Berkshire Blog

Colorado and Connecticut Update State AI Laws, Go Different Directions

Written by Brian Marentette, Ph.D | July 24 2026

States are continuing to fill the gap left by the absence of a single federal AI framework, and the latest updates out of Colorado and Connecticut show two very different paths - both of which matter for HR and talent acquisition teams.

Colorado

Colorado’s newly signed SB 26-189 (Revised Colorado AI Act) (signed May 14, 2026; effective January 1, 2027) repeals the 2024 Colorado AI law (SB 24-205) before it ever took effect, and it scales back the earlier, more prescriptive approach in favor of a more transparency-focused approach to regulating AI in the workplace. The revised law applies to developers and deployers of automated decision-making technologies (ADMT) that materially influence “covered consequential decisions,” explicitly including employment.

Colorado moved away from requiring things like mandatory risk management programs and annual impact assessments, and toward operational requirements you can see in the workflow. These types of requirements include, for example, notice to applicants or employees when AI is used in consequential decisions, disclosures about how automation affects outcomes, options for individuals to seek human review or corrections, recordkeeping, and clear recourse mechanisms when an adverse outcome occurs. The Colorado Attorney General will enforce the new law and there is a 60-day notice and cure period before enforcement (except for recurring violations).

Connecticut

Connecticut went the other direction of Colorado with less of a single “one size fits all” framework, and more of a modular law with targeted rules across use cases, including a set that directly applies to the use of an automated employment-related decision process (AERDP) in employment. The new Connecticut Artificial Intelligence Responsibility and Transparency Act was signed May 29, 2026 and provisions become effective between October 2026 and January 2028, depending on the section.

The law requires that employers using AERDP notify applicants/employees that they are interacting with automated technology. Before making any employment-related decision using AI, a deployer, including employers, must also provide written notice covering, among other items:

  • the purpose of the tool and the nature of the employment-related decision;
  • the personal data being used, how it will be assessed and the sources for that information; and
  • contact information for the deployer.

The law also requires that employers disclose in written notices whether certain reductions in force are “related to the employer’s use of artificial intelligence or another technological change.” This provision of the law applies to RIFs covered by the federal Worker Adjustment and Retraining Notification (WARN) Act, which requires employers with one hundred or more employees to provide advance notice of certain facility closures or large RIFs.

The law places a clear expectation on vendors: developers who market/sell employment AI must provide deployers with the information they need to perform these duties (or contractually assume the duties). Connecticut also adds an important litigation/compliance reality check by amending its Fair Employment Practices Act to state explicitly that using AI is not a defense to a discrimination complaint.

Key Takeaways for Employers

Despite the roll-back of mandatory impact assessments in Colorado, employers should still be prepared to respond to claims of algorithmic bias. As we’ve seen with pay transparency, when employees and applicants have greater insight and more information about an employment practice, they will likely ask more questions and have more information to potentially challenge the process. Employers should establish a monitoring and governance program to stay ahead and manage potential risk with using artificial intelligence tools. Here are a few key considerations for employers in their governance activities:

  • Inventory where AI is a substantial factor in an employment decision process.
  • Vet vendor research and existing bias audits/impact assessments/non-discrimination analytics conducted by the vendor.
  • Align vendor contracts to deliver the required information to respond to transparency requirements.
  • Build applicant/employee notices into the workflow and have alternative applicant assessment methods or human reviews.
  • Conduct regular bias audits/impact assessments/non-discrimination analytics to evaluate possible algorithmic bias.
  • Make sure you still have a defensible, non-discriminatory process, even if the AI tool isn’t making selections or directly ranking applicants or employees.