Connecticut Passes One of Nation Most Comprehensive AI Laws With Landmark SB 5

--- title: "Connecticut Passes One of Nation's Most Comprehensive AI Laws With Landmark SB 5" slug: connecticut-sb5-comprehensive-ai-law category: policy story_number: "14" date: 2026-05-17 ---

# Connecticut Passes One of Nation's Most Comprehensive AI Laws With Landmark SB 5

The Connecticut Artificial Intelligence Responsibility and Transparency Act sailed through both chambers with overwhelming bipartisan margins, setting a new bar for state-level AI regulation even as the Trump administration pushes back against the state-by-state approach.

After years of false starts and a vetoed predecessor, Connecticut lawmakers finally crossed the finish line on May 1, 2026, when the state's House of Representatives voted 131-17 to pass Senate Bill 5 — a sweeping, 71-page artificial intelligence omnibus law that now heads to the desk of Gov. Ned Lamont, who has signaled he will sign it. The Senate had passed the bill 32-4 on April 21 after extensive debate.

The margins tell a story in themselves: nearly unanimous in the Senate, lopsided in the House, bipartisan in both chambers. This was not a party-line skirmish. It was a state legislature arriving at a rare consensus that AI had outrun existing law, and that waiting for Washington was no longer an option.

"There is no longer doubt that the nature of work, the nature of life, is going to change rapidly with the continued evolution of AI," said Rep. Roland Lemar, D-New Haven, co-chair of the General Law Committee, as House debate began. "This is about protecting people without stopping that innovation."

Even Republican members who might have been expected to balk came around. "The time has come," said Rep. David Rutigliano, R-Trumbull. "This isn't overbearing, we're not stifling innovation, we're not doing anything to stifle economic development."

A Deal Years in the Making

The bill's passage represents the resolution of a years-long standoff between the legislature and Lamont, who in 2025 threatened to veto a similar measure over fears that aggressive AI regulation would drive away businesses and hurt innovation. This session, a deal struck between Lamont and the bill's lead author, Sen. James Maroney, D-Milford, broke the logjam.

The final product incorporates provisions from multiple bills — including governor-requested measures on youth social media and AI chatbot safety (Senate Bill 86 and House Bill 5037) — and was formally renamed the Connecticut Artificial Intelligence Responsibility and Transparency Act on the day of the House vote.

"Governor Lamont made it a priority this session to fight for protections for Connecticut residents — especially children — from serious threats posed by emerging technology," said Lamont spokesperson Cathryn Vaulman. "The Governor looks forward to signing SB 5 into law."

Attorney General William Tong, who had thrown his support behind the legislation from the outset, framed the moment with urgency. "Artificial intelligence is exploding everywhere, with potential for good intertwined with severe risks," Tong said in a statement after the House vote. "Neither state nor federal law has kept pace with these developments, and today's vote is an important first step towards protecting Connecticut families."

What the Law Actually Does

The Act establishes several distinct regulatory regimes with staggered effective dates. Unless otherwise specified, provisions take effect October 1, 2026.

Automated Employment Decisions (deployer obligations effective October 1, 2027). The law targets what it calls Automated Employment-Related Decision Processes (AERDPs) — AI systems used as a "substantial factor" in hiring, promotion, discipline, or discharge. Employers deploying such tools must notify workers and applicants that an AI system is in use, explain its purpose, and — when an adverse decision results — provide a plain-language explanation of the principal reasons, including the data types and sources involved. Developers of these tools must furnish deployers with the information needed to comply, or may contractually assume those obligations directly. The law also amends Connecticut's anti-discrimination statutes to classify discriminatory use of AERDPs as an unlawful employment practice, and instructs courts and regulators to consider whether employers conducted anti-bias testing when assessing liability. Employment law firm Littler Mendelson notes the provisions "combine concepts from the current AI regulations in California and the European Union, taking a disclosure-focused approach."

