On April 21, 2026, Connecticut’s Senate passed a comprehensive regulatory proposal for artificial intelligence, marking one of the most detailed legislations of its kind in U.S. state history. The Senate voted 32-4 in favor of the amended Senate Bill 5 after extensive floor discussions. This 64-page bill covers 37 sections and addresses various aspects of AI’s integration into the commercial sphere, including regulations for emotional support chatbots, automated hiring processes, safety standards for frontier models, synthetic content labeling, employment protections, and the establishment of a publicly funded AI training academy. Rather than focusing on a single issue, the bill represents a wide-ranging legislative agenda that emerged from several committees, culminating in a last-minute comprehensive amendment drafted by Senator James Maroney, D-Milford, the bill’s primary author and co-chair of the General Law Committee.
The bill now heads to the House, which opted not to take action on last year’s AI proposal. As reported by CT Mirror, the Senate vote coincided with Hartford officials celebrating “AI Day” downtown.
The legislation features staggered effective dates within a two-year rollout period. Key provisions will take effect on October 1, 2026, while more complex requirements for operators of AI companions and automated job systems will start on either January 1, 2027, or October 1, 2027, depending on the specific section.
Legislative History and the Path to Passage
Connecticut has been debating AI legislation for several years. A previous measure, Senate Bill 2, also authored by Maroney in 2025, was approved by the Senate but stalled after Governor Ned Lamont threatened a veto, citing concerns that stringent regulations could stifle business and innovation. This session, with the end of the legislative calendar approaching and multiple AI-related bills pending, Maroney and his colleagues chose to consolidate various proposals into a single omnibus bill. This approach included elements from the Labor and Public Employees Committee regarding automated employment decisions, as well as two initiatives supported by the governor: Senate Bill 86 and House Bill 5037, which address regulatory sandboxes and youth social media and AI chatbot regulations, respectively.
The new bill mirrors the 2025 effort’s breadth and complexity. Maroney acknowledged the similarities, stating that the limited time remaining necessitated combining measures. “With time constraints, we’re working to merge where possible,” he remarked.
Floor discussions were lengthy, with Senator Paul Cicarella, R-North Haven, dedicating much of the first hour to having Maroney clarify the bill’s details section by section. Senator Henri Martin, R-Bristol, raised concerns about the bill’s extensive scope, stating, “There’s so much here. We may want to simplify and focus.” Senator Rob Sampson, R-Wolcott, voiced procedural concerns, questioning the timing of the last-minute amendment that left lawmakers little time to understand their vote. “I’ve just seen this bill. It feels like yet another last-minute amendment,” he said.
Senate Minority Leader Stephen Harding, R-Brookfield, contended that Connecticut, with its population of approximately 4 million, should defer legislation on rapidly evolving technology to the federal government. Senator Tony Hwang, R-Fairfield, suggested that such regulations might hinder business growth. “We might be creating obstacles to business success and innovation,” he stated.
Supporters of the bill emphasized the need for urgent action. Senator Saud Anwar, D-South Windsor, referenced a tragic case involving a teenager who died by suicide after interacting with an AI chatbot. “We must acknowledge that this is a significant issue,” he noted. “It’s crucial to establish protective measures.” Senate President Pro Tempore Martin Looney, D-New Haven, emphasized the historical implications of inaction, stating, “History will not be kind to those who advocate for waiting,” highlighting the rapid pace of change.
The Governor’s office offered a balanced perspective, affirming that while the advantages of AI must be acknowledged, user safety is paramount. “The Governor maintains that the potential benefits of AI must be weighed against user safety,” remarked spokesperson Cathryn Vaulman.
AI Companion Operators Face New Disclosure and Safety Requirements
Section 5 of the bill, effective January 1, 2027, outlines the most visible obligations for consumer-facing AI products. Under this section, operators are prohibited from providing AI companions without protocols to detect and respond to signs of suicide, self-harm, or imminent violence. If such signs are identified, the system must direct users to suitable mental health services, including the National Suicide Prevention Lifeline at 9-8-8.
