The Colorado Attorney General will circulate a redlined draft of the ADMT Act (SB 26-189) and Chatbot Safety Act (HB 26-1263) rules on Sept. 23, giving deployers roughly 72 hours to read the revision before the rulemaking calendar advances toward the Oct. 26 hearing. Both statutes take effect Jan. 1, 2027.

The unresolved question is definitional. The Aug. 11 draft offered two competing standards for what it means to “materially influence” a consequential decision, and asked commenters which to adopt. Seyfarth Shaw notes that the definition of “materially influence” remains the most important unresolved issue, with the Department proposing two possible standards and asking the public which to adopt. For a 5-to-30-person operation running a chatbot for sales qualification or a model that scores inbound leads, that’s the difference between a marketing stack and a regulated one.

If a workflow is covered, the operational lift is real. Ballard Spahr reads the draft to require adverse-outcome notices within 30 days, delivered through at least two communication methods, disclosing principal reasons and risk-score inputs. Independent human review with override authority, and no AI assist, must follow within 45 days. Federal ECOA or FCRA adverse-action notices can satisfy Colorado if they carry the added ADMT content, but the deployer, not the vendor, owns the disclosure.

The Chatbot Safety Act layers on age-estimation, AI disclosure, minor protections, and an annual AG report due July 1, 2027 covering calendar year 2027.

Colorado isn’t moving alone. Parallel pressure is building elsewhere, including California’s independent-auditor requirement and Texas TRAIGA’s complaint portal. The state-by-state AI compliance map is arriving before most small deployers have finished naming their tools.

Sources

Sources