AI Companions and Minor Protections (effective January 1, 2027). In perhaps its most visceral provisions, the Act prohibits operators of AI companion products from allowing minors to engage in romantic, erotic, or sexually explicit interactions with chatbots; from having AI companions offer unauthorized mental health services to minors; or from deploying companions that discourage minors from seeking help from licensed professionals or trusted adults. Operators that provide AI companions to users under 18 — when it is "reasonably foreseeable" that the companion could engage in any prohibited behavior — face civil penalties of up to $25,000 per violation. Uniquely, the minor-protection provisions carry a private right of action: affected users or their parents may sue within three years of a violation for actual and punitive damages. All AI companion operators must also build in protocols to detect expressions of self-harm or suicidal ideation and refer users to mental health resources, including the National Suicide Prevention Lifeline.

Synthetic Digital Content Watermarking (effective October 1, 2027). Developers of AI systems capable of producing or manipulating synthetic digital content — audio, images, text, or video — must ensure outputs are marked and detectable as AI-generated by the time consumers first encounter them. Technical solutions must be effective, interoperable, robust, and reliable. Exemptions apply to text-only content on matters of public interest unlikely to mislead a reasonable consumer, and to standard editing tools that do not substantially alter input data.

Frontier Model Whistleblower Protections (effective October 1, 2026). Developers training models using more than 10^26 floating-point operations are prohibited from retaliating against employees who raise concerns about "catastrophic risks" — defined to include risks of mass casualties, weapons of mass destruction assistance, or autonomous cyberattacks that evade human oversight. Large frontier developers (annual revenues exceeding $500 million) must establish anonymous internal reporting channels by January 1, 2027, and begin quarterly board-level reporting on flagged risks by May 1, 2027. Civil penalties of up to $1,000 per violation may be imposed by the Commissioner of Consumer Protection.

Subscription Disclosure (effective October 1, 2026). Entities offering AI products on a subscription basis must provide written pre-contract disclosures before any agreement is executed or renewed, covering key terms, conditions, and any usage limitations.

Regulatory Sandbox (plan due July 1, 2027). The Commission of Economic and Community Development must develop a sandbox program permitting limited testing of AI products under reduced licensing and regulatory requirements — a concession that reflects the governor's concern about keeping Connecticut competitive.

States Going It Alone — For Now

Connecticut's move comes against explicit federal resistance. The Trump administration has argued that a patchwork of state AI laws creates compliance confusion and undermines U.S. competitiveness. Connecticut is now joining California and Colorado as states that have enacted comprehensive AI-specific laws regardless, and other state legislatures are watching closely.

For companies, the compliance clock is already running. Firms using AI in hiring decisions — a category that may sweep in generative AI tools used for resume screening or interview analysis — must have disclosure infrastructure in place by October 1, 2027. Operators of AI companion products face the January 1, 2027 deadline for minor-protection protocols. Frontier-scale developers operating in the state face whistleblower obligations starting this October.

Not everyone was persuaded the law would work as intended. "It feels more like theater that we're trying to do something for these kids," said Rep. Bill Buckbee, R-New Milford, skeptical of the chatbot provisions. "There's really not a whole lot we can do with trying to regulate the Internet." Buckbee added a wry footnote: he had used AI to generate the questions he asked on the House floor.

Still, even lawmakers with reservations voted yes in overwhelming numbers. "I cannot vote no and hope that ignoring AI and social media is going to make the state of Connecticut any better," said Rep. Christie Carpino, R-Cromwell. "It's not going to protect our residents if I vote 'No.'"

For a state that spent years unable to translate AI anxiety into law, the passage of SB 5 marks a genuine inflection point — and a signal to other states watching whether the regulatory momentum that began in Europe and California can take hold across the rest of America.

---

Sources: CT Mirror (P.R. Lockhart), Davis Wright Tremaine (Nancy Libin), Littler Mendelson (Allan King and Niloy Ray), Freshfields, DLA Piper