Colorado moved first. Again.
The state that passed the nation’s first comprehensive AI Act in 2024 has now repealed it, replacing SB 24-205 with SB 26-189, a different legal instrument built around a different conceptual framework. Per Baker Botts’ analysis of the bill, the Colorado General Assembly reportedly passed SB 26-189 on May 9, 2026. That passage date is attributed to Baker Botts’ legal analysis, not official Colorado legislative records, and should be confirmed against the General Assembly’s official bill tracking before any formal reliance.
The scope shift is the story. SB 24-205 regulated “high-risk AI systems”, a category defined by application domain and potential harm. SB 26-189 replaces that with “Automated Decision-Making Technology”: AI that “materially influences” consequential decisions. The practical difference isn’t just terminology. The ADMT framing focuses on the decision function rather than the system classification. A tool that would have been analyzed under SB 24-205’s risk tiers might now be evaluated under ADMT’s materiality standard, and vice versa. Legal teams with Colorado exposure need to reclassify their systems under the new framework, not just update their deadline calendars.
The effective date shifts to January 1, 2027, according to Baker Botts’ reading, replacing the prior June 2026 implementation date. That’s roughly seven months of additional runway for organizations building toward Colorado compliance. The runway matters most for employers, financial institutions, and housing providers, whose use of AI in employment screening, credit decisions, and rental applications has been the central focus of Colorado’s legislative efforts from the start.
One provision requires independent verification. The new framework reportedly retains consumer opt-out rights for housing, employment, and credit decisions, but that claim comes from a single source and hasn’t been confirmed against the bill text. Don’t cite this provision as confirmed until the official bill language is reviewed.
The real question is what other state legislatures do with Colorado’s pivot. The “high-risk AI system” framing was adopted from the EU AI Act’s conceptual architecture. ADMT is a distinct American formulation, decision-focused rather than system-focused. If Colorado’s ADMT framework survives legal challenge and produces workable enforcement guidance before January 2027, it gives other state legislatures an alternative model to the EU-derived approach that dominated first- generation AI legislation.
One non-obvious implication worth tracking: the ADMT scope explicitly covers employment decisions. That means SB 26-189 intersects directly with AI-driven workforce automation, not as a labor law, but as a decision-accountability framework. Employers using AI for hiring, scheduling, or performance evaluation in Colorado will need to assess whether their tools “materially influence” consequential employment decisions within the ADMT definition. That’s a compliance question that also functions as a disclosure question, and legal teams building ADMT programs will need to coordinate with HR. For broader context on AI’s role in employment decisions, see the Job Displacement Hub’s coverage of ADMT in employment contexts.
Don’t expect the ADMT concept to stay in Colorado. Watch for similar framing in state AI bills introduced in 2026 legislative sessions, the conceptual migration from system classification to decision function is a shift that benefits from a working precedent, and Colorado just provided one.