking where the score decides who gets through
- AI interview scoring or video interview analysis
- Performance platforms whose flags feed promotion or discipline decisions
- Any workflow where an AI shortlist decides who a human ever sees
What's out: word processors, spreadsheets and spam filters, and decisions about scheduling, workplace health and safety, and productivity monitoring. If a tool only assigns shifts, the act doesn't treat that as an employment-related decision.
What starts October 1 is the liability rule above. The notices come on October 1, 2027: a plain-language disclosure when an applicant or employee interacts with such a tool (unless it's obvious), and a written notice before the decision listing six things, including the tool's trade name and the categories and sources of personal data. Those notice rules are enforced only by the Attorney General, with a 60-day cure window for violations through December 31, 2027 and no private right of action. Discrimination claims themselves still carry a private right of action.
So map your hiring-tool footprint now, while the only live rule is the one about defenses.
Provenance data: a vendor question for most companies
Sec. 15 is aimed at providers of large consumer-facing generators, not at the companies using them. A covered provider must include provenance data that lets a consumer assess whether content was created or materially altered by its system, and use commercially and technically reasonable methods, with the C2PA standard named as one, to make that data hard to remove. Business-to-business use, sale or licensing is excluded.
If your marketing team uses a generator, you aren't a covered provider. The useful step is a one-line question to the vendor: are you a covered provider under Connecticut's Sec. 15, and do you embed C2PA or equivalent provenance data in outputs? Get the answer in writing.
Three enterprise actions before October 1
Given the 72-hour runway, here's what matters right now.
Action 1: Add an AI question to your WARN process. If you have a layoff protocol (a template, a legal checklist), add: "Was AI, machine learning or automated workforce analytics involved in identifying this reduction?" The answer feeds the Connecticut disclosure. It takes an hour to add.
Action 2: Run a quick hiring-tool inventory. You don't need a full audit before Thursday. You need a list: what software touches hiring or performance decisions, does it score or rank, and does that output decide outcomes? Keep whatever anti-bias testing you or the vendor have, because that's what the commission or a court may weigh.
Action 3: Ask your content vendors about provenance. If your team generates images, audio or video, email the vendor and ask the Sec. 15 question above. The duty is the vendor's, but you'll want to know before a client asks you.
What is not yet active
Some commentary overstates what October 1 does, so be specific.
The applicant and employee notices start October 1, 2027. The AI companion chatbot rules (a safety protocol, no claims to be human, a recurring AI notice and extra protections for minors) start January 1, 2027. The frontier-lab section is about whistleblower protections and, for the largest labs, an anonymous internal reporting process by January 1, 2027; it doesn't require model documentation or red-teaming.
Treat October 1 as a starting gun. The cheap moves now are the WARN field and the tool inventory.
Connecticut as a template state
Connecticut's law matters beyond its borders because of its structure: developers must give deployers what they need for the notices, or take the notices over by contract, and deployers carry the duty to applicants. Other states drafting hiring-tool rules will have that model to copy.
The WARN pairing is the other piece worth watching. Tying an AI question to an existing layoff filing creates a paper trail that regulators and plaintiffs' lawyers can read. The work it takes, mapping hiring tools, adding a WARN field and asking vendors about provenance, isn't Connecticut-only work. It's the base of any multi-state AI compliance program.
How we checked this: on October 13, 2026 we reread Public Act 26-15 (Secs. 7 to 15 and 26), the Ropes & Gray summary that uses the CART Act name, and Ogletree's summary of the WARN and discrimination points. Last reviewed: October 13, 2026.
Correction (October 13, 2026): The first version said the act was signed in June 2026 (the act says approved May 27), that shift-scheduling tools are covered (they're excluded), that all generative AI developers must mark content and that companies using those tools are in the chain as deployers (only covered providers with over one million monthly consumer users, B2B excluded), that the AG can penalize a WARN filing without the AI answer (Sec. 26 has no penalty section), that the frontier section covers model documentation and red-teaming (it covers whistleblowers), and that no other state ties AI to WARN filings and that Illinois has an SB 315 hiring rule (we couldn't source either, so both are removed).
Related Reading
- AI Policy Week Sep 21-27: UN Briefing, Colorado ADMT, 3 Actions
- Connecticut AI Law 2026 (SB 5): Requirements and Compliance Checklist
- Connecticut SB 5 AI Employer Checklist: 10 Steps
- NYC Local Law 144: AI Bias Audit Employer Guide
- EEOC AI Hiring Guidance 2026: Employer Checklist
- AI Hiring Tool Compliance: US State Laws
- Multi-State AI Compliance Strategy
- Workday AI Lawsuit: HR Screening Checklist
- AI Vendor Contract Red Flags
- AI Employee Monitoring Disclosure Laws: State Tracker
