- within Strategy topic(s)
Highlights
- Connecticut Senate Bill 5 is a wide-ranging artificial intelligence (AI) bill, but the provisions most likely to affect businesses are the new requirements governing the use of AI in employment decisions.
- The employment decision requirements focus on transparency and disclosure without imposing risk assessments or internal governance obligations, and the law does not address "algorithmic discrimination," a term that has drawn scrutiny in other jurisdictions.
- The law takes effect on a staggered basis beginning October 1, 2026, with employment-related AI obligations applying to systems deployed on or after October 1, 2027.
Connecticut Gov. Ned Lamont signed Senate Bill (SB) 5 into law on May 29, 2026, establishing a novel framework regulating numerous use cases involving the use and sale of artificial intelligence (AI) technology. The wide-ranging bill contains obligations related to provision of subscription-based AI services, frontier models, AI companions that could be used by minors, use of AI in hiring decisions and watermarking of synthetic content, among other topics.
Scope of Requirements
The 74-page bill covers a wide range of issues:
Consumer Protection
Subscription-based providers of AI technology must disclose key terms and conditions to consumers before entering into or renewing a subscription.
Similar Existing Laws: None specific to AI, though multiple states have laws regulating auto renewal.
Frontier Model Safety
Frontier model developers must establish anonymous reporting channels for employees, investigate reports for catastrophic risk, take immediate action to eliminate dangers, and provide regular updates to whistleblowers and quarterly reports to officers and directors. Retaliation against whistleblowers is prohibited.
Similar Existing Laws: California's Transparency in Frontier Artificial Intelligence Act contains whistleblower protections, among broader obligations.
AI Regulatory Sandbox
Connecticut's Commissioner of Economic and Community Development must develop a plan for a regulatory sandbox allowing applicants to temporarily test innovative AI products under reduced regulatory requirements.
Similar Existing Laws: Utah, Texas and Delaware all have laws establishing regulatory sandbox programs.
AI Companion Regulation and Child Safety
Operators of AI companions must implement protocols to detect and address user expressions indicating suicide, self-harm or imminent violence risk, and must disclose to users that they are communicating with AI. The bill prohibits providing AI companions to minors if the companion could foreseeably encourage harmful behaviors, provide unauthorized mental health services, engage in romantic or sexual interactions, or use manipulative engagement tactics.
The bill uses a standard definition of AI companions, carves out customer service chatbots, and provides a private right of action for aggrieved users or their parents/guardians.
Similar Existing Laws: California (SB 243), New York (N.Y. Gen. Business Law Section 1700, et seq.), Washington (House Bill 2225), Georgia (SB 540), Idaho (SB 1297), Iowa (Senate File 2417), Nebraska (Legislative Bill 525) and Oregon (SB 1546) all regulate AI interfaces that are capable of sustaining relationships with users across multiple interactions using natural language.
Automated Employment Decision Processes (AERDP)
Employers must disclose to employees and applicants when they are interacting with an "automated employment-related decision process" and describe its general nature, unless obvious to a reasonable person.
AERDP is defined broadly to include computational processes generating any output – including constraints, ranks, scores, recommendations or classifications – that is not a de minimis factor in employment decisions. The definition excludes word processing, spreadsheets, data storage, spellchecking and calculators if these tools do not make an employment decision.
"Employment-Related Decision" includes any decision based on personal data to recruit, hire, promote, discipline or discharge an individual; review employment; select for training or apprenticeship; or determine tenure, terms, privileges or conditions of employment.
Employers must also provide pre-use notices when automated processes are a substantial factor in employment decisions. Notices must state that an AERDP has been deployed, describe the decision's purpose and nature, and inform employees/applicants of their right to opt out of profiling under Connecticut's data privacy law (though this right does not apply in the employment context).
If the AERDP is a substantial factor in an adverse decision, employers must provide a high-level explanation of the principal reasons, the AERDP's role, the type and source of data used, and an opportunity to examine and correct personal data. "Substantial factor" includes any factor that assists in making and is capable of altering an employment decision's outcome.
Developers are required to provide deployers with information necessary for compliance.
Violations are considered unfair or deceptive trade practices enforceable solely by the state's attorney general, with a 60-day cure period and no private right of action. The bill also incorporates AERDPs into Connecticut's existing employment discrimination prohibitions.
