Skip to content
Menu
Patent pending

The law library · Connecticut, United States · US state law

Connecticut: Public Act 26-15 (SB 5), online safety and AI

AN ACT CONCERNING ONLINE SAFETY · Substitute Senate Bill No. 5 · signed by the Governor 2 June 2026 · staged effective dates from 1 July 2026 to 1 January 2028

Page verified 13 August 2026 against the enrolled Public Act PDF on cga.ct.gov and the Governor’s signing announcement on portal.ct.gov. Every date below is read from the “Effective” line of the section it belongs to.

The AI half of Connecticut’s 2026 pair. One act, several regimes, each with its own clock: frontier-model catastrophic risk duties and automated employment-decision technology rules effective 1 October 2026, AI companion protections from 1 January 2027, provenance duties for large generative AI systems, restrictions on state agency AI use, and covered-minor protections from 1 January 2028.

Status

  • 2 Jun 2026Signed by Governor Lamont (Governor’s office announcement, portal.ct.gov).
  • 1 Jul 2026Secs. 17 and 18 (AI definitions, general-purpose AI model provisions) effective.
  • 1 Oct 2026Secs. 1 (AI subscription providers), 2 (frontier models), 7 to 12 (automated employment-related decision technology), 15 (generative AI provenance), and 38 (state agency AI limits) effective.
  • 1 Jan 2027Secs. 4 to 6 (artificial intelligence companions) effective.
  • 1 Oct 2027Deployer duties under Secs. 9 and 10 attach to automated employment-related decision technology deployed on or after this date.
  • 1 Jan 2028Sec. 39 (covered minors) effective.

The regimes inside it

  • FrontierSec. 2 defines “catastrophic risk”: a foreseeable, material risk that a frontier model provides expert-level assistance in creating a chemical, biological, radiological or nuclear weapon, or engages, with no meaningful human oversight, intervention or supervision, in conduct amounting to a cyberattack or a serious crime, with a threshold of more than one billion dollars of covered property in a single incident.
  • CompanionsSecs. 4 to 6: operators of artificial intelligence companions face duties keyed to minors and to users the operator knows or has reason to know are at risk.
  • Hiring AISecs. 7 to 12: developers and deployers of automated employment-related decision technology owe documentation and disclosure duties; violations are unfair or deceptive trade practices under Sec. 12. Trade secrets are protected by Sec. 11.
  • ProvenanceSec. 15: a “covered provider”, one whose generative AI system has more than one million monthly users and is publicly accessible to consumers, owes provenance duties for generated images, audio and video.
  • The stateSec. 38: limits on state agencies using AI in certain functions.

Why it is on this list

PA 26-15 puts Connecticut alongside Illinois SB 315, California SB 53 and New York’s RAISE Act in the frontier-model column, and alongside NYC Local Law 144 in the hiring column, in one statute. For an employer, the practical clock is 1 October 2027: deployments of automated employment-decision technology from that date carry the deployer duties.

The privacy half of the pair is Public Act 26-64 (SB 4).

What the signed record shows

Every regime in this act is answered with records of what the system did.

Frontier duties, hiring-AI disclosure, provenance for generated media: each one turns on being able to show, later, what the model produced, what a person oversaw, and when. A signed, tamper-evident record made at the time, verifiable by anyone with public mathematics, is that evidence in its cheapest form.

Regulayer for enterpriseThe AI audit trail →

Sources

Related

Checked against the source 13 August 2026: the enrolled Public Act PDF, section “Effective” lines read individually, and the Governor’s signing announcement. Sections not listed here carry their own dates in the act. Information, not legal advice.

Information, not legal advice. Every entry is verified against the issuing body’s own document; where a source is reporting rather than the document, we say so.