What's Happening?
On October 6, the Colorado Attorney General (AG) released an interim redline of its proposed rules for implementing the state’s Automated Decision-Making Technology Act (SB 26-189) and Chatbot Safety Act (HB 26-1263). This release precedes an October 26 rulemaking
hearing, with the goal of having rules adopted before the January 1, 2027, effective dates of these laws. The revised draft introduces a new rule that establishes when a system’s output is presumed to 'Materially Influence' a consequential decision. This presumption applies if a decision-maker reviews an output—such as a score, ranking, recommendation, or threshold—or uses it to screen data, and the output aligns with the final outcome. A deployer can rebut this presumption by demonstrating the output was only a 'De Minimis Factor.' Tools used solely for summarizing, organizing, or presenting information for human review are excluded from the statute's definition of the technology. The draft also clarifies that deployers remain responsible even when a third party operates the technology on their behalf, provided the deployer uses the output for decision-making. Additionally, the redline revises adverse outcome disclosure requirements, removing the previous mandate for deployers to state 'principal reason(s)' for an adverse outcome and disclose underlying inferences or scores. Instead, deployers would describe the technology’s role with specificity.
Why It's Important?
These proposed rules are critical for establishing a regulatory framework for artificial intelligence and automated decision-making in Colorado, impacting businesses that utilize such technologies and the consumers affected by them. The concept of 'Materially Influence' is central, as it defines the scope of accountability for deployers of AI systems. By clarifying when an AI's output is considered influential, the rules aim to ensure transparency and fairness in decisions made by or with the assistance of AI. The revisions to disclosure requirements are particularly important for balancing consumer rights with business operational realities, potentially streamlining compliance for companies while still providing consumers with essential information. This regulatory effort positions Colorado at the forefront of state-level AI governance, potentially influencing how other states approach similar challenges. The rules will affect various sectors, including finance (e.g., credit unions), employment, and other areas where automated systems make consequential decisions, ensuring that these technologies are deployed responsibly and ethically.
What's Next?
The Colorado AG will hold a rulemaking hearing on October 26, where stakeholders can provide further input on the revised proposed rules. Written comments will be accepted until this date. Following the hearing, the AG's office will consider all feedback and may make additional revisions before finalizing the rules. The ultimate goal is to adopt these rules before the January 1, 2027, effective dates of the Automated Decision-Making Technology Act and the Chatbot Safety Act. Once adopted, businesses operating in Colorado that use automated decision-making technologies will need to review and adjust their practices to ensure compliance. This includes re-evaluating their disclosure processes, understanding the 'Materially Influence' presumption, and potentially modifying their AI systems to meet the new requirements. Consumers will gain new rights regarding access to and correction of personal data used in automated decisions, although some aspects of these rights have been refined in the latest draft. The implementation of these rules will mark a significant step in regulating AI in the U.S.
Beyond the Headlines
The Colorado AG's efforts to regulate automated decision-making technology delve into complex ethical and societal implications of AI. The debate over 'Materially Influence' and the extent of disclosure requirements reflects a broader societal discussion about algorithmic transparency and accountability. By defining when an AI's output is considered influential, the state is attempting to draw a line between mere data processing and actual decision-making, which has profound implications for legal liability and consumer recourse. The removal of the requirement to stay adverse outcomes pending correction or human review, and the exclusion of opinions and predictions from correction rights, suggest a pragmatic approach to avoid overly burdensome regulations that could stifle innovation. However, it also raises questions about the extent of consumer protection. This regulatory initiative highlights the ongoing challenge of balancing technological advancement with the need to safeguard individual rights and ensure equitable outcomes in an increasingly automated world. It sets a precedent for how states might navigate the ethical landscape of AI, potentially shaping future national and international regulatory discussions.
