Section 1 - Description
Section 2 - Scope
Section 3 - Requirements
Section 4 - Penalties
Section 5 - Status
Connecticut’s SB1103 regulates the usage of AI and automated decision systems by state agencies. The bill establishes an “Office of Artificial Intelligence”, tasked with developing procedures that state government agencies must follow when building, buying or using AI and automated decision systems. The bill also requires all state agencies to inventory, monitor and periodically assess their usage of AI.
This law only applies to the public sector in Connecticut and covers all state agencies, which is defined as any “department, board, commission, council, institution, office, constituent unit of the state system of higher education, technical education and career school or other agency in the executive, legislative or judicial branch of state government.”
The bill defines an “automated decision system” as any machine-based system or application used to “make, inform, or materially support” a critical decision.
Critical decisions domains include:
- Education and vocational training
- Essential utilities
- Family planning services
- Financial services, including any services from a creditor or mortgage broker
- Legal services
- Government benefits
- Public services
Any state agency must:
- Inventory and assess their usages of automated decision systems every two years. Any usages that are inconsistent with their initial procedures must be deactivated.
- Share any relevant information with the Office of Artificial Intelligence, including advance notice of 60 days if the agency plans to develop or acquire a new system
- Implement safeguards to ensure that such automated systems are properly applied, utilized and functioning, and provides appropriate training to all personnel responsible for designing, utilizing or procuring such automated systems.
The Office of Artificial Intelligence must:
- Inventory all automated systems used, developed or procured by state agencies. This inventory must include the name, description, training data used, output, and financial impact of the system. It must also include any known instances of discrimination caused by the system. The Office must review this inventory to determine if any systems harmed the rights of Connecticut citizens or posed a risk to the state.
- Develop and require the procedures that state agencies must follow when developing, using or procuring automated decision systems. The Office may also force agencies to halt development or usage of any system that doesn’t comply with the automated systems procedures.
No penalties are set. The bill applies to the public sector only.
The bill has been passed in both the Connecticut House and Senate signed by the governor.
Enforcement of the bill starts at different times:
Before January 2024, The Office of Artificial Intelligence must conduct its first inventory of the state government’s AI usage. Starting January 2024, state agencies may not use automated decision systems that don’t follow the state’s procedures. Before January 2025, state agencies must also inventory their automated decision systems.