Connecticut’s Comprehensive AI Law: An In-Depth Look at Substitute Senate Bill No. 5
On May 27, 2026, Connecticut took a bold step into the future by enacting a comprehensive artificial intelligence (AI) law known as Substitute Senate Bill No. 5 (Public Act No. 26-15). This far-reaching legislation introduces a variety of regulations designed to address the complexities and ethical dimensions of AI technology in areas such as companion chatbots, frontier model governance, and AI applications in employment decisions. With the law’s effective dates staggered to begin in October 2026, let’s delve into its key provisions and implications.
AI Companion Safeguards
As society becomes increasingly intertwined with AI, ensuring user well-being is paramount. The Act mandates that operators of AI companions implement robust protocols for several critical areas. First and foremost, the law requires these operators to establish mechanisms to detect and respond to user expressions of self-harm. They must also prevent outputs that could encourage harmful behavior and mitigate anthropomorphic deception—ensuring users are not misled into believing they are interacting with a human.
Crucially, if an AI companion could be mistaken for a human user, operators have a duty to provide clear disclosures. This information must be continuously visible or presented at defined intervals: at the start of the first interaction per day and subsequently at least hourly for minors. This transparency aims to inform users, particularly vulnerable individuals, of their interactions with AI technology.
Moreover, added safeguards are required for users identified as minors. Operators must restrict certain types of interactions with AI companions and furnish minors and their guardians with tools to manage screen time and account settings, promoting a safer online environment for younger users.
Frontier Model Governance
A unique aspect of the Act is its definition of "frontier developers," which includes those who train foundation models using substantial computational power—specifically exceeding 10^26 integer or floating-point operations. This contrasts with California’s legislation that applies a similar compute-based threshold but defines "frontier developers" in a standalone capacity.
The law distinguishes between "frontier developers" and "large frontier developers," the latter being those with annual gross revenues exceeding $500 million. A significant portion of the governance requirements centers on whistleblower protections. All frontier developers are prohibited from retaliating against employees who highlight risks associated with catastrophic outcomes. Large frontier developers face more stringent obligations, such as creating formal internal reporting systems by January 1, 2027.
These systems must include anonymous reporting channels and regular updates on investigations, ensuring accountability within organizations working with advanced AI models.
AI in Employment Decisions
The impact of AI technologies on the workforce is profound, hence the Act requires transparency in automated employment-related decision-making. Employers must inform candidates when they are interacting with such technologies—unless it would be inherently obvious.
Before any automated decision-making technology influences hiring or employment decisions, employers must furnish applicants or employees with a notice. This must outline critical details, including the technology’s name, its purpose, the nature of the decision-making process, the categories of personal data processed, and data sources. Additionally, employers must provide their contact information, thereby creating an open line for inquiries.
Generative AI Provenance
In an age where content authenticity is ever more essential, the Act stipulates that covered providers—specifically those who produce generative AI systems engaging over a million users monthly—must incorporate metadata or similar signals where feasible. These signals will help users identify whether content has been AI-generated or significantly altered, contributing to greater transparency and accountability in digital media.
Disclosures for AI Subscriptions
Subscription services related to AI technology face new obligations under the law. Providers are prohibited from entering into or renewing subscriptions without first delivering written notice that discloses the key terms and conditions. This step is aimed at ensuring that consumers are fully informed and that their agreement is based on clear comprehension of what they’re signing up for.
Regulatory Sandbox Program and Enforcement
In a bid to promote innovation while maintaining regulatory oversight, the Act directs the Connecticut Commissioner of Economic and Community Development to establish an AI regulatory sandbox program. This initiative allows businesses to test innovative products or services in a controlled environment, alleviating some regulatory burdens temporarily.
Finally, the enforcement of this comprehensive framework falls primarily to the Connecticut Attorney General, ensuring that accountability mechanisms are in place to uphold the law’s stipulations.
Connecticut’s AI legislation represents a significant step forward in the responsible governance of artificial intelligence technology. As these regulations begin to take effect, the state positions itself as a leader in ensuring that AI serves the public interest while promoting safety, accountability, and innovation in this rapidly evolving sector.