United States of America: Introduced Bill for an Act Concerning Consumer Protections in Interactions with Artificial Intelligence Systems (SB 24-205) including testing requirement

On 10 April 2024, the Bill for an Act Concerning Consumer Protections in Interactions with Artificial Intelligence Systems (SB 24-205), including testing requirements, was introduced to the Colorado Senate. The Bill includes measures for developers and deployers of high-risk Artificial Intelligence (AI) systems. The high-risk AI system is defined as systems developed or substantially modified to make consequential decisions that impact consumer's access to or the availability, cost, or terms of various aspects of their life, including criminal justice remedies, education, employment, essential goods or services, financial or lending services, government services, healthcare, housing, insurance, or legal services. In particular, the developers will be required to provide deployers with information about the high-risk system, including the information necessary to conduct an impact assessment and issue public statements listing the types of high-risk systems developed or modified, along with details on known or potential risks of algorithmic discrimination and measures of addressing them. Additionally, the developers would have to notify the attorney general and known deployers of any discovered or anticipated algorithmic discrimination risks. Furthermore, the deployers would have to develop a risk management policy, conduct impact assessments, inform consumers of consequential decisions, and publicly disclose system details while reporting discrimination discoveries to authorities.Finally, the developers of general-purpose AI models are mandated to maintain documentation including policies for copyright compliance and detailed summaries of training data. They must also provide deployers with comprehensive information regarding the model's capabilities, technical requirements, design specifications, key design choices, optimization objectives, and data used for training, testing, and validation.