TL;DR Connecticut's AI Responsibility and Transparency Act (CART Act, PA 26-15) activates three employer obligations on October 1, 2026. Mass layoff WARN notices must now disclose AI involvement. The AEDT enforcement framework goes live. Synthetic content provenance marking is required. If you have employees or applicants in Connecticut, all three apply to you regardless of where you are headquartered.
Three days. That is what employers have to get three distinct AI compliance obligations in order before Connecticut's AI Responsibility and Transparency Act kicks in.
The Connecticut CART Act (Senate Bill 5, Public Act 26-15) was signed by Governor Ned Lamont in June 2026. It is one of the most comprehensive state AI laws in the US, covering employment automation tools, synthetic content, chatbot transparency, and frontier AI developer obligations. October 1, 2026 is not the full enforcement date for every provision. But it is the date three specific obligations become active, and at least one of them -- the requirement to disclose AI involvement in mass layoff notices -- has caught most HR and legal teams unprepared.
What goes live on October 1
The CART Act takes a phased approach to compliance. October 1, 2026 is the first major activation date. Three things happen:
1. The AEDT statutory framework is active. The definitions, enforcement structure, and trade-secret limitations that govern Automated Employment Decision Technology all take effect. Critically, so does the provision that "using an AEDT is not a defense to employment discrimination claims." From October 1, a discriminatory hiring decision made with the help of an AI screening tool carries exactly the same legal exposure as a discriminatory decision made by a human manager -- and potentially more, if you also failed to meet notice obligations.
2. The WARN Act AI disclosure obligation begins. Connecticut employers filing mass layoff notices under the federal Worker Adjustment and Retraining Notification Act must now include a statement disclosing whether AI tools or other technological changes contributed to the layoffs. The disclosure goes to the Connecticut Department of Labor alongside the existing WARN notice.
3. Synthetic content provenance requirements take effect. Developers of generative AI tools must ensure that AI-generated audio, image, and video output is marked with embedded provenance metadata to the extent technically feasible. Enterprises using those tools to produce marketing or communications content for Connecticut audiences are in the compliance chain as deployers.
The WARN Act disclosure: the obligation nobody prepared for
The WARN Act disclosure is the provision generating the most surprise among HR and employment counsel. The federal WARN Act requires covered employers (generally 100 or more employees) to give 60 days advance notice before mass layoffs or plant closings. Connecticut's CART Act adds a new data field to that notice: was this layoff related to your use of artificial intelligence or another technological change?
This is a disclosure obligation, not a prohibition. Connecticut is not telling employers they cannot use AI in workforce planning. It is telling them they must say so when they do.
The practical problem: most companies have never audited the degree to which their workforce planning was shaped by AI tools. When a company runs a headcount reduction, it typically uses a combination of performance data, financial modeling, skills-gap analysis, and manager judgment. If any of those inputs ran through an AI system -- a workforce analytics platform, an AI-assisted budget model, an automated skills-matching tool -- the WARN notice filed after October 1 should reflect that.
Companies that file a WARN notice without the disclosure and are later found to have used AI in the underlying decision face potential enforcement action from the Connecticut AG, and the factual inconsistency itself becomes a liability in any subsequent wrongful termination litigation.
The AEDT framework: you are probably already a deployer
The CART Act defines Automated Employment Decision Technology as any technology that uses computation to generate output -- predictions, recommendations, classifications, rankings, or scores -- that is used to make, guide, or assist employment decisions. If you use any of the following, you are a deployer:
- Resume screening software with AI-assisted ranking (HireVue, Workday Recruiting, Greenhouse)
- AI-assisted interview scoring or video interview analysis
- Performance management platforms that use machine learning to flag underperformers
- Scheduling tools that use AI to assign shifts or evaluate attendance patterns
- Any workflow that routes candidates through an AI-generated shortlist before human review
The reach is wide. Many companies that would not describe themselves as "using AI for hiring" are, in fact, deployers under this definition because their applicant tracking system includes AI ranking by default.