Similar Existing Laws: The Colorado AI Act, as amended by SB 26-189 on May 15, 2026, includes employment decisions as a regulated use case, as does the California Consumer Privacy Act (CCPA).
Labeling Synthetic Content/Watermarking
Developers of "artificial intelligence systems" and general-purpose AI models must ensure synthetic digital content is marked and detectable as such. The disclosure excludes text-only content, assistive editing functions and content unlikely to mislead a reasonable person. "Artificial intelligence system" means any machine-based system that infers from inputs how to generate outputs, including synthetic digital content.
For audio, image or video content forming part of an evidently artistic, creative, satirical or fictional work, the disclosure can be limited so as not to hinder display or enjoyment.
Similar Existing Laws: Multiple states have enacted laws requiring disclosure of AI-generated content, including California (Assembly Bill 2655 and AB 2839).
Collective Bargaining Agreements
State employers may not use AI in ways that "modify or impair" collective bargaining unit members' rights, benefits or privileges (e.g., to reduce wages or assume employee duties).
Workforce Development and Education
The law promotes AI workforce training and requires employers serving Worker Adjustment and Retraining Notification (WARN) Act layoff notices to disclose whether layoffs relate to AI or other technological changes.
AI Safe Harbor Program
SB 5 establishes a voluntary safe harbor under which AI users may apply to the Department of Consumer Protection (DCP) for approval. Participation provides a presumption of compliance with applicable consumer protection and data privacy laws, plus a cure period of at least 10 days before enforcement action.
When Does the Law Go into Effect?
The law has staggered effective dates, with most provisions taking effect on October 1, 2026. AI companion requirements are effective on January 1, 2027, and obligations related to automated employment decision processes apply to systems deployed on or after October 1, 2027.
How Do the Obligations Related to Use of AI in Employment Decisions Compare to Existing Laws?
Connecticut's SB 5 joins a small but growing cluster of state laws addressing AI in impactful decisions. The closest comparisons are Colorado's AI Act (SB 205, amended by SB26-189, now effective January 1, 2027) and the California Consumer Privacy Act regulations (effective January 2027) – the only state frameworks covering employment decisions. SB 5 is narrower than these laws in that it regulates only employment decisions, whereas the Colorado and California laws also cover significant consumer-facing decisions.
State comprehensive privacy laws include opt-out rights for profiling via automated decision-making that produces legally or similarly significant effects. However, although some laws include employment actions as covered decisions, all except California exclude individuals acting in an employment context from their scope – making this a null obligation outside California.
All three frameworks – California, Colorado and Connecticut – are built on transparency and ability to correct decisions. Though details vary, all three require pre-use notice to impacted employees/applicants. Note, however, that Colorado enacted a bill this month (SB 26-189) that will lighten Colorado AI Act obligations.
| California | Colorado | Connecticut | |
|---|---|---|---|
| Pre-Use Notice | Required | Required | Required |
| Right to Opt Out | Businesses must offer applicants and employees either a right to opt out of the use or to appeal an adverse decision. | No. The obligation was struck by SB 26-189. | Deployers must inform employees of their right to opt out of personal data processing under Connecticut's privacy law, but that right does not apply in the employment context under Connecticut law. |
| Right to Appeal | Businesses must offer applicants and employees either a right to opt out of the use or to appeal an adverse decision. | Yes. SB 26-189 includes a right to human review and reconsideration of adverse decisions. | No |
| Right to Access and Correct | Individuals have a right to "access ADMT" which would provide them with information about how AI has been used, as well as a right to access and correct personal information. | Yes. SB 26-189 creates a right to access and correct personal information when the use of an ADMT results in an adverse outcome | Individuals have a right to access and correct personal information used in the decision-making process. |
| Risk Assessment Requirement | Yes | No. The requirement was struck by SB 26-189. | No |
Will SB 5 Face Legal Challenges?
The Trump Administration previously issued an executive order stating a policy to prevent a patchwork of state AI regulations, so it is possible that the administration may seek to block implementation through litigation. Notably, the legislation does not reference "algorithmic discrimination," a term that drew legal opposition in other jurisdictions, including a legal challenge to the Colorado AI Act before the act was amended to narrow its focus.
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.
[View Source]
