On May 1, the Connecticut legislature passed an artificial intelligence (“AI”) safety, transparency, and consumer protection bill (“SB 5”). While the Colorado legislature takes steps to streamline existing requirements for developers and deployers of AI systems, Connecticut has passed a multi-part framework that will impose requirements on large frontier developers, operators of AI companions, developers of general purpose models capable of generating synthetic digital content, and developers and deployers of automated employment-related decision technologies. Governor Ned Lamont (D) is expected to sign the bill into law.
Key provisions in the law include the following:
- Employee Reporting Protections: Frontier developers, defined as any person doing business in Connecticut who intends to train, initiates the training of, or trains a foundation model using, or intending to use, a quantity of computing power greater than 10^26 power integer or floating point operations, will be required to implement reporting procedures and protections for employees. For example, frontier developers will be prohibited from retaliating against employees who report certain public health or safety risks. Further, large frontier developers (frontier developers with more than $500M in revenue) must implement internal reporting processes for employees to report certain public health or safety risks and must provide periodic updates to reporting employees. The Attorney General can enforce this section and seek civil penalties of up to $1,000 per violation.
- AI Companions. SB 5 will prohibit operators (entities who provide an AI companion to or operate an AI companion for a user) from providing AI companions (i.e., AI with a natural language interface that provides adaptive, human-like responses to user inputs and is able to sustain a relationship across multiple interactions) unless certain conditions are met. For example, operators must implement a protocol that uses evidence-based methods to detect indicators of risk of suicide, self-harm or imminent physical violence and implement reasonable measures to prevent the AI companion from generating outputs that encourage these activities. Additionally, operators must implement reasonable measures to prohibit and prevent an AI companion from claiming to be a human or generating any outputs that refute or conflict with any disclosure that the AI companion is not a human being. Violations of this section will be enforceable by the Attorney General under the state’s unfair and deceptive practices statute.
- AI Companions Provided to Minors. Operators will be prohibited from providing AI companions to any user that the operator knows, or has reason to believe, is under 18 years of age unless certain protective measures have been put into place that meet or exceed industry standards. For example, such measures must take steps to prevent the AI companion from (a) encouraging the user to harm others, (b) discouraging the user from seeking mental health services from a licensed mental health professional, or (c) discouraging the user from seeking assistance from an appropriate adult. A violation of this section will be enforceable by the Attorney General under the state’s unfair and deceptive practices statute.
- Automated Employment-related Decision Technology. The law will create requirements for developers and deployers of automated employment-related decision technology, defined as technology that processes personal data and uses computation to generate any output to make or materially influence an employment-related decision. Among other things, deployers will be required to provide employees and applicants a written notice with information about the use of the technology, and developers will be required to provide all information deployers need to comply with their obligations under the law. This provision of the law will be enforceable by the Attorney General under the state’s unfair and deceptive practices statute.
We will continue to update you on meaningful developments in these quarterly updates and across our blogs.