Twenty days from now, Connecticut becomes the first state to make a legal argument that will reshape how enterprise deployers treat automated systems in the workplace. On October 1, 2026, key provisions of Public Act 26-15 take effect — and the most consequential one is a single sentence buried in an amendment to the Connecticut Fair Employment Practices Act.
The amendment to CFEPA § 46a-60(b) states that the use of Automated Employment-related Decision Technology shall not be a defense against discrimination complaints. That clause does something no other state has done: it removes the algorithmic shield. If an AEDT tool — defined as any technology that processes personal data and uses computation to generate predictions, recommendations, classifications, rankings, or scores that substantially influence employment decisions — produces a discriminatory outcome, the employer cannot point to the software as an excuse. The liability stays with the human entity deploying it.
This matters because of what AEDT tools actually are in practice. A resume screener that filters candidates by pattern-matching against historical hiring data is an employment decision agent. A promotion-ranking engine that scores employees on inferred potential is an employment decision agent. A disciplinary trigger that flags workers based on productivity metrics is an employment decision agent. Connecticut’s statute does not name these tools as agents, but the functional definition captures them all — and it makes their deployment a source of legal exposure rather than a buffer against it.
The contrast with California sharpens the picture. As I covered in No Robo Bosses, California’s SB 947 takes a procedural approach: it mandates human verification for AI-informed terminations and imposes $500 penalties with a private right of action. Connecticut’s approach is substantive. It does not ask whether a human reviewed the decision. It asks whether the decision was discriminatory — and it refuses to let the employer blame the tool. The procedural route creates a compliance checklist. The substantive route changes the liability calculus for every company deploying AEDT in Connecticut.
Enforcement runs through the Connecticut Unfair Trade Practices Act. The Attorney General holds sole enforcement authority over notice and disclosure violations — there is no private right of action for those provisions. For violations occurring through December 31, 2027, the AG must first issue a 60-day cure letter before filing suit. This is not a litigation trap designed to punish early movers. It is an institutional ramp that gives employers time to audit their AEDT deployments, document their decision logic, and build the compliance infrastructure that the new regime requires.
The October 1 deadline also activates two other provisions that connect directly to the enforcement calendar I mapped in The Enforcement Wave. First, employers filing WARN Act notices must disclose whether workforce reductions are related to the employer’s use of AI or another technological change — Connecticut is the first state to fold an AI-specific question into its mini-WARN reporting regime. Second, frontier developer whistleblower protections take effect: entities developing foundational models are prohibited from retaliating against employees who report AI safety concerns, with civil penalties up to $1,000 per violation.
The federal gap makes state-level action like this structurally significant. The Congressional Research Service has confirmed that no known U.S. government guidance specifically addresses agentic AI. In the absence of federal standards, Connecticut is filling the employment layer of that void — not with a broad AI governance statute, but with a targeted amendment to existing employment discrimination law that changes who bears the risk when algorithmic tools make consequential decisions about people’s livelihoods.
Connecticut has also built institutional capacity to sustain enforcement beyond the initial deadline. The statute establishes an AI Policy Office overseen by an AI Policy Director, an AI Learning Laboratory Program, a Connecticut AI Academy, an AI working group, and a Technology Advisory Board. These are not decorative advisory structures. They represent a long-term commitment to monitoring how AEDT tools are deployed and whether employers are meeting their obligations.
For builders shipping AEDT tools into enterprise environments, the practical challenge is clear. Explainability moves from a technical preference to a legal requirement. If a resume screener rejects a candidate and that rejection correlates with a protected characteristic, the employer will need to demonstrate that the tool’s logic was non-discriminatory — and the tool builder will need to provide the evidence to support that defense. The black-box model that cannot produce a traceable rationale for its outputs becomes a liability, not an asset, in Connecticut’s legal framework.
The enforcement wave starts here. CRA Article 14 arrives tomorrow. The FTC personalized pricing comment period closes September 25. Maryland’s HB 895 takes effect October 1 alongside Connecticut. Colorado remains frozen by litigation. The EU AI Act has been active for five weeks with zero enforcement actions. But Connecticut is the first state to tell employers that the tool they bought to make employment decisions will not protect them when those decisions go wrong. The builders who understand that distinction will adapt. The ones who treat it as another compliance checkbox will learn otherwise.