Key Takeaways
- Connecticut’s Senate Bill 5, passed May 11, 2026, creates the “Automated Employment-related Decision Technology” (AEDT) framework, requiring employers to comply with new AI hiring and personnel regulations from October 1, 2026.
- The Act explicitly states that using an automated system is not a defence against discrimination claims, placing the burden on employers to conduct and document anti-bias testing, effective October 1, 2026.
- By October 1, 2027, employers using AEDT must provide written notices to applicants and employees detailing the AI’s role in adverse decisions and the data types processed; failure to comply is enforceable by the Connecticut Attorney General under CUTPA, with no private right of action for individuals. Connecticut is poised to become among the strictest states in the country on AI hiring tools, after the General Assembly passed Senate Bill 5 on May 11, 2026, with Governor Ned Lamont expected to sign it into law.
Defining Automated Employment-related Decision Technology
The law’s employment framework turns on a deliberately broad definition of Automated Employment-related Decision Technology. Under SB 5, AEDT covers any computational process that produces an output, a score, rank, recommendation, constraint or classification, that serves as a “substantial factor” in an employment-related decision. Hiring, promotion, discipline, compensation, performance evaluations and termination all fall within scope.
That definition reaches resume screening software, third-party hiring platforms, assessment tools and performance analytics systems. Generic software, spreadsheets, word processors, is excluded, as are tools used only incidentally or for purely descriptive purposes. The practical focus is on predictive AI that carries algorithmic bias risk. Generative AI tools could also fall under the framework if deployed in a way that produces discriminatory outcomes.
Shifting Liability and the Role of Bias Testing
The sharpest edge of SB 5 is its amendment to Connecticut’s anti-discrimination statute: as of October 1, 2026, employers cannot use the fact that an automated system made a recommendation as a defence against a discrimination claim. If a third-party AI tool contributed to a discriminatory outcome, the employer remains liable.
The law does offer a practical mitigation route. Courts and agencies may consider evidence of bias testing when assessing an employer’s conduct, including the quality, recency and scope of testing, the results obtained, and what the employer did in response. That language stops well short of a safe harbour, but it makes pre-deployment and ongoing AI bias testing a near-essential risk management step for any employer using these tools in Connecticut. The approach echoes aspects of California’s Fair Employment and Housing rules, which have similarly pushed employers toward proactive AI auditing.
The statute also draws a clear line between developers and deployers. From October 1, 2026, developers must provide employers with enough information to meet their compliance obligations, where the technology is marketed or intended to materially influence employment decisions. Developers can contractually assume certain notice obligations, but primary accountability stays with the employer deploying the system. This mirrors the compliance architecture emerging in other AI liability frameworks, where the organisation closest to the affected individual bears the greater burden. The deployment of AI agents in regulated decisions is drawing similar scrutiny in financial services.
Transparency and Disclosure Mandates
By October 1, 2027, employers using AEDT that interacts directly with applicants or employees must provide written notice explaining the technology’s role. The notice requirement is general for routine use, but the obligations become more specific when a system contributes to an adverse decision.
When an employer uses AEDT to reach a negative employment outcome, the affected individual must receive a “high-level statement” setting out the principal reasons for that decision. That statement must address the degree to which the AI output contributed to the outcome, the type of data the system processed, and the source of that data. The intent is to give individuals enough information to understand, and potentially challenge, how AI shaped a consequential decision about their employment.
A trade secret safe harbour is built into the framework, allowing developers and deployers to withhold genuinely proprietary information, provided they notify the recipient that information is being withheld and on what basis. That carve-out will likely become a point of contention as workers and advocates test its limits in practice.
Pilot Program for Independent Verification and Enforcement
SB 5 creates a pilot programme for independent verification organisations, launching July 1, 2027. These third-party bodies, approved by the Connecticut Department of Consumer Protection, will assess whether AI systems meet defined risk mitigation and safety standards. Verification by one of these organisations can be offered as evidence in civil proceedings, but it does not create a safe harbour or presumption of compliance in enforcement actions. The programme sunsets in 2030 and is explicitly framed as a testbed for future AI auditing and certification regimes.
Enforcement of the employment provisions sits exclusively with the Connecticut Attorney General, acting under the Connecticut Unfair Trade Practices Act. There is no private right of action for individuals under these sections. For violations occurring on or before December 31, 2027, employers may be offered a 60-day cure period to address compliance failures, a grace period that reflects the novelty of the obligations rather than any softening of the underlying liability standard.
The legislation also adds an AI disclosure requirement to federal WARN notices, effective October 1, 2026. Employers must state in those notices whether layoffs are connected to the use of AI or other technological changes, extending the Act’s transparency logic beyond hiring and into workforce reduction decisions. Connecticut’s approach to AI in employment sits within a broader national debate over algorithmic accountability; the growing political pressure for AI oversight legislation suggests more states may follow.
Preparing for Compliance
With the first compliance deadline less than six months away, employers operating in Connecticut face a real time pressure on preparation. The immediate priority is an inventory of every automated tool used in employment decisions, mapped against the AEDT definition to identify which systems fall within scope. Vendor contracts should be reviewed now to confirm that developers can supply the information employers will need to meet their obligations.
From there, organisations need a phased plan covering anti-bias testing protocols, internal documentation procedures, and the notice and disclosure workflows required by the October 2027 deadline. Cross-functional coordination between HR, legal and IT will determine how smoothly that preparation goes. The cure period available through December 2027 provides some room for course correction, but it is not a substitute for early action, particularly given that the Attorney General’s enforcement mandate under CUTPA carries real reputational and financial risk for non-compliant employers. For more coverage of AI policy and regulation, visit our AI Policy & Regulation section.
Originally published at https://autonainews.com/connecticut-lawmakers-pass-sb-5-on-ai-hiring-tools/
Top comments (0)