The labeling requirement is explicit. Operators must clearly inform users that they are interacting with an AI and not a human, doing so at the start of each interaction and at least once every hour during ongoing sessions. This requirement for periodic reminders exceeds California’s SB-243, which mandated similar disclosures but did not specify an hourly frequency. Breaching Section 5 could lead to civil penalties of up to $15,000 per day, enforceable by the Attorney General.
Section 6 introduces additional requirements for users under 18. Operators are forbidden from providing AI companions to minors if it’s foreseeable that the system could encourage self-harm, suicidal thoughts, violence, or unhealthy behaviors. This prohibition extends to romantic or sexually explicit interactions, as well as systems leveraging specific engagement strategies to maximize user interaction. Violations involving minors incur fines of $25,000 per infraction. Parents or guardians may also initiate civil actions, with a three-year statute of limitations from the violation date.
The definition of an AI companion under this bill is carefully limited. It pertains to any AI model that communicates in natural language, simulating human conversation through text, audio, or video. Systems exclusively for internal business functions or straightforward customer service tasks are specifically excluded.
Automated Employment Decisions Require Layered Disclosure
Sections 7 through 13 establish a distinct regulatory framework for automated employment-related decision processes, effective October 1, 2027. This framework differentiates between developers, who create these systems, and deployers, who use them to evaluate employees and job applicants in Connecticut.
Automated employment decisions are defined broadly. This includes any computational process that influences hiring, promotions, discipline, or similar outcomes. This encompasses tools analyzing facial expressions, word usage, or vocal tones during online interviews, as well as systems screening resumes for specific keywords or patterns and targeting job advertisements to certain demographics.
Deployers have three distinct responsibilities once an automated decision-making process is triggered. Firstly, they must inform all employees or applicants that they are interacting with an automated system using clear language. Secondly, they must provide written disclosure outlining the system’s purpose, the nature of the decision being made, and the applicant’s option to opt-out of personal data processing. Thirdly, if the outcome is unfavorable, the deployer must give a brief explanation of the primary reasons for the decision, the role of the automated output in the outcome, the types of data processed, and the data’s source. If personal data not submitted by the applicant was utilized, the applicant is entitled to review and correct that information.
These responsibilities can be delegated to developers via a binding written contract specifying the duties transferred. Section 12 safeguards trade secrets, clarifying that trade secret information is not required to be disclosed, though the withholding party must inform the affected individual that some information is being withheld and provide a rationale.
Violations of these provisions are considered unfair or deceptive practices under Connecticut law. Enforcement falls solely to the Attorney General, who must first issue a notice of violation, allowing 60 days for resolution before further action is taken.
Frontier Developer Obligations and the Catastrophic Risk Threshold
Section 2 addresses frontier developers, defined as businesses in Connecticut using over 10^26 operations for training or fine-tuning frontier models. This threshold mirrors the computational bar established in the Illinois SB3444 frontier model bill, as previously reported.
The bill precisely defines catastrophic risk as any foreseeable risk that the use or deployment of a “foundation model” could lead to the deaths or serious injuries of over 50 people or cause $1 billion or more in damage from a single incident. Three scenarios are specifically listed: models enabling the creation or release of weapons of mass destruction; those engaging in destructive cyber activities or severe criminal conduct without appropriate human oversight; and scenarios where control of the model evades the developer or user.
However, certain risks are exempted. Risks from publicly available information that has been reproduced elsewhere are excluded from this definition, as are lawful federal government actions and instances where combining software did not substantially elevate risk.
Large frontier developers—those with over $500 million in annual revenue—must also implement specific internal protocols by January 1, 2027, including establishing an anonymous reporting mechanism for employees to disclose information indicating public health dangers posed by catastrophic risks. Companies must provide monthly updates to employees reporting such dangers and are obligated to investigate and address legitimate concerns. From May 1, 2027, they we’ll have to send quarterly reports to their leadership that summarize these disclosures and any actions taken in response. Officer or director allegations must be handled separately, ensuring those individuals do not receive relevant reports.