What does the AEDT framework actually require from deployers starting October 1? The short answer is that the full notice-to-applicants obligation -- the requirement to proactively notify each job candidate that their application is being evaluated by an AEDT -- is on a phased timeline, with broader applicant notice requirements taking effect later. What is active now is the enforcement structure: the Connecticut AG has authority to investigate, the non-defense provision is live, and companies that have not audited their AEDT stack have no grace period on the discrimination liability question.
This means the time to map your AEDT footprint is now, not when the applicant notice obligations fully activate.
Synthetic content provenance: the obligation that travels through your supply chain
The synthetic content provision targets developers of generative AI tools -- the Adobes, the OpenAIs, the Midjourney equivalents. They must embed provenance metadata into AI-generated audio, image, and video output, making it technically detectable as machine-generated.
For enterprise marketing and communications teams, the compliance question is two levels removed: does your AI content vendor already embed this metadata? If not, you are relying on a tool that does not yet meet Connecticut's requirement, and any content you produce for Connecticut audiences using that tool is potentially non-compliant.
The "to the extent technically feasible" qualifier matters here. Vendors cannot simply ignore the requirement on the grounds that embedding provenance is difficult -- they must demonstrate the technical barrier if they claim the exception. For enterprise teams, this means asking vendors directly: do you embed C2PA provenance or equivalent metadata in your generative outputs? The answer should be in writing.
Three enterprise actions before October 1
Given the 72-hour runway, here is what actually matters right now.
Action 1: Add an AI-use flag to your WARN process. If your company has a mass layoff protocol (a document, a template, a legal team checklist), add one question now: "Was AI, machine learning, or automated workforce analytics involved in identifying this workforce reduction?" The answer goes into the Connecticut WARN notice. This takes one hour to add to your protocol and potentially saves significant litigation exposure later.
Action 2: Run a quick AEDT inventory. You do not need a full audit before Thursday. You need a list: what software touches your hiring or performance management workflows, and does any of it use AI or automated scoring? Ask your HR and talent acquisition teams to name the tools they use. Cross-reference with those tools' documentation. Companies that can show they began this inventory in good faith before October 1 are in a materially better position than those who never looked.
Action 3: Ask your AI content vendors about provenance. If your team uses generative AI to produce images, audio, or video -- for advertising, internal training materials, social media -- email the vendor today and ask whether they embed provenance metadata in their outputs. If the answer is no or unclear, flag that for your legal team. The synthetic content obligation applies to developers, but your use of non-compliant tools creates reputational and regulatory exposure.
What is not yet active
It is worth being specific about what October 1 does not do, because some CART Act commentary overstates the immediate obligations.
Several CART Act provisions have later effective dates. The full pre-decision applicant notice (notifying each candidate before an AEDT decision is made) has a phased timeline. Chatbot safety obligations for companies operating consumer-facing bots in Connecticut also take effect later. Frontier AI developer obligations around model documentation and red-teaming have their own schedule.
The October 1 date is a starting gun, not a finish line. Legal teams in Connecticut-adjacent companies have been slow to act on this law because its comprehensive scope made it feel like a 2027 problem. It is not. Companies that treat it as the moment to begin their CART Act assessment will already be behind on the obligations that are live today. Companies that use it as a forcing function to at least inventory their AEDT stack and update their WARN process will have done the minimum necessary to avoid the most immediate enforcement risk.
Connecticut as a template state
Connecticut's CART Act matters beyond its borders for two reasons. First, it is one of the most comprehensive state AI laws enacted so far, and its structure -- developer obligations flowing documentation to deployers, deployers bearing the direct compliance and disclosure burden -- is likely to influence how other states draft their own laws. Illinois (SB 315, effective January 1, 2027), New York, and other states are watching Connecticut's implementation closely.
Second, the WARN Act + AI disclosure pairing is novel. No other state has linked AI use disclosure to an existing employment law filing requirement in this way. If it works as an enforcement mechanism -- creating a paper trail that plaintiffs' attorneys can use in wrongful termination cases -- other states will copy it.
Enterprise teams that build their AI documentation and disclosure practices around Connecticut's requirements now will be in a stronger position when the next state's law activates on a similar model. The effort to map your AEDT vendors, update your WARN checklist, and confirm provenance compliance in your content tools is not Connecticut-specific work. It is the foundation of a multi-state AI compliance posture that will matter in Illinois, New York, and beyond starting in 2027.