Whistleblower protections are strictly enforced. Companies cannot enforce policies that penalize employees for reporting catastrophic risks as per established Connecticut whistleblower protections. Civil penalties for violations amount to $1,000 each.
Synthetic Digital Content Must Be Labeled
Section 15 introduces mandatory labeling for synthetic digital content, effective October 1, 2027. Developers of AI systems capable of generating synthetic content must ensure that these outputs are identifiable as synthetic before consumers first interact with them. This labeling must be accessible and conform to existing accessibility standards.
The labeling standard specifies that solutions must be effective, reliable, and consistent with recognized technical standards while considering implementation costs for feasibility.
Exemptions apply for artistic works in audio, image, or video formats, where disclosure may be minimal. Text-only synthetic content aimed at informing the public and unlikely to mislead a reasonable person is also exempt, as are AI systems used for standard editing that do not significantly modify input data.
Connecticut AI Academy and Workforce Infrastructure
Section 19 mandates the establishment of a Connecticut AI Academy by the Board of Regents for Higher Education by December 31, 2026. This academy will curate online AI courses, promote digital literacy, and provide resources for individuals aged 13 to 20.
The academy will also support small businesses, nonprofits, educators, and individuals receiving unemployment benefits, ensuring they are informed about its existence and activities upon filing claims. The Secretary of State is tasked with disseminating information about the academy’s offerings for small businesses, while the Department of Housing will inform housing authority residents about the academy’s services.
Section 33 requires the Labor Commissioner to set up an Artificial Intelligence Workforce Research Hub within the Labor Department by July 1, 2026. This hub will analyze AI’s impact on Connecticut’s workforce and provide annual reports on its findings starting October 1, 2026, directed to relevant legislative committees.
AI Regulatory Sandbox and Working Group
Effective July 1, 2027, Section 3 directs the Commissioner of Economic and Community Development to create a plan for an AI regulatory sandbox program, allowing innovators to test their AI products under reduced regulatory requirements. A report with implementation recommendations must be submitted to the Governor and legislative committees by January 1, 2028.
Section 20 establishes a working group effective July 1, 2026, to recommend best practices for AI deployment in state services. This group will also propose regulations for general-purpose AI models and synthetic content disclosures on social media. Initial appointments must be made by July 31, 2026, with the first meeting occurring by August 31, 2026. The final report is due by February 1, 2027.
The working group will consist of 15 voting members from various sectors, including AI developers, industry representatives, educators specializing in tech policy, labor organizations, and small businesses. Non-voting ex-officio members will include the Attorney General, the Comptroller, the Treasurer, and others.
Connecticut Technology Advisory Board
Section 25, effective July 1, 2026, creates a Connecticut Technology Advisory Board within the Legislative Department. Eight voting members will be appointed, with equal representation from the House speaker, Senate president pro tempore, and minority leaders from both chambers. All members must possess credentials in artificial intelligence, technology, or related fields. The board’s responsibilities include formulating a state technology strategy to be updated at least biennially and providing recommendations to the legislative, executive, and judicial sectors.
What This Means for Marketing and Advertising Professionals
State-level AI legislation is becoming an essential compliance issue for technology firms and marketing platforms operating on a large scale. As documented by PPC Land, xAI filed a federal lawsuit in April 2026 to challenge Colorado’s SB24-205, claiming that state-level AI bias regulations infringe on freedom of speech concerning model outputs. Connecticut’s bill lacks the broad algorithmic discrimination framework xAI contested in Colorado; however, its provisions—particularly regarding automated employment disclosures and risk reporting—impose substantial compliance obligations that necessitate detailed operational records.
The automated employment provisions directly relate to systems that target job advertisements based on demographics, which is closely aligned with programmatic advertising practices. Any platform utilizing computational processes to determine which recruitment ads are displayed to which users—if those decisions significantly impact employment outcomes—may fall under the automated employment decision framework, requiring careful legal review.
The labeling requirements for synthetic content, effective October 2027, will have implications for AI-generated advertising content. Any system capable of producing AI-generated media must implement detectable labeling solutions compliant with interoperability standards—whatever those standards may evolve to be by that timeline. This aligns closely with the EU AI Act’s transparency obligations under Article 50, suggesting that both European and U.S. markets may adopt synthetic content labeling requirements simultaneously by 2027.
Timeline
- May 17, 2024: Colorado Governor signs SB24-205, the high-risk AI law challenged in federal court by xAI.
- August 1, 2024: EU AI Act comes into effect.
- 2025: Connecticut Senate passes Senate Bill 2, a comprehensive AI proposal by Senator Maroney, but it ultimately is halted due to a veto threat from Governor Lamont.
- October 13, 2025: California Governor signs SB-243, which requires disclosures by companion chatbot AIs.
- September 4, 2025: European Commission opens consultations on AI transparency guidelines under Article 50 of the AI Act.
- August 5, 2025: Colorado Attorney General publicly critiques SB24-205 as “problematic” and in need of revision.
- August 2025: A special Colorado legislative session delays SB24-205’s effective date until June 30, 2026.
- December 29, 2025: xAI files a federal lawsuit against California’s AB 2013 AI training data transparency law.
- February 2026: Connecticut General Assembly opens its February session and introduces Senate Bill 5.
- April 9, 2026: xAI files a federal lawsuit against Colorado to block SB24-205.
- April 2026: Illinois SB3444 is introduced, with backing from OpenAI, shielding frontier AI firms from massive liability.
- April 21, 2026: Connecticut Senate passes amended Senate Bill 5 with a 32-4 vote and sends it to the House.
- April 26, 2026: The substitute to Bill No. 5 is published, receiving joint endorsements from the Government and Labor (GL), Judiciary (JUD), and Appropriations (APP) committees.
- July 1, 2026: Sections 19, 20, 25, and 33 of the Connecticut bill come into effect, launching the Connecticut AI Academy, working group, Technology Advisory Board, and AI Workforce Research Hub.
- October 1, 2026: Subscription AI disclosure (Section 1), whistleblower protections for frontier developers (Section 2), definitions for automated employment frameworks (Sections 7-13), and synthetic content developer requirements (Section 15) take effect.
- January 1, 2027: AI companion safety and minor protection measures (Sections 4-6) become effective, while large frontier developers are required to establish anonymous reporting systems.
- October 1, 2027: Automated employment decision disclosure requirements apply to processes deployed on or after this date.
- January 1, 2028: A report with AI regulatory sandbox recommendations must be submitted by the Commissioner of Economic and Community Development to the Governor and relevant legislative committees.
Summary
Who: Connecticut General Assembly, advancing Substitute Bill No. 5 from the February 2026 legislative session. The bill received favorable recommendations from three committees: Government and Labor, Judiciary, and Appropriations.
What: A 64-page, 37-section omnibus AI bill addressing subscription-based AI disclosures, whistleblower protections for frontier developers, safety and minor protections for AI companions, automated employment decision transparency, synthetic digital content labeling, and the establishment of a Connecticut AI Academy and Workforce Research Hub, along with a state Technology Advisory Board, and a plan for an AI regulatory sandbox.
When: The Connecticut Senate approved the bill on April 21, 2026, with a 32-4 vote. The amended text of the bill was published on April 26, 2026, and it is now under consideration in the House. Effective dates for various provisions span from July 1, 2026, to October 1, 2027.
Where: Connecticut. The law applies to entities conducting business within the state that provide AI technology to consumers present in Connecticut and to frontier developers utilizing computational resources exceeding 10^26 operations.
Why: This bill responds to the rapid integration of AI systems across employment, consumer, and public health sectors, where Connecticut lawmakers determined existing legal frameworks were inadequate to ensure accountability, transparency, or protection—particularly for vulnerable users, such as minors engaging with AI companions and workers influenced by automated decision systems.
Share this article
